Authors: Yuexin Liao
Abstract: This paper introduces a differentiable framework that embeds the axiomatic structure of Random Utility Models (RUM) directly into deep neural networks. Although projecting empirical choice data onto the RUM polytope is NP-hard in general, we uncover an isomorphism between RUM consistency and flow conservation on the Boolean lattice. Leveraging this combinatorial structure, we derive a novel Tree-Preconditioned Conjugate Gradient solver. By exploiting the spanning tree of the constraint graph, our preconditioner effectively "whitens" the ill-conditioned Hessian spectrum induced by the Interior Point Method barrier, achieving superlinear convergence and scaling to problem sizes previously deemed unsolvable. We further formulate the projection as a differentiable layer via the Implicit Function Theorem, where the exact Jacobian propagates geometric constraints during backpropagation. Empirical results demonstrate that this "Axioms-as-Layers" paradigm eliminates the structural overfitting inherent in penalty-based methods, enabling models that are jointly trainable, provably rational, and capable of generalizing from sparse data regimes where standard approximations fail.
Authors: Zeming Du, Qitan Shao, Hongfei Liu, Yong Zhang
Abstract: While accurate traffic forecasting is vital for Intelligent Transportation Systems (ITS), effectively communicating predicted conditions via natural language for human-centric decision support remains a challenge and is often handled separately. To address this, we propose CrossTrafficLLM, a novel GenAI-driven framework that simultaneously predicts future spatiotemporal traffic states and generates corresponding natural language descriptions, specifically targeting conditional abnormal event summaries. We tackle the core challenge of aligning quantitative traffic data with qualitative textual semantics by leveraging Large Language Models (LLMs) within a unified architecture. This design allows generative textual context to improve prediction accuracy while ensuring generated reports are directly informed by the forecast. Technically, a text-guided adaptive graph convolutional network is employed to effectively merge high-level semantic information with the traffic network structure. Evaluated on the BJTT dataset, CrossTrafficLLM demonstrably surpasses state-of-the-art methods in both traffic forecasting performance and text generation quality. By unifying prediction and description generation, CrossTrafficLLM delivers a more interpretable, and actionable approach to generative traffic intelligence, offering significant advantages for modern ITS applications.
Authors: Peter Wang, Neelesh Gupta, Viktor Prasanna
Abstract: The need for long-context reasoning has led to alternative neural network architectures besides Transformers and self-attention, a popular model being Hyena, which employs causal 1D-convolutions implemented with FFTs. Long convolutions enable efficient global context mixing, but requirements for intermediate results exceed the 2-3 MB Block RAM capacity of FPGAs. We present a chunked FFT convolution approach enabling 450K length sequence by 450K length filter convolutions on an Alveo U200 FPGA with 2.8 MB BRAM through chunking and overlap-add reconstruction. We find that throughput scales proportionally with chunk size while degrading minimally by 7% for our longest sequences, demonstrating that careful memory management enables deployment of long-context primitives on edge FPGAs without sacrificing performance.
Authors: Ali Rahimi
Abstract: We describe an exact algorithm for solving linear systems $Hx=b$ where $H$ is the Hessian of a deep net. The method computes Hessian-inverse-vector products without storing the Hessian or its inverse in time and storage that scale linearly in the number of layers. Compared to the naive approach of first computing the Hessian, then solving the linear system, which takes storage that's quadratic in the number of parameters and cubically many operations, our Hessian-inverse-vector product method scales roughly like Pearlmutter's algorithm for computing Hessian-vector products.
Authors: Andrew Kiruluta
Abstract: We present a theory-first framework that interprets inference-time adaptation in large language models (LLMs) as online Bayesian state estimation. Rather than modeling rapid adaptation as implicit optimization or meta-learning, we formulate task- and context-specific learning as the sequential inference of a low-dimensional latent adaptation state governed by a linearized state-space model. Under Gaussian assumptions, adaptation follows a Kalman recursion with closed-form updates for both the posterior mean and covariance. This perspective elevates epistemic uncertainty to an explicit dynamical variable. We show that inference-time learning is driven by covariance collapse, i.e., rapid contraction of posterior uncertainty induced by informative tokens, which typically precedes convergence of the posterior mean. Using observability conditions on token-level Jacobians, we establish stability of the Bayesian filter, prove exponential covariance contraction rates, and derive mean-square error bounds. Gradient descent, natural-gradient methods, and meta-learning updates arise as singular, noise-free limits of the filtering dynamics, positioning optimization-based adaptation as a degenerate approximation of Bayesian inference. The resulting theory provides a unified probabilistic account of in-context learning, parameter-efficient adaptation, and test-time learning without parameter updates. It yields explicit guarantees on stability and sample efficiency, offers a principled interpretation of prompt informativeness via information accumulation, and clarifies the role of uncertainty dynamics absent from existing accounts. Minimal illustrative experiments corroborate the qualitative predictions of the theory.
Authors: Muhammed Yusuf Kocyigit, Caglar Yildirim
Abstract: We present a controlled study of how dataset contamination interacts with the post-training stages now standard in large language model training pipelines. Starting from clean checkpoints of Qwen2.5 (0.5B/1.5B) and Gemma3 (1B/4B), we inject five copies of GSM8K and MBPP test items into the first 2B tokens of an otherwise 25B token extended pre-training dataset. We then compare the contaminated and clean models both immediately after pre-training and again after two popular post-training methods: supervised fine-tuning (SFT) and reinforcement learning (RL) with group relative policy optimization (GRPO). The applied post-training steps do not have any contamination. Across math and coding benchmarks, we find three consistent patterns: (i) Contamination causes performance spikes that are gradually diminished with continued pre-training. After even 25B tokens the apparent performance inflation of contamination can become close to zero. (ii) Both SFT and GRPO resurface the leaked information, but with different external validity: SFT inflates scores only on the contaminated tasks, whereas GRPO also inflates performance on uncontaminated counterparts (GSMPlus, HumanEval). (iii) Model scale amplifies these tendencies, larger Supervised Fine Tuned models memorize more, while larger GRPO models translate leakage into more generalizable capabilities. Our results underscore the need for contamination audits \emph{after} post-training and suggest that RL-based post-training, although not immune, can help alleviate contamination-related over-estimation problems.
Authors: Tanvi Jois, Hussain Ahmad, Fatima Noor, Faheem Ullah
Abstract: Bushfires are among the most destructive natural hazards in Australia, causing significant ecological, economic, and social damage. Accurate prediction of bushfire intensity is therefore essential for effective disaster preparedness and response. This study examines the predictive capability of spatio-temporal environmental data for identifying high-risk bushfire zones across Australia. We integrated historical fire events from NASA FIRMS, daily meteorological observations from Meteostat, and vegetation indices such as the Normalized Difference Vegetation Index (NDVI) from Google Earth Engine for the period 2015-2023. After harmonizing the datasets using spatial and temporal joins, we evaluated several machine learning models, including Random Forest, XGBoost, LightGBM, a Multi-Layer Perceptron (MLP), and an ensemble classifier. Under a binary classification framework distinguishing 'low' and 'high' fire risk, the ensemble approach achieved an accuracy of 87%. The results demonstrate that combining multi-source environmental features with advanced machine learning techniques can produce reliable bushfire intensity predictions, supporting more informed and timely disaster management.
Authors: Min-Han Shih, Yu-Hsin Wu, Yu-Wei Chen
Abstract: We propose a dedicated multimodal Judge Model designed to provide reliable, explainable evaluation across a diverse suite of tasks. Our benchmark spans text, audio, image, and video modalities, drawing from carefully sampled public datasets with fixed seeds to ensure reproducibility and minimize train test leakage. Instead of simple scoring, our framework aggregates multimodal judgments, analyzes the quality and reasoning consistency of model outputs, and generates diagnostic feedback. We evaluate several MLLMs, including Gemini 2.5, Phi 4, and Qwen 2.5, across 280 multimodal samples and compare judge model assessments with human annotators. Results show strong alignment between the Judge Model and human scores, demonstrating its potential as a scalable, interpretable evaluation pipeline for future multimodal AI research.
Authors: Jinwoong Kim, Sangjin Park
Abstract: Multivariate time-series models achieve strong predictive performance in healthcare, industry, energy, and finance, but how they combine cross-variable interactions with temporal dynamics remains unclear. SHapley Additive exPlanations (SHAP) are widely used for interpretation. However, existing time-series variants typically treat the feature and time axes independently, fragmenting structural signals formed jointly by multiple variables over specific intervals. We propose GroupSegment SHAP (GS-SHAP), which constructs explanatory units as group-segment players based on cross-variable dependence and distribution shifts over time, and then quantifies each unit's contribution via Shapley attribution. We evaluate GS-SHAP across four real-world domains: human activity recognition, power-system forecasting, medical signal analysis, and financial time series, and compare it with KernelSHAP, TimeSHAP, SequenceSHAP, WindowSHAP, and TSHAP. GS-SHAP improves deletion-based faithfulness (DeltaAUC) by about 1.7x on average over time-series SHAP baselines, while reducing wall-clock runtime by about 40 percent on average under matched perturbation budgets. A financial case study shows that GS-SHAP identifies interpretable multivariate-temporal interactions among key market variables during high-volatility regimes.
Authors: HyunJun Jeon
Abstract: This paper presents a large-scale stress test of machine learning systems using structured mathematical data as a benchmark. We evaluate the robustness of tree-based classifiers at an unprecedented scale, utilizing ten billion deterministic samples and five billion adversarial counterexamples. Our framework introduces three primary contributions: first, a high-throughput pipeline that reformulates Pythagorean triple generation into a single-parameter index stream, significantly improving computational efficiency over classical methods; second, the Hypothesis-driven Negative Dataset (HND), which categorizes nine classes of adversarial attacks designed to exploit both arithmetic precision and structural patterns; and third, a fault-tolerant infrastructure for reliable large-scale training. Experimental results demonstrate that while LightGBM achieves 99.99% accuracy, feature attribution reveals that the model prioritizes underlying quadratic patterns over direct algebraic verification. These findings suggest that learned heuristics can effectively identify structural representations in numerical data, potentially serving as efficient preprocessors for formal verification methods.
Authors: Dileepa Pitawela, Gustavo Carneiro, Hsiang-Ting Chen
Abstract: Recent research highlights the potential of machine learning models to learn to complement (L2C) human strengths; however, generalizing this capability to unseen users remains a significant challenge. Existing L2C methods oversimplify interaction between human and AI by relying on a single, global user model that neglects individual user variability, leading to suboptimal cooperative performance. Addressing this, we introduce L2CU, a novel L2C framework for human-AI cooperative classification with unseen users. Given sparse and noisy user annotations, L2CU identifies representative annotator profiles capturing distinct labeling patterns. By matching unseen users to these profiles, L2CU leverages profile-specific models to complement the user and achieve superior joint accuracy. We evaluate L2CU on datasets (CIFAR-10N, CIFAR-10H, Fashion-MNIST-H, Chaoyang and AgNews), demonstrating its effectiveness as a model-agnostic solution for improving human-AI cooperative classification.
Authors: Lucio M. Dery, Zohar Yahav, Henry Prior, Qixuan Feng, Jiajun Shen, Arthur Szlam
Abstract: Solving increasingly complex problems with large language models (LLMs) necessitates a move beyond individual models and towards multi-model systems that can effectively collaborate. While text has traditionally served as the medium for inter-model communication, a richer and more efficient exchange is possible if models can access each other's internal states directly. In this paper, we propose learning a shared representation space that aligns the k-v caches of multiple models, creating a high-bandwidth channel for collaboration without altering the underlying pre-trained parameters. We do so by augmenting each model with adapters to translate its state into and out of this shared space. Via a suite of experiments with Gemma-2 models, we demonstrate that this approach not only enables seamless inter-model communication but also improves individual model performance. We also show that the shared space allows for the direct transfer of learned skills, such as soft prompts, between different models. Our work represents a significant step towards a future where models can fluidly share knowledge and capabilities.
Authors: Adewumi Augustine Adepitan, Christopher J. Haruna, Morayo Ogunsina, Damilola Olawoyin Yussuf, Ayooluwatomiwa Ajiboye
Abstract: Accurate roadway travel-time prediction is foundational to transportation systems analysis, yet widespread reliance on either data-intensive congestion models or overly na\"ive heuristics limits scalability and practical adoption in engineering workflows. This paper develops a lightweight estimator for minimally-congested car travel times that integrates open road-network data, speed constraints, and sparse control/turn features within a random forest framework to correct bias from shortest-path traversal-time baselines. Using an urban testbed, the pipeline: (i) constructs drivable networks from volunteered geographic data; (ii) solves Dijkstra routes minimizing edge traversal time; (iii) derives sparse operational features (signals, stops, crossings, yield, roundabouts; left/right/slight/U-turn counts); and (iv) trains a regression ensemble on limited high-quality reference times to generalize predictions beyond the training set. Out-of-sample evaluation demonstrates marked improvements over traversal-time baselines across mean absolute error, mean absolute percentage error, mean squared error, relative bias, and explained variance, with no significant mean bias under minimally congested conditions and consistent k-fold stability indicating negligible overfitting. The resulting approach offers a practical middle ground for transportation engineering: it preserves point-to-point fidelity at metropolitan scale, reduces resource requirements, and supplies defensible performance estimates where congestion feeds are inaccessible or cost-prohibitive, supporting planning, accessibility, and network performance applications under low-traffic operating regimes.
Authors: SM Ashfaq uz Zaman, Faizan Qamar, Masnizah Mohd, Nur Hanis Sabrina Suhaimi, Amith Khandakar
Abstract: Automatic Identification System (AIS) data are vital for maritime domain awareness, yet they often suffer from domain shifts, data sparsity, and class imbalance, which hinder the performance of predictive models. In this paper, we propose a robust data augmentation method, AISCycleGen, based on Cycle-Consistent Generative Adversarial Networks (CycleGAN), which is tailored for AIS datasets. Unlike traditional methods, AISCycleGen leverages unpaired domain translation to generate high-fidelity synthetic AIS data sequences without requiring paired source-target data. The framework employs a 1D convolutional generator with adaptive noise injection to preserve the spatiotemporal structure of AIS trajectories, enhancing the diversity and realism of the generated data. To demonstrate its efficacy, we apply AISCycleGen to several baseline regression models, showing improvements in performance across various maritime domains. The results indicate that AISCycleGen outperforms contemporary GAN-based augmentation techniques, achieving a PSNR value of 30.5 and an FID score of 38.9. These findings underscore AISCycleGen's potential as an effective and generalizable solution for augmenting AIS datasets, improving downstream model performance in real-world maritime intelligence applications.
Authors: Wonhyeok Choi, Minwoo Choi, Jungwan Woo, Kyumin Hwang, Jaeyeul Kim, Sunghoon Im
Abstract: Diffusion policies have emerged as a powerful approach for robotic control, demonstrating superior expressiveness in modeling multimodal action distributions compared to conventional policy networks. However, their integration with online reinforcement learning remains challenging due to fundamental incompatibilities between diffusion model training objectives and standard RL policy improvement mechanisms. This paper presents the first comprehensive review and empirical analysis of current Online Diffusion Policy Reinforcement Learning (Online DPRL) algorithms for scalable robotic control systems. We propose a novel taxonomy that categorizes existing approaches into four distinct families -- Action-Gradient, Q-Weighting, Proximity-Based, and Backpropagation Through Time (BPTT) methods -- based on their policy improvement mechanisms. Through extensive experiments on a unified NVIDIA Isaac Lab benchmark encompassing 12 diverse robotic tasks, we systematically evaluate representative algorithms across five critical dimensions: task diversity, parallelization capability, diffusion step scalability, cross-embodiment generalization, and environmental robustness. Our analysis identifies key findings regarding the fundamental trade-offs inherent in each algorithmic family, particularly concerning sample efficiency and scalability. Furthermore, we reveal critical computational and algorithmic bottlenecks that currently limit the practical deployment of online DPRL. Based on these findings, we provide concrete guidelines for algorithm selection tailored to specific operational constraints and outline promising future research directions to advance the field toward more general and scalable robotic learning systems.
Authors: Jiquan Wang, Sha Zhao, Yangxuan Zhou, Yiming Kang, Shijian Li, Gang Pan
Abstract: Electroencephalography (EEG) foundation models hold significant promise for universal Brain-Computer Interfaces (BCIs). However, existing approaches often rely on end-to-end fine-tuning and exhibit limited efficacy under frozen-probing protocols, lacking the intrinsic universality required for broad generalization. This limitation stems from adapting general-purpose sequence architectures that overlook the biophysical and dynamical principles of neural activity. To bridge this gap, we propose DeeperBrain, a neuro-grounded foundation model integrating domain-specific inductive biases into its model design and learning objectives. Architecturally, DeeperBrain incorporates a volume conduction-aware channel encoding to model spatial mixing via 3D geometry, and a neurodynamics-aware temporal encoding capturing slow adaptations using oscillatory and exponential bases. For pretraining, we introduce a dual-objective strategy combining Masked EEG Reconstruction (MER) for local fidelity and Neurodynamics Statistics Prediction (NSP). NSP enforces alignment with macroscopic brain states by predicting interpretable order parameters, including spectral power, functional connectivity, cross-frequency coupling, and dynamic complexity. Extensive experiments demonstrate that DeeperBrain achieves state-of-the-art or highly competitive performance under end-to-end fine-tuning. Crucially, it maintains superior efficacy under a rigorous frozen-probing protocol, verifying that embedding neuroscientific first principles endows learned representations with the intrinsic universality essential for universal BCI. The code will be publicly available.
Authors: Zhaowen Fan
Abstract: This work introduces Adaptive Density Fields (ADF), a geometric attention framework that formulates spatial aggregation as a query-conditioned, metric-induced attention operator in continuous space. By reinterpreting spatial influence as geometry-preserving attention grounded in physical distance, ADF bridges concepts from adaptive kernel methods and attention mechanisms. Scalability is achieved via FAISS-accelerated inverted file indices, treating approximate nearest-neighbor search as an intrinsic component of the attention mechanism. We demonstrate the framework through a case study on aircraft trajectory analysis in the Chengdu region, extracting trajectory-conditioned Zones of Influence (ZOI) to reveal recurrent airspace structures and localized deviations.
Authors: Yifang Zhang, Shengwu Xiong, Henan Wang, Wenjie Yin, Jiawang Peng, Duan Zhou, Yuqiang Zhang, Chen Zhou, Hua Chen, Qile Zhao, Pengfei Duan
Abstract: Global navigation satellite systems (GNSS) station-based Precipitation Nowcasting aims to predict rainfall within the next 0-6 hours by leveraging a GNSS station's historical observations of precipitation, GNSS-PWV, and related meteorological variables, which is crucial for disaster mitigation and real-time decision-making. In recent years, time-series forecasting approaches have been extensively applied to GNSS station-based precipitation nowcasting. However, the highly imbalanced temporal distribution of precipitation, marked not only by the dominance of non-rainfall events but also by the scarcity of extreme precipitation samples, significantly limits model performance in practical applications. To address the dual imbalance problem in precipitation nowcasting, we propose a continuous probability modeling-based framework, RainBalance. This plug-and-play module performs clustering for each input sample to obtain its cluster probability distribution, which is further mapped into a continuous latent space via a variational autoencoder (VAE). By learning in this continuous probabilistic space, the task is reformulated from fitting single and imbalance-prone precipitation labels to modeling continuous probabilistic label distributions, thereby alleviating the imbalance issue. We integrate this module into multiple state-of-the-art models and observe consistent performance gains. Comprehensive statistical analysis and ablation studies further validate the effectiveness of our approach.
Authors: Rohit Kaushik, Eva Kaushik
Abstract: Cardiovascular disease (CVD) continues to be the major cause of death globally, calling for predictive models that not only handle diverse and high-dimensional biomedical signals but also maintain interpretability and privacy. We create a single multimodal learning framework that integrates cross modal transformers with graph neural networks and causal representation learning to measure personalized CVD risk. The model combines genomic variation, cardiac MRI, ECG waveforms, wearable streams, and structured EHR data to predict risk while also implementing causal invariance constraints across different clinical subpopulations. To maintain transparency, we employ SHAP based feature attribution, counterfactual explanations and causal latent alignment for understandable risk factors. Besides, we position the design in a federated, privacy, preserving optimization protocol and establish rules for convergence, calibration and uncertainty quantification under distributional shift. Experimental studies based on large-scale biobank and multi institutional datasets reveal state discrimination and robustness, exhibiting fair performance across demographic strata and clinically distinct cohorts. This study paves the way for a principled approach to clinically trustworthy, interpretable and privacy respecting CVD prediction at the population level.
Authors: Felix Biggs, Samuel Willis
Abstract: Meta-learning methods for regression like Neural (Diffusion) Processes achieve impressive results, but with these models it can be difficult to incorporate expert prior knowledge and information contained in metadata. Large Language Models (LLMs) are trained on giant corpora including varied real-world regression datasets alongside their descriptions and metadata, leading to impressive performance on a range of downstream tasks. Recent work has extended this to regression tasks and is able to leverage such prior knowledge and metadata, achieving surprisingly good performance, but this still rarely matches dedicated meta-learning methods. Here we introduce a general method for sampling from a product-of-experts of a diffusion or flow matching model and an `expert' with binned probability density; we apply this to combine neural diffusion processes with LLM token probabilities for regression (which may incorporate textual knowledge), exceeding the empirical performance of either alone.
Authors: Naomi Fridman, Berta Ben Shachar
Abstract: Intrapartum cardiotocography (CTG) is widely used for fetal monitoring during labor, yet its interpretation suffers from high inter-observer variability and limited predictive accuracy. Deep learning approaches have been constrained by the scarcity of CTG recordings with clinical outcome labels. We present the first application of self-supervised pre-training to intrapartum CTG analysis, leveraging 2,444 hours of unlabeled recordings for masked pre-training followed by fine-tuning on the 552-recording CTU-UHB benchmark. Using a PatchTST transformer architecture with a channel-asymmetric masking scheme designed for fetal heart rate reconstruction, we achieve an area under the receiver operating characteristic curve of 0.83 on the full test set and 0.853 on uncomplicated vaginal deliveries, exceeding previously reported results on this benchmark (0.68-0.75). Error analysis reveals that false-positive alerts typically correspond to CTG patterns judged concerning on retrospective clinical review, suggesting clinically meaningful predictions even when umbilical pH is normal. We release standardized dataset splits and model weights to enable reproducible benchmarking. Our results demonstrate that self-supervised pre-training can address data scarcity in fetal monitoring, offering a path toward reliable decision support in the labor room.
Authors: Varun Kotte
Abstract: Structured extraction with LLMs fails in production not because models lack understanding, but because output formatting is unreliable across models and prompts. A prompt that returns clean JSON on GPT-4 may produce fenced, prose-wrapped, or malformed output on Llama, causing strict parsers to reject otherwise correct extractions. We formalize this as format collapse and introduce a dual-metric evaluation framework: ROS (strict parsing, measuring operational reliability) and CSS (post-canonicalization, measuring semantic capability). On a 37,346-example camera metadata benchmark across six model families, we find severe format collapse (for example, Gemma-2B: ROS 0.116 versus CSS 0.246) and large cross-model portability gaps (0.4 to 0.6 F1). We then present PromptPort, a reliability layer combining deterministic canonicalization with a lightweight verifier (DistilBERT) and a safe-override policy. PromptPort recovers format failures (plus 6 to 8 F1), adds verifier-driven semantic selection (plus 14 to 16 F1 beyond canonicalization), and approaches per-field oracle performance (0.890 versus 0.896 in zero-shot) without modifying base models. The method generalizes to held-out model families and provides explicit abstention when uncertain, enabling reliable structured extraction in production deployments.
Authors: Kapil Wanaskar, Gaytri Jena, Vinija Jain, Aman Chadha, Amitava Das
Abstract: Alignment in large language models (LLMs) is still largely static: after training, the policy is frozen. DPO, GRPO methods typically imprint one behavior into the weights, leaving little runtime control beyond prompt hacks or expensive re-alignment. We introduce ECLIPTICA, which treats alignment as instruction-driven and runtime-controllable: natural-language alignment instructions act as an explicit behavioral contract (stance, refusal boundary, verbosity) that modulates behavior on the fly under evolving safety requirements, user roles, and governance constraints. We introduce CITA (Contrastive Instruction-Tuned Alignment), combining SFT with contrastive preference optimization under an explicit geometric anchor to a reference model. This yields a stable Riemannian chart and keeps instruction updates within a shared neighborhood, so regimes stay nearby and traversable for reliable switching. To isolate policy switching from ordinary instruction following, we release the ECLIPTICA benchmark: 3000 controlled cases (300 prompts x 10 instruction types) where the user request is fixed and only the alignment instruction changes. On Llama-3.1-8B across five suites (ECLIPTICA, TruthfulQA, Conditional Safety, Length Control, LITMUS), CITA reaches 86.7% instruction-alignment efficiency, beating DPO (56.1%), GRPO (36.1%), and PPO (20.4%).
Authors: Niklas Jacobs, Manuel C. Voelkle, Norbert Kathmann, Kevin Hilbert
Abstract: In the context of personalized medicine, machine learning algorithms are growing in popularity. These algorithms require substantial information, which can be acquired effectively through the usage of previously gathered data. Open data and the utilization of synthetization techniques have been proposed to address this. In this paper, we propose and evaluate alternative approach that uses additional simulated data based on summary statistics published in the literature. The simulated data are used to pretrain random forests, which are afterwards fine-tuned on a real dataset. We compare the predictive performance of the new approach to random forests trained only on the real data. A Monte Carlo Cross Validation (MCCV) framework with 100 iterations was employed to investigate significance and stability of the results. Since a first study yielded inconclusive results, a second study with improved methodology (i.e., systematic information extraction and different prediction outcome) was conducted. In Study 1, some pretrained random forests descriptively outperformed the standard random forest. However, this improvement was not significant (t(99) = 0.89, p = 0.19). Contrary to expectations, in Study 2 the random forest trained only with the real data outperformed the pretrained random forests. We conclude with a discussion of challenges, such as the scarcity of informative publications, and recommendations for future research.
Authors: Kaiyuan Deng, Gen Li, Yang Xiao, Bo Hui, Xiaolong Ma
Abstract: Text-to-image diffusion models have achieved remarkable progress, yet their use raises copyright and misuse concerns, prompting research into machine unlearning. However, extending multi-concept unlearning to large-scale scenarios remains difficult due to three challenges: (i) conflicting weight updates that hinder unlearning or degrade generation; (ii) imprecise mechanisms that cause collateral damage to similar content; and (iii) reliance on additional data or modules, creating scalability bottlenecks. To address these, we propose Scalable-Precise Concept Unlearning (ScaPre), a unified framework tailored for large-scale unlearning. ScaPre introduces a conflict-aware stable design, integrating spectral trace regularization and geometry alignment to stabilize optimization, suppress conflicts, and preserve global structure. Furthermore, an Informax Decoupler identifies concept-relevant parameters and adaptively reweights updates, strictly confining unlearning to the target subspace. ScaPre yields an efficient closed-form solution without requiring auxiliary data or sub-models. Comprehensive experiments on objects, styles, and explicit content demonstrate that ScaPre effectively removes target concepts while maintaining generation quality. It forgets up to $\times \mathbf{5}$ more concepts than the best baseline within acceptable quality limits, achieving state-of-the-art precision and efficiency for large-scale unlearning.
Authors: Anshum Rankawat
Abstract: Parent-Guided Adaptive Reliability (PGAR) is a lightweight behavioural meta-learning framework that adds a supervisory "parent" layer on top of a standard learner to improve stability, calibration, and recovery under disturbances. PGAR computes three reflex-level signals (incident detection, overconfidence correction, and recovery memory) and fuses them into a bounded reliability index in [0,1]. This index continuously modulates the learner's effective learning rate, reducing update magnitude during instability and restoring it as reliability improves. We provide a Lyapunov-based proof sketch establishing bounded adaptation of the reliability dynamics under mild assumptions (smooth loss, descent direction, and bounded reflex outputs). Empirical evaluations on representative learning tasks show improved calibration, reduced loss variance, and faster recovery compared to standard optimizers, while retaining computational simplicity. PGAR functions as a plug-in reliability layer for existing optimization and learning pipelines, supporting interpretable reliability traces in safety-relevant settings.
Authors: Saki Imai, Pedram Heydari, Anthony Sicilia, Asteria Kaeberlein, Katherine Atwell, Malihe Alikhani
Abstract: Preference based alignment objectives implicitly assume that all human preferences are expressed with equal strength. In practice, however, preference strength varies across individuals and contexts -- a phenomenon established in behavioral economics and discrete choice theory. This mismatch limits the ability of existing objectives to faithfully capture heterogeneous human judgments. Inspired by this literature, we introduce Mixed Logit Direct Preference Optimization (MixDPO), a generalization of Direct Preference Optimization that models variation in preference strength. MixDPO enables alignment objectives to capture heterogeneity in how strongly preferences are expressed across training examples. We evaluate MixDPO on three preference datasets using two open-weight language models. Across datasets, MixDPO improves aggregate alignment performance (+11.2 points on Pythia-2.8B) while preserving subgroup level preferences, with the largest gains appearing in settings with higher inferred preference heterogeneity. MixDPO makes preference heterogeneity explicit through learned strength distributions. We release our code for reproducibility.
Authors: Oliver T. Schmidt, Aaron Towne, Adrian Lozano-Duran, Scott T. M. Dawson, Ricardo Vinuesa
Abstract: We introduce a community challenge designed to facilitate direct comparisons between data-driven methods for compression, forecasting, and sensing of complex aerospace flows. The challenge is organized into three tracks that target these complementary capabilities: compression (compact representations for large datasets), forecasting (predicting future flow states from a finite history), and sensing (inferring unmeasured flow states from limited measurements). Across these tracks, multiple challenges span diverse flow datasets and use cases, each emphasizing different model requirements. The challenge is open to anyone, and we invite broad participation to build a comprehensive and balanced picture of what works and where current methods fall short. To support fair comparisons, we provide standardized success metrics, evaluation tools, and baseline implementations, with one classical and one machine-learning baseline per challenge. Final assessments use blind tests on withheld data. We explicitly encourage negative results and careful analyses of limitations. Outcomes will be disseminated through an AIAA Journal Virtual Collection and invited presentations at AIAA conferences.
Authors: Sean P. Engelstad, Sameul R. Darr, Matthew Taliaferro, Vinay K. Goyal
Abstract: Supporting Go/No-Go decisions prior to launch requires assessing real-time telemetry data against redline limits established during the design qualification phase. Family data from ground testing or previous flights is commonly used to detect initiating failure modes and their timing; however, this approach relies heavily on engineering judgment and is more error-prone for new launch vehicles. To address these limitations, we utilize Long-Term Short-Term Memory (LSTM) networks for supervised classification of time-series anomalies. Although, initial training labels derived from simulated anomaly data may be suboptimal due to variations in anomaly strength, anomaly settling times, and other factors. In this work, we propose a novel statistical detector based on the Mahalanobis distance and forward-backward detection fractions to adjust the supervised training labels. We demonstrate our method on digital twin simulations of a ground-stage propulsion system with 20.8 minutes of operation per trial and O(10^8) training timesteps. The statistical data relabeling improved precision and recall of the LSTM classifier by 7% and 22% respectively.
Authors: Wei Ai, Yun Peng, Yuntao Shou, Tao Meng, Keqin Li
Abstract: With the rapid growth of IoT devices and latency-sensitive applications, the demand for both real-time and energy-efficient computing has surged, placing significant pressure on traditional cloud computing architectures. Mobile edge computing (MEC), an emerging paradigm, effectively alleviates the load on cloud centers and improves service quality by offloading computing tasks to edge servers closer to end users. However, the limited computing resources, non-continuous power provisioning (e.g., battery-powered nodes), and highly dynamic systems of edge servers complicate efficient task scheduling and resource allocation. To address these challenges, this paper proposes a multi-agent deep reinforcement learning algorithm, TG-DCMADDPG, and constructs a collaborative computing framework for multiple edge servers, aiming to achieve joint optimization of fine-grained task partitioning and offloading. This approach incorporates a temporal graph neural network (TimeGNN) to model and predict time series of multi-dimensional server state information, thereby reducing the frequency of online interactions and improving policy predictability. Furthermore, a multi-agent deterministic policy gradient algorithm (DC-MADDPG) in a discrete-continuous hybrid action space is introduced to collaboratively optimize task partitioning ratios, transmission power, and priority scheduling strategies. Extensive simulation experiments confirm that TG-DCMADDPG achieves markedly faster policy convergence, superior energy-latency optimization, and higher task completion rates compared with existing state-of-the-art methods, underscoring its robust scalability and practical effectiveness in dynamic and constrained MEC scenarios.
Authors: Qing He (Ant Group, Hangzhou, China), Dongsheng Bi (Ant Group, Hangzhou, China), Jianrong Lu (Ant Group, Hangzhou, China, Zhejiang University, Hangzhou, China), Minghui Yang (Ant Group, Hangzhou, China), Zixiao Chen (Ant Group, Hangzhou, China), Jiacheng Lu (Ant Group, Hangzhou, China), Jing Chen (Ant Group, Hangzhou, China), Nannan Du (Ant Group, Hangzhou, China), Xiao Cu (Ant Group, Hangzhou, China), Sijing Wu (Health Information Center of Zhejiang Province, Hangzhou, China), Peng Xiang (Department of AI and IT, The Second Affiliated Hospital, School of Medicine, Zhejiang University, Hangzhou, China), Yinyin Hu (Health Information Center of Zhejiang Province, Hangzhou, China), Yi Guo (Health Information Center of Zhejiang Province, Hangzhou, China), Chunpu Li (Health Information Center of Zhejiang Province, Hangzhou, China), Shaoyang Li (Ant Group, Hangzhou, China), Zhuo Dong (Ant Group, Hangzhou, China), Ming Jiang (Ant Group, Hangzhou, China), Shuai Guo (Ant Group, Hangzhou, China), Liyun Feng (Ant Group, Hangzhou, China), Jin Peng (Ant Group, Hangzhou, China), Jian Wang (Ant Group, Hangzhou, China), Jinjie Gu (Ant Group, Hangzhou, China), Junwei Liu (Ant Group, Hangzhou, China, School of Software and Microelectronics, Peking University, Beijing, China)
Abstract: The proliferation of Large Language Models (LLMs) presents transformative potential for healthcare, yet practical deployment is hindered by the absence of frameworks that assess real-world clinical utility. Existing benchmarks test static knowledge, failing to capture the dynamic, application-oriented capabilities required in clinical practice. To bridge this gap, we introduce a Medical LLM Benchmark MLB, a comprehensive benchmark evaluating LLMs on both foundational knowledge and scenario-based reasoning. MLB is structured around five core dimensions: Medical Knowledge (MedKQA), Safety and Ethics (MedSE), Medical Record Understanding (MedRU), Smart Services (SmartServ), and Smart Healthcare (SmartCare). The benchmark integrates 22 datasets (17 newly curated) from diverse Chinese clinical sources, covering 64 clinical specialties. Its design features a rigorous curation pipeline involving 300 licensed physicians. Besides, we provide a scalable evaluation methodology, centered on a specialized judge model trained via Supervised Fine-Tuning (SFT) on expert annotations. Our comprehensive evaluation of 10 leading models reveals a critical translational gap: while the top-ranked model, Kimi-K2-Instruct (77.3% accuracy overall), excels in structured tasks like information extraction (87.8% accuracy in MedRU), performance plummets in patient-facing scenarios (61.3% in SmartServ). Moreover, the exceptional safety score (90.6% in MedSE) of the much smaller Baichuan-M2-32B highlights that targeted training is equally critical. Our specialized judge model, trained via SFT on a 19k expert-annotated medical dataset, achieves 92.1% accuracy, an F1-score of 94.37%, and a Cohen's Kappa of 81.3% for human-AI consistency, validating a reproducible and expert-aligned evaluation protocol. MLB thus provides a rigorous framework to guide the development of clinically viable LLMs.
Authors: Wei Li, Wei Zhang, Qingyu Yan
Abstract: Battery capacity degradation prediction has long been a central topic in battery health analytics, and most studies focus on state of health (SoH) estimation and end of life (EoL) prediction. This study extends the scope to online refinement of the entire capacity fade trajectory (CFT) through EntroLnn, a framework based on entropy-guided transformable liquid neural networks (LNNs). EntroLnn treats CFT refinement as an integrated process rather than two independent tasks for pointwise SoH and EoL. We introduce entropy-based features derived from online temperature fields, applied for the first time in battery analytics, and combine them with customized LNNs that model temporal battery dynamics effectively. The framework enhances both static and dynamic adaptability of LNNs and achieves robust and generalizable CFT refinement across different batteries and operating conditions. The approach provides a high fidelity battery health model with lightweight computation, achieving mean absolute errors of only 0.004577 for CFT and 18 cycles for EoL prediction. This work establishes a foundation for entropy-informed learning in battery analytics and enables self-adaptive, lightweight, and interpretable battery health prediction in practical battery management systems.
Authors: Bodla Krishna Vamshi, Rohan Bhatnagar, Haizhao Yang
Abstract: Large language models (LLMs) frequently generate factually incorrect or unsupported content, commonly referred to as hallucinations. Prior work has explored decoding strategies, retrieval augmentation, and supervised fine-tuning for hallucination detection, while recent studies show that in-context learning (ICL) can substantially influence factual reliability. However, existing ICL demonstration selection methods often rely on surface-level similarity heuristics and exhibit limited robustness across tasks and models. We propose MB-ICL, a manifold-based demonstration sampling framework for selecting in-context demonstrations that leverages latent representations extracted from frozen LLMs. By jointly modeling local manifold structure and class-aware prototype geometry, MB-ICL selects demonstrations based on their proximity to learned prototypes rather than lexical or embedding similarity alone. Across factual verification (FEVER) and hallucination detection (HaluEval) benchmarks, MB-ICL outperforms standard ICL selection baselines in the majority of evaluated settings, with particularly strong gains on dialogue and summarization tasks. The method remains robust under temperature perturbations and model variation, indicating improved stability compared to heuristic retrieval strategies. While lexical retrieval can remain competitive in certain question-answering regimes, our results demonstrate that manifold-based prototype selection provides a reliable and training light approach for hallucination detection without modifying LLM parameters, offering a principled direction for improved ICL demonstration selection.
Authors: Fang Wu, Stan Z. Li
Abstract: Protein-protein interaction (PPI) represents a central challenge within the biology field, and accurately predicting the consequences of mutations in this context is crucial for drug design and protein engineering. Deep learning (DL) has shown promise in forecasting the effects of such mutations, but is hindered by two primary constraints. First, the structures of mutant proteins are often elusive to acquire. Secondly, PPI takes place dynamically, which is rarely integrated into the DL architecture design. To address these obstacles, we present a novel framework named Refine-PPI with two key enhancements. First, we introduce a structure refinement module trained by a mask mutation modeling (MMM) task on available wild-type structures, which is then transferred to produce the inaccessible mutant structures. Second, we employ a new kind of geometric network, called the probability density cloud network (PDC-Net), to capture 3D dynamic variations and encode the atomic uncertainty associated with PPI. Comprehensive experiments on SKEMPI.v2 substantiate the superiority of Refine-PPI over all existing tools for predicting free energy change. These findings underscore the effectiveness of our hallucination strategy and the PDC module in addressing the absence of mutant protein structure and modeling geometric uncertainty.
Authors: Nejad Alagha, Anis Salwa Mohd Khairuddin, Obada Al-Khatib, Abigail Copiaco
Abstract: Wind turbines play a critical role in the shift toward sustainable energy generation. Their operation relies on multiple interconnected components, and a failure in any of these can compromise the entire system's functionality. Detecting faults accurately is challenging due to the intricate, non-linear, and non-stationary nature of vibration signals, influenced by dynamic loading, environmental variations, and mechanical interactions. As such, effective signal processing techniques are essential for extracting meaningful features to enhance diagnostic accuracy. This study presents a hybrid approach for fault detection in wind turbine gearboxes, combining Complete Ensemble Empirical Mode Decomposition with Adaptive Noise (CEEMDAN) and a Multiscale Convolutional Neural Network (MSCNN). CEEMDAN is employed to decompose vibration signals into intrinsic mode functions, isolating critical features at different time-frequency scales. These are then input into the MSCNN, which performs deep hierarchical feature extraction and classification. The proposed method achieves an F1 Score of 98.95\%, evaluated on real-world datasets, and demonstrates superior performance in both detection accuracy and computational speed compared to existing approaches. This framework offers a balanced solution for reliable and efficient fault diagnosis in wind turbine systems.
Authors: Cheng Yan, Wuyang Zhang, Zhiyuan Ning, Fan Xu, Ziyang Tao, Lu Zhang, Bing Yin, Yanyong Zhang
Abstract: The rapid proliferation of Large Language Models (LLMs) has led to a fragmented and inefficient ecosystem, a state of ``model lock-in'' where seamlessly integrating novel models remains a significant bottleneck. Current routing frameworks require exhaustive, costly retraining, hindering scalability and adaptability. We introduce ZeroRouter, a new paradigm for LLM routing that breaks this lock-in. Our approach is founded on a universal latent space, a model-agnostic representation of query difficulty that fundamentally decouples the characterization of a query from the profiling of a model. This allows for zero-shot onboarding of new models without full-scale retraining. ZeroRouter features a context-aware predictor that maps queries to this universal space and a dual-mode optimizer that balances accuracy, cost, and latency. Our framework consistently outperforms all baselines, delivering higher accuracy at lower cost and latency.
Authors: Zhi Wang, Yanni Li, Pingping Zheng, Yiyuan Jiao
Abstract: Clustering temporal and dynamically changing multivariate time series from real-world fields, called temporal clustering for short, has been a major challenge due to inherent complexities. Although several deep temporal clustering algorithms have demonstrated a strong advantage over traditional methods in terms of model learning and clustering results, the accuracy of the few algorithms are not satisfactory. None of the existing algorithms can continuously learn new tasks and deal with the dynamic data effectively and efficiently in the sequential tasks learning. To bridge the gap and tackle these issues, this paper proposes a novel algorithm \textbf{L}ifelong \textbf{D}eep \textbf{T}emporal \textbf{C}lustering (\textbf{LDTC}), which effectively integrates dimensionality reduction and temporal clustering into an end-to-end deep unsupervised learning framework. Using a specifically designed autoencoder and jointly optimizing for both the latent representation and clustering objective, the LDTC can achieve high-quality clustering results. Moreover, unlike any previous work, the LDTC is uniquely equipped with the fully dynamic model expansion and rehearsal-based techniques to effectively learn new tasks and to tackle the dynamic data in the sequential tasks learning without the catastrophic forgetting or degradation of the model accuracy. Experiments on seven real-world multivariate time series datasets show that the LDTC is a promising method for dealing with temporal clustering issues effectively and efficiently.
Authors: Zenghao Duan, Zhiyi Yin, Zhichao Shi, Liang Pang, Shaoling Jing, Zihe Huang, Jiayi Wu, Yu Yan, Jingcheng Deng, Huawei Shen, Xueqi Cheng
Abstract: Large language models (LLMs) exhibit exceptional performance but pose inherent risks of generating toxic content, restricting their safe deployment. While traditional methods (e.g., alignment) adjust output preferences, they fail to eliminate underlying toxic regions in parameters, leaving models vulnerable to adversarial attacks. Prior mechanistic studies characterize toxic regions as "toxic vectors" or "layer-wise subspaces", yet our analysis identifies critical limitations: i) Removed toxic vectors can be reconstructed via linear combinations of non-toxic vectors, demanding targeting of entire toxic subspace; ii) Contrastive objective over limited samples inject noise into layer-wise subspaces, hindering stable extraction. These highlight the challenge of identifying robust toxic subspace and removing them. Therefore, we propose GLOSS (GLobal tOxic Subspace Suppression), a lightweight method that mitigates toxicity by identifying and eliminating this global subspace from FFN parameters. Experiments on LLMs (e.g., Qwen3) show GLOSS achieves SOTA detoxification while preserving general capabilities without requiring large-scale retraining. WARNING: This paper contains context which is toxic in nature.
Authors: Dhivya Dharshini Kannan, Wei Li, Zhang Wei, Jianbiao Wang, Zhi Wei Seh, Man-Fai Ng
Abstract: Battery management systems increasingly require accurate battery health prognostics under strict on-device constraints. This paper presents DLNet, a practical framework with dual-stage distillation of liquid neural networks that turns a high-capacity model into compact and edge-deployable models for battery health prediction. DLNet first applies Euler discretization to reformulate liquid dynamics for embedded compatibility. It then performs dual-stage knowledge distillation to transfer the teacher model's temporal behavior and recover it after further compression. Pareto-guided selection under joint error-cost objectives retains student models that balance accuracy and efficiency. We evaluate DLNet on a widely used dataset and validate real-device feasibility on an Arduino Nano 33 BLE Sense using int8 deployment. The final deployed student achieves a low error of 0.0066 when predicting battery health over the next 100 cycles, which is 15.4% lower than the teacher model. It reduces the model size from 616 kB to 94 kB with 84.7% reduction and takes 21 ms per inference on the device. These results support a practical smaller wins observation that a small model can match or exceed a large teacher for edge-based prognostics with proper supervision and selection. Beyond batteries, the DLNet framework can extend to other industrial analytics tasks with strict hardware constraints.
Authors: Ingo Schmitt
Abstract: An artificial neural network (ANN) is a numerical method used to solve complex classification problems. Due to its high classification power, the ANN method often outperforms other classification methods in terms of accuracy. However, an ANN model lacks interpretability compared to methods that use the symbolic paradigm. Our idea is to derive a symbolic representation from a simple ANN model trained on minterm values of input objects. Based on ReLU nodes, the ANN model is partitioned into cells. We convert the ANN model into a cell-based, three-dimensional bit tensor. The theory of Formal Concept Analysis applied to the tensor yields concepts that are represented as logic trees, expressing interpretable attribute interactions. Their evaluations preserve the classification power of the initial ANN model.
Authors: Arion Das, Partha Pratim Saha, Amit Dhanda, Vinija Jain, Aman Chadha, Amitava Das
Abstract: Direct Preference Optimization (DPO) is a principled, scalable alternative to RLHF for aligning large language models from pairwise preferences, but its internal geometric footprint remains undercharacterized, limiting audits, checkpoint comparisons, and failure prediction. We introduce SPINAL (Scaling-law and Preference Integration in Neural Alignment Layers), a diagnostic that measures how alignment reshapes representations across depth by tracing localized structural change layer by layer. Across model families, DPO produces a layerwise calibration effect concentrated in the final decoder blocks (often layers 21-30), where preference gradients most directly affect the next-token distribution. SPINAL encodes each checkpoint as a depth trace over (layer index, contraction score, transport score). The contraction score summarizes how quickly the tail of a layer's spectrum decays (how fast small modes vanish); higher values indicate stronger contraction into fewer effective directions. The transport score summarizes how much the token distribution shifts between adjacent layers using a bounded overlap measure; lower values indicate shorter, smoother steps through representation space. Aligned checkpoints show a late-layer ramp-up in contraction and a smooth reduction in transport, consistent with tightened and stabilized policy mass, while unaligned models trace higher-curvature, more entropic, and geometrically incoherent depth paths. Overall, alignment is geometrically localized: the final layers encode the dominant preference-induced corrections. SPINAL turns this localization into a practical audit signal, quantifying where alignment concentrates, how strongly it manifests, and when it begins to destabilize during training.
Authors: Tianhao Xu, Yiming Liu, Xianglong Lu, Yijia Zhao, Xuting Zhou, Aichen Feng, Yiyi Chen, Yi Shen, Qin Zhou, Xumeng Chen, Ilya Sherstyuk, Haorui Li, Rishi Thakkar, Ben Hamm, Yuanzhe Li, Xue Huang, Wenpeng Wu, Anish Shanbhag, Harry Kim, Chuan Chen, Junjie Lai
Abstract: Optimizing Large Language Model (LLM) inference in production systems is increasingly difficult due to dynamic workloads, stringent latency/throughput targets, and a rapidly expanding configuration space. This complexity spans not only distributed parallelism strategies (tensor/pipeline/expert) but also intricate framework-specific runtime parameters such as those concerning the enablement of CUDA graphs, available KV-cache memory fractions, and maximum token capacity, which drastically impact performance. The diversity of modern inference frameworks (e.g., TRT-LLM, vLLM, SGLang), each employing distinct kernels and execution policies, makes manual tuning both framework-specific and computationally prohibitive. We present AIConfigurator, a unified performance-modeling system that enables rapid, framework-agnostic inference configuration search without requiring GPU-based profiling. AIConfigurator combines (1) a methodology that decomposes inference into analytically modelable primitives - GEMM, attention, communication, and memory operations while capturing framework-specific scheduling dynamics; (2) a calibrated kernel-level performance database for these primitives across a wide range of hardware platforms and popular open-weights models (GPT-OSS, Qwen, DeepSeek, LLama, Mistral); and (3) an abstraction layer that automatically resolves optimal launch parameters for the target backend, seamlessly integrating into production-grade orchestration systems. Evaluation on production LLM serving workloads demonstrates that AIConfigurator identifies superior serving configurations that improve performance by up to 40% for dense models (e.g., Qwen3-32B) and 50% for MoE architectures (e.g., DeepSeek-V3), while completing searches within 30 seconds on average. Enabling the rapid exploration of vast design spaces - from cluster topology down to engine specific flags.
Authors: Zhe Jia, Xiaotian Zhang, Junpeng Li
Abstract: Inferring high-dimensional physical states from sparse, ad-hoc sensor arrays is a fundamental challenge across AI for Science, as they are complicated by irregular geometries and the profound Sim-to-Real gap in physical modeling. Taking earthquake source characterization as a representative challenge, we address limitations in conventional deep learning: CNNs demand fixed grids, while pooling-based architectures (e.g., DeepSets) struggle to capture the relational wave physics. Here, we propose SourceNet, a Transformer-based framework that treats the sensor array as a flexible set to model arbitrary geometries. To bridge the reality gap, we introduce Physics-Structured Domain Randomization (PSDR). Instead of forcing feature alignment, PSDR randomizes the governing physical dynamics by varying velocity structures, propagation effects, and sensor availability, to force the model to learn robust representations invariant to unmodeled environmental heterogeneity. By pre-training on 100,000 synthetic events and fine-tuning on ~2,000 real world events, SourceNet achieves state-of-the-art precision on held-out real data. This demonstrates exceptional data efficiency, and matches classical solvers while enabling real-time processing. Remarkably, interpretability analysis reveals that the model shows scientific-agent-like features: it autonomously discovers geometric information bottlenecks and learns an attention policy that prioritizes sparse sensor placements, effectively recovering principles of optimal experimental design from data alone.
Authors: Benjamin Turtel, Paul Wilczewski, Danny Franklin, Kris Skothiem
Abstract: Many real-world prediction problems lack labels observable at prediction time, creating a temporal gap between prediction and outcome that yields supervision only after events resolve. To address this setting, we extend reinforcement learning with verifiable rewards to temporally resolved real-world prediction, and use it to train language models to make probabilistic forecasts under causally masked information with retrospective evaluation using proper scoring rules. Supervision is derived solely from post-resolution outcomes, preserving delayed-reward semantics. On real-world forecasting benchmarks, Qwen3-32B trained using Foresight Learning improves Brier score by 27% and halves calibration error relative to its pretrained baseline, and outperforms Qwen3-235B on both constructed future-event prediction tasks and the Metaculus benchmark despite a 7x parameter disadvantage.
Authors: Tara Bogavelli, Oluwanifemi Bamgbose, Gabrielle Gauthier Melan\c{c}on, Fanny Riols, Roshnee Sharma
Abstract: Enterprise LLM applications require consistently high quality and reliable performance across diverse scenarios, demanding robustness to minor variations. Existing research shows that even small prompt changes can lead to substantial differences in output, but has mainly focused on a narrow set of perturbations with small academic datasets, limiting their relevance to real-world applications. To address this, we present a comprehensive benchmark suite that evaluates robustness across multiple perturbation types, including general text edits (e.g., punctuation, whitespace), formatting changes (e.g., JSON, YAML), multilingual and cross-lingual inputs, and positional variations in instructions. Evaluating 11 models ranging from 4B to 120B+ parameters, we find that minor perturbations reduce performance by up to 40 percentage points on key enterprise metrics. Critically, we demonstrate that the relationship between model size and robustness is more nuanced than conventional assumptions suggest: an 8B parameter model (Ministral 3 8B) outperforms most larger models, while another 8B model (Llama 3.1 8B) performs worst overall.
Authors: Siqi Zhu, Joshua D. Kaggie
Abstract: Federated Learning (FL) enables collaborative model training across decentralized devices while preserving data privacy. However, real-world FL deployments face critical challenges such as data imbalances, including label noise and non-IID distributions. RHFL+, a state-of-the-art method, was proposed to address these challenges in settings with heterogeneous client models. This work investigates the robustness of RHFL+ under class imbalances through three key contributions: (1) reproduction of RHFL+ along with all benchmark algorithms under a unified evaluation framework; (2) extension of RHFL+ to real-world medical imaging datasets, including CBIS-DDSM, BreastMNIST and BHI; (3) a novel implementation using NVFlare, NVIDIA's production-level federated learning framework, enabling a modular, scalable and deployment-ready codebase. To validate effectiveness, extensive ablation studies, algorithmic comparisons under various noise conditions and scalability experiments across increasing numbers of clients are conducted.
Authors: Philipp Baumann, Olivier Goldschmidt, Dorit S. Hochbaum, Jason Yang
Abstract: The anticlustering problem is to partition a set of objects into K equal-sized anticlusters such that the sum of distances within anticlusters is maximized. The anticlustering problem is NP-hard. We focus on anticlustering in Euclidean spaces, where the input data is tabular and each object is represented as a D-dimensional feature vector. Distances are measured as squared Euclidean distances between the respective vectors. Applications of Euclidean anticlustering include social studies, particularly in psychology, K-fold cross-validation in which each fold should be a good representative of the entire dataset, the creation of mini-batches for gradient descent in neural network training, and balanced K-cut partitioning. In particular, machine-learning applications involve million-scale datasets and very large values of K, making scalable anticlustering algorithms essential. Existing algorithms are either exact methods that can solve only small instances or heuristic methods, among which the most scalable is the exchange-based heuristic fast_anticlustering. We propose a new algorithm, the Assignment-Based Anticlustering algorithm (ABA), which scales to very large instances. A computational study shows that ABA outperforms fast_anticlustering in both solution quality and running time. Moreover, ABA scales to instances with millions of objects and hundreds of thousands of anticlusters within short running times, beyond what fast_anticlustering can handle. As a balanced K-cut partitioning method for tabular data, ABA is superior to the well-known METIS method in both solution quality and running time. The code of the ABA algorithm is available on GitHub.
Authors: Nusrat Jahan Prottasha, Md Kowsher, Chun-Nam Yu, Chen Chen, Ozlem Garibay
Abstract: Mixture-of-experts variants of parameter-efficient fine-tuning enable per-token specialization, but they introduce additional trainable routers and expert parameters, increasing memory usage and training cost. This undermines the core goal of parameter-efficient fine-tuning. We propose Monkey Jump, a method that brings mixture-of-experts-style specialization to parameter-efficient fine-tuning without introducing extra trainable parameters for experts or routers. Instead of adding new adapters as experts, Monkey Jump treats the adapters already present in each Transformer block (such as query, key, value, up, and down projections) as implicit experts and routes tokens among them. Routing is performed using k-means clustering with exponentially moving averaged cluster centers, requiring no gradients and no learned parameters. We theoretically show that token-wise routing increases expressivity and can outperform shared adapters by avoiding cancellation effects. Across multi-task experiments covering 14 text, 14 image, and 19 video benchmarks, Monkey Jump achieves competitive performance with mixture-of-experts-based parameter-efficient fine-tuning methods while using 7 to 29 times fewer trainable parameters, up to 48 percent lower memory consumption, and 1.5 to 2 times faster training. Monkey Jump is architecture-agnostic and can be applied to any adapter-based parameter-efficient fine-tuning method.
Authors: Thomas Vaitses Fontanari, Mariana Recamonde-Mendoza
Abstract: This study explores the use of graph neural networks (GNNs) with hierarchical pooling and multiple convolution layers for cancer classification based on RNA-seq data. We combine gene expression data from The Cancer Genome Atlas (TCGA) with a precomputed STRING protein-protein interaction network to classify tissue origin and distinguish between normal and tumor samples. The model employs Chebyshev graph convolutions (K=2) and weighted pooling layers, aggregating gene clusters into 'supernodes' across multiple coarsening levels. This approach enables dimensionality reduction while preserving meaningful interactions. Saliency methods were applied to interpret the model by identifying key genes and biological processes relevant to cancer. Our findings reveal that increasing the number of convolution and pooling layers did not enhance classification performance. The highest F1-macro score (0.978) was achieved with a single pooling layer. However, adding more layers resulted in over-smoothing and performance degradation. However, the model proved highly interpretable through gradient methods, identifying known cancer-related genes and highlighting enriched biological processes, and its hierarchical structure can be used to develop new explainable architectures. Overall, while deeper GNN architectures did not improve performance, the hierarchical pooling structure provided valuable insights into tumor biology, making GNNs a promising tool for cancer biomarker discovery and interpretation
Authors: Shenghong Cai, Zihua Yang, Yang Lu, Mengke Li, Yuzhu Ji, Yiqun Zhang, Yiu-Ming Cheung
Abstract: Driven by the growth of Web-scale decentralized services, Federated Clustering (FC) aims to extract knowledge from heterogeneous clients in an unsupervised manner while preserving the clients' privacy, which has emerged as a significant challenge due to the lack of label guidance and the Non-Independent and Identically Distributed (non-IID) nature of clients. In real scenarios such as personalized recommendation and cross-device user profiling, the global cluster may be fragmented and distributed among different clients, and the clusters may exist at different granularities or even nested. Although Hierarchical Clustering (HC) is considered promising for exploring such distributions, the sophisticated recursive clustering process makes it more computationally expensive and vulnerable to privacy exposure, thus relatively unexplored under the federated learning scenario. This paper introduces an efficient one-shot hierarchical FC framework that performs client-end distribution exploration and server-end distribution aggregation through one-way prototype-level communication from clients to the server. A fine partition mechanism is developed to generate successive clusterlets to describe the complex landscape of the clients' clusters. Then, a multi-granular learning mechanism on the server is proposed to fuse the clusterlets, even when they have inconsistent granularities generated from different clients. It turns out that the complex cluster distributions across clients can be efficiently explored, and extensive experiments comparing state-of-the-art methods on ten public datasets demonstrate the superiority of the proposed method.
Authors: Liming Liu, Binxuan Huang, Xin Liu, Bing Yin, Tuo Zhao
Abstract: Masked Diffusion Language Models (DLMs) achieve significant speed by generating multiple tokens in parallel. However, this parallel sampling approach, especially when using fewer inference steps, will introduce strong dependency errors and cause quality to deteriorate rapidly as the generation step size grows. As a result, reliable self-correction becomes essential for maintaining high-quality multi-token generation. To address this, we propose Decoupled Self-Correction (DSC), a novel two-stage methodology. DSC first fully optimizes the DLM's generative ability before freezing the model and training a specialized correction head. This decoupling preserves the model's peak SFT performance and ensures the generated errors used for correction head training are of higher quality. Additionally, we introduce Future-Context Augmentation (FCA) to maximize the correction head's accuracy. FCA generalizes the error training distribution by augmenting samples with ground-truth tokens, effectively training the head to utilize a richer, future-looking context. This mechanism is used for reliably detecting the subtle errors of the high-fidelity base model. Our DSC framework enables the model, at inference time, to jointly generate and revise tokens, thereby correcting errors introduced by multi-token generation and mitigating error accumulation across steps. Experiments on mathematical reasoning and code generation benchmarks demonstrate that our approach substantially reduces the quality degradation associated with larger generation steps, allowing DLMs to achieve both high generation speed and strong output fidelity.
Authors: Zhen Liu, Yucheng Wang, Boyuan Li, Junhao Zheng, Emadeldeen Eldele, Min Wu, Qianli Ma
Abstract: Foundation models pre-trained on large-scale source datasets are reshaping the traditional training paradigm for time series classification. However, existing time series foundation models primarily focus on forecasting tasks and often overlook classification-specific challenges, such as modeling interpretable shapelets that capture class-discriminative temporal features. To bridge this gap, we propose UniShape, a unified shape-aware foundation model designed for time series classification. UniShape incorporates a shape-aware adapter that adaptively aggregates multiscale discriminative subsequences (shapes) into class tokens, effectively selecting the most relevant subsequence scales to enhance model interpretability. Meanwhile, a prototype-based pretraining module is introduced to jointly learn instance- and shape-level representations, enabling the capture of transferable shape patterns. Pre-trained on a large-scale multi-domain time series dataset comprising 1.89 million samples, UniShape exhibits superior generalization across diverse target domains. Experiments on 128 UCR datasets and 30 additional time series datasets demonstrate that UniShape achieves state-of-the-art classification performance, with interpretability and ablation analyses further validating its effectiveness.
Authors: Hengliang Wu, Youming Tao, Anhao Zhou, Shuzhen Chen, Falko Dressler, Dongxiao Yu
Abstract: Driven by the right to be forgotten (RTBF), machine unlearning has become an essential requirement for privacy-preserving machine learning. However, its realization in decentralized federated learning (DFL) remains largely unexplored. In DFL, clients exchange local updates only with neighbors, causing model information to propagate and mix across the network. As a result, when a client requests data deletion, its influence is implicitly embedded throughout the system, making removal difficult without centralized coordination. We propose a novel certified unlearning framework for DFL based on Newton-style updates. Our approach first quantifies how a client's data influence propagates during training. Leveraging curvature information of the loss with respect to the target data, we then construct corrective updates using Newton-style approximations. To ensure scalability, we approximate second-order information via Fisher information matrices. The resulting updates are perturbed with calibrated noise and broadcast through the network to eliminate residual influence across clients. We theoretically prove that our approach satisfies the formal definition of certified unlearning, ensuring that the unlearned model is difficult to distinguish from a retrained model without the deleted data. We also establish utility bounds showing that the unlearned model remains close to retraining from scratch. Extensive experiments across diverse decentralized settings demonstrate the effectiveness and efficiency of our framework.
Authors: Ramnath Kumar, Kyle Ritscher, Junmin Judy, Lawrence Liu, Cho-Jui Hsieh
Abstract: Learning activation functions has emerged as a promising direction in deep learning, allowing networks to adapt activation mechanisms to task-specific demands. In this work, we introduce a novel framework that employs the Gumbel-Softmax trick to enable discrete yet differentiable selection among a predefined set of activation functions during training. Our method dynamically learns the optimal activation function independently of the input, thereby enhancing both predictive accuracy and architectural flexibility. Experiments on synthetic datasets show that our model consistently selects the most suitable activation function, underscoring its effectiveness. These results connect theoretical advances with practical utility, paving the way for more adaptive and modular neural architectures in complex learning scenarios.
Authors: Suvo Banik, Troy D. Loeffler, Henry Chan, Sukriti Manna, Orcun Yildiz, Tom Peterka, Subramanian Sankaranarayanan
Abstract: High-dimensional design spaces underpin a wide range of physics-based modeling and computational design tasks in science and engineering. These problems are commonly formulated as constrained black-box searches over rugged objective landscapes, where function evaluations are expensive, and gradients are unavailable or unreliable. Conventional global search engines and optimizers struggle in such settings due to the exponential scaling of design spaces, the presence of multiple local basins, and the absence of physical guidance in sampling. We present a physics-informed Monte Carlo Tree Search (MCTS) framework that extends policy-driven tree-based reinforcement concepts to continuous, high-dimensional scientific optimization. Our method integrates population-level decision trees with surrogate-guided directional sampling, reward shaping, and hierarchical switching between global exploration and local exploitation. These ingredients allow efficient traversal of non-convex, multimodal landscapes where physically meaningful optima are sparse. We benchmark our approach against standard global optimization baselines on a suite of canonical test functions, demonstrating superior or comparable performance in terms of convergence, robustness, and generalization. Beyond synthetic tests, we demonstrate physics-consistent applicability to (i) crystal structure optimization from clusters to bulk, (ii) fitting of classical interatomic potentials, and (iii) constrained engineering design problems. Across all cases, the method converges with high fidelity and evaluation efficiency while preserving physical constraints. Overall, our work establishes physics-informed tree search as a scalable and interpretable paradigm for computational design and high-dimensional scientific optimization, bridging discrete decision-making frameworks with continuous search in scientific design workflows.
Authors: Xuezhe Ma, Shicheng Wen, Linghao Jin, Bilge Acun, Ruihang Lai, Bohan Hou, Will Lin, Hao Zhang, Songlin Yang, Ryan Lee, Mengxi Wu, Jonathan May, Luke Zettlemoyer, Carole-Jean Wu
Abstract: Designing a unified neural network to efficiently and inherently process sequential data with arbitrary lengths is a central and challenging problem in sequence modeling. The design choices in Transformer, including quadratic complexity and weak length extrapolation, have limited their ability to scale to long sequences. In this work, we propose Gecko, a neural architecture that inherits the design of Mega and Megalodon (exponential moving average with gated attention), and further introduces multiple technical components to improve its capability to capture long range dependencies, including timestep decay normalization, sliding chunk attention mechanism, and adaptive working memory. In a controlled pretraining comparison with Llama2 and Megalodon in the scale of 7 billion parameters and 2 trillion training tokens, Gecko achieves better efficiency and long-context scalability. Gecko reaches a training loss of 1.68, significantly outperforming Llama2-7B (1.75) and Megalodon-7B (1.70), and landing close to Llama2-13B (1.67). Notably, without relying on any context-extension techniques, Gecko exhibits inherent long-context processing and retrieval capabilities, stably handling sequences of up to 4 million tokens and retrieving information from contexts up to $4\times$ longer than its attention window. Code: https://github.com/XuezheMax/gecko-llm
Authors: Chutian Huang, Chang Ma, Kaibo Wang, Yang Xiang
Abstract: Learning solution operators for differential equations with neural networks has shown great potential in scientific computing, but ensuring their stability under input perturbations remains a critical challenge. This paper presents a robust self-supervised neural operator framework that enhances stability through adversarial training while preserving accuracy. We formulate operator learning as a min-max optimization problem, where the model is trained against worst-case input perturbations to achieve consistent performance under both normal and adversarial conditions. We demonstrate that our method not only achieves good performance on standard inputs, but also maintains high fidelity under adversarial perturbed inputs. The results highlight the importance of stability-aware training in operator learning and provide a foundation for developing reliable neural PDE solvers in real-world applications, where input noise and uncertainties are inevitable.
Authors: Alan Oursland
Abstract: Gradient descent on log-sum-exp (LSE) objectives performs implicit expectation--maximization (EM): the gradient with respect to each component output equals its responsibility. The same theory predicts collapse without volume control analogous to the log-determinant in Gaussian mixture models. We instantiate the theory in a single-layer encoder with an LSE objective and InfoMax regularization for volume control. Experiments confirm the theory's predictions. The gradient--responsibility identity holds exactly; LSE alone collapses; variance prevents dead components; decorrelation prevents redundancy. The model exhibits EM-like optimization dynamics in which lower loss does not correspond to better features and adaptive optimizers offer no advantage. The resulting decoder-free model learns interpretable mixture components, confirming that implicit EM theory can prescribe architectures.
Authors: Qiang Zhang, Boli Chen, Fanrui Zhang, Ruixue Ding, Shihang Wang, Qiuchen Wang, Yinfeng Huang, Haonan Zhang, Rongxiang Zhu, Pengyong Wang, Ailin Ren, Xin Li, Pengjun Xie, Jiawei Liu, Ning Guo, Jingren Zhou, Zheng-Jun Zha
Abstract: Reinforcement learning has substantially improved the performance of LLM agents on tasks with verifiable outcomes, but it still struggles on open-ended agent tasks with vast solution spaces (e.g., complex travel planning). Due to the absence of objective ground-truth for these tasks, current RL algorithms largely rely on reward models that assign scalar scores to individual responses. We contend that such pointwise scoring suffers from an inherent discrimination collapse: the reward model struggles to distinguish subtle advantages among different trajectories, resulting in scores within a group being compressed into a narrow range. Consequently, the effective reward signal becomes dominated by noise from the reward model, leading to optimization stagnation. To address this, we propose ArenaRL, a reinforcement learning paradigm that shifts from pointwise scalar scoring to intra-group relative ranking. ArenaRL introduces a process-aware pairwise evaluation mechanism, employing multi-level rubrics to assign fine-grained relative scores to trajectories. Additionally, we construct an intra-group adversarial arena and devise a tournament-based ranking scheme to obtain stable advantage signals. Empirical results confirm that the built seeded single-elimination scheme achieves nearly equivalent advantage estimation accuracy to full pairwise comparisons with O(N^2) complexity, while operating with only O(N) complexity, striking an optimal balance between efficiency and precision. Furthermore, to address the lack of full-cycle benchmarks for open-ended agents, we build Open-Travel and Open-DeepResearch, two high-quality benchmarks featuring a comprehensive pipeline covering SFT, RL training, and multi-dimensional evaluation. Extensive experiments show that ArenaRL substantially outperforms standard RL baselines, enabling LLM agents to generate more robust solutions for complex real-world tasks.
Authors: Sang T. Truong, Duc Q. Nguyen, Willie Neiswanger, Ryan-Rhys Griffiths, Stefano Ermon, Nick Haber, Sanmi Koyejo
Abstract: Bayesian optimization (BO) is a common framework for optimizing black-box functions, yet most existing methods assume static query costs and rely on myopic acquisition strategies. We introduce LookaHES, a nonmyopic BO framework designed for dynamic, history-dependent cost environments, where evaluation costs vary with prior actions, such as travel distance in spatial tasks or edit distance in sequence design. LookaHES combines a multi-step variant of $H$-Entropy Search with pathwise sampling and neural policy optimization, enabling long-horizon planning beyond twenty steps without the exponential complexity of existing nonmyopic methods. The key innovation is the integration of neural policies, including large language models, to effectively navigate structured, combinatorial action spaces such as protein sequences. These policies amortize lookahead planning and can be integrated with domain-specific constraints during rollout. Empirically, LookaHES outperforms strong myopic and nonmyopic baselines across nine synthetic benchmarks from two to eight dimensions and two real-world tasks: geospatial optimization using NASA night-light imagery and protein sequence design with constrained token-level edits. In short, LookaHES provides a general, scalable, and cost-aware solution for robust long-horizon optimization in complex decision spaces, which makes it a useful tool for researchers in machine learning, statistics, and applied domains. Our implementation is available at https://github.com/sangttruong/nonmyopia.
Authors: Stavros Tsimpoukis, Dimitrios Tyrovolas, Sotiris Ioannidis, Maria Kafesaki, Ian F. Akyildiz, George K. Karagiannidis, Christos K. Liaskos
Abstract: Contemporary industrial Non-Destructive Inspection (NDI) methods require sensing capabilities that operate in occluded, hazardous, or access restricted environments. Yet, the current visual inspection based on optical cameras offers limited quality of service to that respect. In that sense, novel methods for workpiece inspection, suitable, for smart manufacturing are needed. Programmable Wireless Environments (PWE) could help towards that direction, by redefining the wireless Radio Frequency (RF) wave propagation as a controllable inspector entity. In this work, we propose a novel approach to Non-Destructive Inspection, leveraging an RF sensing pipeline based on RF wavefront encoding for retrieving workpiece-image entries from a designated database. This approach combines PWE-enabled RF wave manipulation with machine learning (ML) tools trained to produce visual outputs for quality inspection. Specifically, we establish correlation relationships between RF wavefronts and target industrial assets, hence yielding a dataset which links wavefronts to their corresponding images in a structured manner. Subsequently, a Generative Adversarial Network (GAN) derives visual representations closely matching the database entries. Our results indicate that the proposed method achieves an SSIM 99.5% matching score in visual outputs, paving the way for next-generation quality control workflows in industry.
Authors: Bowen Zhang, Hongda Tian, Adam Berry, A. Craig Roussac
Abstract: Accurate forecasting of the grid carbon intensity factor (CIF) is critical for enabling demand-side management and reducing emissions in modern electricity systems. Leveraging multiple interrelated time series, CIF prediction is typically formulated as a multivariate time series forecasting problem. Despite advances in deep learning-based methods, it remains challenging to capture the fine-grained local-temporal dependencies, dynamic higher-order cross-variable dependencies, and complex multi-frequency patterns for CIF forecasting. To address these issues, we propose a novel model that integrates two parallel modules: 1) one enhances the extraction of local-temporal dependencies under multi-frequency by applying multiple wavelet-based convolutional kernels to overlapping patches of varying lengths; 2) the other captures dynamic cross-variable dependencies under multi-frequency to model how inter-variable relationships evolve across the time-frequency domain. Evaluations on four representative electricity markets from Australia, featuring varying levels of renewable penetration, demonstrate that the proposed method outperforms the state-of-the-art models. An ablation study further validates the complementary benefits of the two proposed modules. Designed with built-in interpretability, the proposed model also enables better understanding of its predictive behavior, as shown in a case study where it adaptively shifts attention to relevant variables and time intervals during a disruptive event.
Authors: Qi Dong, Rubing Huang, Ling Zhou, Dave Towey, Jinyu Tian, Jianzhou Wang
Abstract: Diffusion models have emerged as a powerful method in various applications. However, their application to Short-Term Electricity Load Forecasting (STELF) -- a typical scenario in energy systems -- remains largely unexplored. Considering the nonlinear and fluctuating characteristics of the load data, effectively utilizing the powerful modeling capabilities of diffusion models to enhance STELF accuracy remains a challenge. This paper proposes a novel diffusion model with multi-frequency reconstruction for STELF, referred to as the Multi-Frequency-Reconstruction-based Diffusion (MFRD) model. The MFRD model achieves accurate load forecasting through four key steps: (1) The original data is combined with the decomposed multi-frequency modes to form a new data representation; (2) The diffusion model adds noise to the new data, effectively reducing and weakening the noise in the original data; (3) The reverse process adopts a denoising network that combines Long Short-Term Memory (LSTM) and Transformer to enhance noise removal; and (4) The inference process generates the final predictions based on the trained denoising network. To validate the effectiveness of the MFRD model, we conducted experiments on two data platforms: Australian Energy Market Operator (AEMO) and Independent System Operator of New England (ISO-NE). The experimental results show that our model consistently outperforms the compared models.
Authors: Liang Zheng, Bowen Shi, Yitao Hu, Jiawei Zhang, Ruofan Li, Sheng Chen, Wenxin Li, Keqiu Li
Abstract: Diffusion-based large language models (dLLMs) have emerged as a promising paradigm, utilizing simultaneous denoising to enable global planning and iterative refinement. While these capabilities are particularly advantageous for long-context generation, deploying such models faces a prohibitive memory capacity barrier stemming from severe system inefficiencies. We identify that existing inference systems are ill-suited for this paradigm: unlike autoregressive models constrained by the cumulative KV-cache, dLLMs are bottlenecked by transient activations recomputed at every step. Furthermore, general-purpose memory reuse mechanisms lack the global visibility to adapt to dLLMs' dynamic memory peaks, which toggle between logits and FFNs. To address these mismatches, we propose Mosaic, a memory-efficient inference system that shifts from local, static management to a global, dynamic paradigm. Mosaic integrates a mask-only logits kernel to eliminate redundancy, a lazy chunking optimizer driven by an online heuristic search to adaptively mitigate dynamic peaks, and a global memory manager to resolve fragmentation via virtual addressing. Extensive evaluations demonstrate that Mosaic achieves an average 2.71$\times$ reduction in the memory peak-to-average ratio and increases the maximum inference sequence length supportable on identical hardware by 15.89-32.98$\times$. This scalability is achieved without compromising accuracy and speed, and in fact reducing latency by 4.12%-23.26%.
Authors: Huyen Khanh Vo, Isabel Valera
Abstract: Multimodal variational autoencoders (VAEs) are widely used for weakly supervised generative learning with multiple modalities. Predominant methods aggregate unimodal inference distributions using either a product of experts (PoE), a mixture of experts (MoE), or their combinations to approximate the joint posterior. In this work, we revisit multimodal inference through the lens of probabilistic opinion pooling, an optimization-based approach. We start from H\"older pooling with $\alpha=0.5$, which corresponds to the unique symmetric member of the $\alpha\text{-divergence}$ family, and derive a moment-matching approximation, termed Hellinger. We then leverage such an approximation to propose HELVAE, a multimodal VAE that avoids sub-sampling, yielding an efficient yet effective model that: (i) learns more expressive latent representations as additional modalities are observed; and (ii) empirically achieves better trade-offs between generative coherence and quality, outperforming state-of-the-art multimodal VAE models.
Authors: Maciej Glowacki
Abstract: Learning robust and generalisable abstractions from high-dimensional input data is a central challenge in machine learning and its applications to high-energy physics (HEP). Solutions of lower functional complexity are known to produce abstractions that generalise more effectively and are more robust to input perturbations. In complex hypothesis spaces, inductive biases make such solutions learnable by shaping the loss geometry during optimisation. In a HEP classification task, we show that a soft symmetry respecting inductive bias creates approximate degeneracies in the loss, which we identify as pseudo-Goldstone modes. We quantify functional complexity using metrics derived from first principles Hessian analysis and via compressibility. Our results demonstrate that solutions of lower complexity give rise to abstractions that are more generalisable, robust, and efficiently distillable.
Authors: Nicola Aladrah, Emanuele Ballarin, Matteo Biagetti, Alessio Ansuini, Alberto d'Onofrio, Fabio Anselmi
Abstract: A central problem in machine learning theory is to characterize how learning dynamics select particular solutions among the many compatible with the training objective, a phenomenon, called implicit bias, which remains only partially characterized. In the present work, we identify a general mechanism, in terms of an explicit geometric correction of the learning dynamics, for the emergence of implicit biases, arising from the interaction between continuous symmetries in the model's parametrization and stochasticity in the optimization process. Our viewpoint is constructive in two complementary directions: given model symmetries, one can derive the implicit bias they induce; conversely, one can inverse-design a wide class of different implicit biases by computing specific redundant parameterizations. More precisely, we show that, when the dynamics is expressed in the quotient space obtained by factoring out the symmetry group of the parameterization, the resulting stochastic differential equation gains a closed form geometric correction in the stationary distribution of the optimizer dynamics favoring orbits with small local volume. We compute the resulting symmetry induced bias for a range of architectures, showing how several well known results fit into a single unified framework. The approach also provides a practical methodology for deriving implicit biases in new settings, and it yields concrete, testable predictions that we confirm by numerical simulations on toy models trained on synthetic data, leaving more complex scenarios for future work. Finally, we test the implicit bias inverse-design procedure in notable cases, including biases toward sparsity in linear features or in spectral properties of the model parameters.
Authors: Rishiraj Saha Roy, Chris Hinze, Luzian Hahn, Fabian Kuech
Abstract: We demonstrate CEDAR, an application for automating data science (DS) tasks with an agentic setup. Solving DS problems with LLMs is an underexplored area that has immense market value. The challenges are manifold: task complexities, data sizes, computational limitations, and context restrictions. We show that these can be alleviated via effective context engineering. We first impose structure into the initial prompt with DS-specific input fields, that serve as instructions for the agentic system. The solution is then materialized as an enumerated sequence of interleaved plan and code blocks generated by separate LLM agents, providing a readable structure to the context at any step of the workflow. Function calls for generating these intermediate texts, and for corresponding Python code, ensure that data stays local, and only aggregate statistics and associated instructions are injected into LLM prompts. Fault tolerance and context management are introduced via iterative code generation and smart history rendering. The viability of our agentic data scientist is demonstrated using canonical Kaggle challenges.
Authors: Zhangqi Duan, Nigel Fernandez, Andrew Lan
Abstract: Open-ended tasks, such as coding problems that are common in computer science education, provide detailed insights into student knowledge. However, training large language models (LLMs) to simulate and predict possible student errors in their responses to these problems can be challenging: they often suffer from mode collapse and fail to fully capture the diversity in syntax, style, and solution approach in student responses. In this work, we present KASER (Knowledge-Aligned Student Error Simulator), a novel approach that aligns errors with student knowledge. We propose a training method based on reinforcement learning using a hybrid reward that reflects three aspects of student code prediction: i) code similarity to the ground-truth, ii) error matching, and iii) code prediction diversity. On two real-world datasets, we perform two levels of evaluation and show that: At the per-student-problem pair level, our method outperforms baselines on code and error prediction; at the per-problem level, our method outperforms baselines on error coverage and simulated code diversity.
Authors: Quan Minh Nguyen, Min-Seon Kim, Hoang M. Ngo, Trong Nghia Hoang, Hyuk-Yoon Kwon, My T. Thai
Abstract: Membership inference attack (MIA) poses a significant privacy threat in federated learning (FL) as it allows adversaries to determine whether a client's private dataset contains a specific data sample. While defenses against membership inference attacks in standard FL have been well studied, the recent shift toward federated fine-tuning has introduced new, largely unexplored attack surfaces. To highlight this vulnerability in the emerging FL paradigm, we demonstrate that federated prompt-tuning, which adapts pre-trained models with small input prefixes to improve efficiency, also exposes a new vector for privacy attacks. We propose PromptMIA, a membership inference attack tailored to federated prompt-tuning, in which a malicious server can insert adversarially crafted prompts and monitors their updates during collaborative training to accurately determine whether a target data point is in a client's private dataset. We formalize this threat as a security game and empirically show that PromptMIA consistently attains high advantage in this game across diverse benchmark datasets. Our theoretical analysis further establishes a lower bound on the attack's advantage which explains and supports the consistently high advantage observed in our empirical results. We also investigate the effectiveness of standard membership inference defenses originally developed for gradient or output based attacks and analyze their interaction with the distinct threat landscape posed by PromptMIA. The results highlight non-trivial challenges for current defenses and offer insights into their limitations, underscoring the need for defense strategies that are specifically tailored to prompt-tuning in federated settings.
Authors: Joe Dwyer
Abstract: Research in machine learning has questioned whether increases in training token counts reliably produce proportional performance gains in large language models. Building on prior work introducing an energy-aware parameter efficiency metric, this study empirically examines the effects of increasing training token counts under fixed hardware and training conditions. The significance of this work lies in the explicit integration of power consumption and execution duration, as reflected by the power sampling frequency, into token-scale analysis. This addresses a gap in prior studies emphasizing performance outcomes while underrepresenting computational and energy costs. Using a repeated-measures experimental design on a constant GPU instance with an identical model architecture, optimizer settings, and epoch counts, a 1.1-billion-parameter TinyLlama model was trained at three token counts (500K, 1M, and 2M). While conventional performance metrics exhibited inconsistent or diminishing returns across token scales, the inclusion of power consumption and execution duration revealed a strictly monotonic decline in training efficiency as token count increased. Repeated-measures ANOVA demonstrated a strong effect of token count on parameter efficiency, with all pairwise comparisons remaining significant following Bonferroni correction. These findings indicate that increases in training token counts may be energetically inefficient even when marginal performance improvements are observed, underscoring the importance of efficiency-aware evaluation in large language model training.
Authors: Md Nafees Fuad Rafi, Samiul Hasan
Abstract: Real-time traffic prediction is critical for managing transportation systems during hurricane evacuations. Although data-driven graph-learning models have demonstrated strong capabilities in capturing the complex spatiotemporal dynamics of evacuation traffic at a network level, they mostly consider a single dimension (e.g., travel-time or distance) to construct the underlying graph. Furthermore, these models often lack interpretability, offering little insight into which input variables contribute most to their predictive performance. To overcome these limitations, we develop a novel Reinforcement Learning-guided Dynamic Multi-Graph Fusion (RL-DMF) framework for evacuation traffic prediction. We construct multiple dynamic graphs at each time step to represent heterogeneous spatiotemporal relationships between traffic detectors. A dynamic multi-graph fusion (DMF) module is employed to adaptively learn and combine information from these graphs. To enhance model interpretability, we introduce RL-based intelligent feature selection and ranking (RL-IFSR) method that learns to mask irrelevant features during model training. The model is evaluated using a real-world dataset of 12 hurricanes affecting Florida from 2016 to 2024. For an unseen hurricane (Milton, 2024), the model achieves a 95% accuracy (RMSE = 293.9) for predicting the next 1-hour traffic flow. Moreover, the model can forecast traffic flow for up to next 6 hours with 90% accuracy (RMSE = 426.4). The RL-DMF framework outperforms several state-of-the-art traffic prediction models. Furthermore, ablation experiments confirm the effectiveness of dynamic multi-graph fusion and RL-IFSR approaches for improving model performance. This research provides a generalized and interpretable model for real-time evacuation traffic forecasting, with significant implications for evacuation traffic management.
Authors: Zohaib Khan, Omer Tafveez, Zoha Hayat Bhatti
Abstract: Recent advances in mathematical reasoning typically rely on massive scale, yet the question remains: can strong reasoning capabilities be induced in small language models ($\leq1.5\text{B}$) under extreme constraints? We investigate this by training models on a single A40 GPU (48GB) for under 24 hours using Reinforcement Learning with Verifiable Rewards (RLVR) and Low-Rank Adaptation (LoRA). We find that the success of this ``micro-budget" regime depends critically on the interplay between adapter capacity and model initialization. While low-rank adapters ($r=8$) consistently fail to capture the complex optimization dynamics of reasoning, high-rank adapters ($r=256$) unlock significant plasticity in standard instruction-tuned models. Our best result achieved an impressive 40.0\% Pass@1 on AIME 24 (an 11.1\% absolute improvement over baseline) and pushed Pass@16 to 70.0\%, demonstrating robust exploration capabilities. However, this plasticity is not universal: while instruction-tuned models utilized the budget to elongate their chain-of-thought and maximize reward, heavily math-aligned models suffered performance collapse, suggesting that noisy, low-budget RL updates can act as destructive interference for models already residing near a task-specific optimum.
Authors: Poushali Sengupta, Rabindra Khadka, Sabita Maharjan, Frank Eliassen, Yan Zhang, Shashi Raj Pandey, Pedro G. Lind, Anis Yazidi
Abstract: Complex AI systems make better predictions but often lack transparency, limiting trustworthiness, interpretability, and safe deployment. Common post hoc AI explainers, such as LIME, SHAP, HSIC, and SAGE, are model agnostic but are too restricted in one significant regard: they tend to misrank correlated features and require costly perturbations, which do not scale to high dimensional data. We introduce ExCIR (Explainability through Correlation Impact Ratio), a theoretically grounded, simple, and reliable metric for explaining the contribution of input features to model outputs, which remains stable and consistent under noise and sampling variations. We demonstrate that ExCIR captures dependencies arising from correlated features through a lightweight single pass formulation. Experimental evaluations on diverse datasets, including EEG, synthetic vehicular data, Digits, and Cats-Dogs, validate the effectiveness and stability of ExCIR across domains, achieving more interpretable feature explanations than existing methods while remaining computationally efficient. To this end, we further extend ExCIR with an information theoretic foundation that unifies the correlation ratio with Canonical Correlation Analysis under mutual information bounds, enabling multi output and class conditioned explainability at scale.
Authors: Dushan N. Wadduwage, Dineth Jayakody, Leonidas Zimianitis
Abstract: Machine learning (ML) models show strong promise for new biomedical prediction tasks, but concerns about trustworthiness have hindered their clinical adoption. In particular, it is often unclear whether a model relies on true clinical cues or on spurious hierarchical correlations in the data. This paper introduces a simple yet broadly applicable trustworthiness test grounded in stochastic proof-by-contradiction. Instead of just showing high test performance, our approach trains and tests on spurious labels carefully permuted based on a potential outcomes framework. A truly trustworthy model should fail under such label permutation; comparable accuracy across real and permuted labels indicates overfitting, shortcut learning, or data leakage. Our approach quantifies this behavior through interpretable Fisher-style p-values, which are well understood by domain experts across medical and life sciences. We evaluate our approach on multiple new bacterial diagnostics to separate tasks and models learning genuine causal relationships from those driven by dataset artifacts or statistical coincidences. Our work establishes a foundation to build rigor and trust between ML and life-science research communities, moving ML models one step closer to clinical adoption.
Authors: Anca Muresan, Mihaela Cardei, Ionut Cardei
Abstract: Early identification of student success is crucial for enabling timely interventions, reducing dropout rates, and promoting on time graduation. In educational settings, AI powered systems have become essential for predicting student performance due to their advanced analytical capabilities. However, effectively leveraging diverse student data to uncover latent and complex patterns remains a key challenge. While prior studies have explored this area, the potential of dynamic data features and multi category entities has been largely overlooked. To address this gap, we propose a framework that integrates heterogeneous graph deep learning models to enhance early and continuous student performance prediction, using traditional machine learning algorithms for comparison. Our approach employs a graph metapath structure and incorporates dynamic assessment features, which progressively influence the student success prediction task. Experiments on the Open University Learning Analytics (OULA) dataset demonstrate promising results, achieving a 68.6% validation F1 score with only 7% of the semester completed, and reaching up to 89.5% near the semester's end. Our approach outperforms top machine learning models by 4.7% in validation F1 score during the critical early 7% of the semester, underscoring the value of dynamic features and heterogeneous graph representations in student success prediction.
Authors: Harsh Parikh
Abstract: The Rashomon effect -- the existence of multiple, distinct models that achieve nearly equivalent predictive performance -- has emerged as a fundamental phenomenon in modern machine learning and statistics. In this paper, we explore the causes underlying the Rashomon effect, organizing them into three categories: statistical sources arising from finite samples and noise in the data-generating process; structural sources arising from non-convexity of optimization objectives and unobserved variables that create fundamental non-identifiability; and procedural sources arising from limitations of optimization algorithms and deliberate restrictions to suboptimal model classes. We synthesize insights from machine learning, statistics, and optimization literature to provide a unified framework for understanding why the multiplicity of good models arises. A key distinction emerges: statistical multiplicity diminishes with more data, structural multiplicity persists asymptotically and cannot be resolved without different data or additional assumptions, and procedural multiplicity reflects choices made by practitioners. Beyond characterizing causes, we discuss both the challenges and opportunities presented by the Rashomon effect, including implications for inference, interpretability, fairness, and decision-making under uncertainty.
Authors: Anay Sinhal, Arpana Sinhal, Amit Sinhal
Abstract: Deep learning models for radiology interpretation increasingly rely on multi-institutional data, yet privacy regulations and distribution shift across hospitals limit central data pooling. Federated learning (FL) allows hospitals to collaboratively train models without sharing raw images, but current FL algorithms typically assume a static data distribution. In practice, hospitals experience continual evolution in case mix, annotation protocols, and imaging devices, which leads to catastrophic forgetting when models are updated sequentially. Federated continual learning (FCL) aims to reconcile these challenges but existing methods either ignore the stringent privacy constraints of healthcare or rely on replay buffers and public surrogate datasets that are difficult to justify in clinical settings. We study FCL for chest radiography classification in a setting where hospitals are clients that receive temporally evolving streams of cases and labels. We introduce DP-FedEPC (Differentially Private Federated Elastic Prototype Consolidation), a method that combines elastic weight consolidation (EWC), prototype-based rehearsal, and client-side differential privacy within a standard FedAvg framework. EWC constrains updates along parameters deemed important for previous tasks, while a memory of latent prototypes preserves class structure without storing raw images. Differentially private stochastic gradient descent (DP-SGD) at each client adds calibrated Gaussian noise to clipped gradients, providing formal privacy guarantees for individual radiographs.
Authors: Sofiia Huraka, Vakhtang Putkaradze
Abstract: Wide-spread adoption of unmanned vehicle technologies requires the ability to predict the motion of the combined vehicle operation from observations. While the general prediction of such motion for an arbitrary control mechanism is difficult, for a particular choice of control, the dynamics reduces to the Lie-Poisson equations [33,34]. Our goal is to learn the phase-space dynamics and predict the motion solely from observations, without any knowledge of the control Hamiltonian or the nature of interaction between vehicles. To achieve that goal, we propose the Control Optimal Lie-Poisson Neural Networks (CO-LPNets) for learning and predicting the dynamics of the system from data. Our methods learn the mapping of the phase space through the composition of Poisson maps, which are obtained as flows from Hamiltonians that could be integrated explicitly. CO-LPNets preserve the Poisson bracket and thus preserve Casimirs to machine precision. We discuss the completeness of the derived neural networks and their efficiency in approximating the dynamics. To illustrate the power of the method, we apply these techniques to systems of $N=3$ particles evolving on ${\rm SO}(3)$ group, which describe coupled rigid bodies rotating about their center of mass, and ${\rm SE}(3)$ group, applicable to the movement of unmanned air and water vehicles. Numerical results demonstrate that CO-LPNets learn the dynamics in phase space from data points and reproduce trajectories, with good accuracy, over hundreds of time steps. The method uses a limited number of points ($\sim200$/dimension) and parameters ($\sim 1000$ in our case), demonstrating potential for practical applications and edge deployment.
Authors: Min Chen, Zihan Wang, Canyu Chen, Zeguan Wu, Manling Li, Junyu Liu
Abstract: Large language models (LLMs) can be adapted to new tasks using parameter-efficient fine-tuning (PEFT) methods that modify only a small number of trainable parameters, often through low-rank updates. In this work, we adopt a quantum-information-inspired perspective to understand their effectiveness. From this perspective, low-rank parameterizations naturally correspond to low-dimensional Matrix Product States (MPS) representations, which enable entanglement-based characterizations of parameter structure. Thereby, we term and measure "Artificial Entanglement", defined as the entanglement entropy of the parameters in artificial neural networks (in particular the LLMs). We first study the representative low-rank adaptation (LoRA) PEFT method, alongside full fine-tuning (FFT), using LLaMA models at the 1B and 8B scales trained on the Tulu3 and OpenThoughts3 datasets, and uncover: (i) Internal artificial entanglement in the updates of query and value projection matrices in LoRA follows a volume law with a central suppression (termed as the "Entanglement Valley"), which is sensitive to hyper-parameters and is distinct from that in FFT; (ii) External artificial entanglement in attention matrices, corresponding to token-token correlations in representation space, follows an area law with logarithmic corrections and remains robust to LoRA hyper-parameters and training steps. Drawing a parallel to the No-Hair Theorem in black hole physics, we propose that although LoRA and FFT induce distinct internal entanglement signatures, such differences do not manifest in the attention outputs, suggesting a "no-hair" property that results in the effectiveness of low rank updates. We further provide theoretical support based on random matrix theory, and extend our analysis to an MPS Adaptation PEFT method, which exhibits qualitatively similar behaviors.
Authors: Malavika Pradeep, Akshay Sasi, Nusaibah Farrukh, Rahul Venugopal, Elizabeth Sherly
Abstract: The electroencephalogram (EEG) has been the gold standard for quantifying mental workload; however, due to its complexity and non-portability, it can be constraining. ECG signals, which are feasible on wearable equipment pieces such as headbands, present a promising method for cognitive state monitoring. This research explores whether electrocardiogram (ECG) signals are able to indicate mental workload consistently and act as surrogates for EEG-based cognitive indicators. This study investigates whether ECG-derived features can serve as surrogate indicators of cognitive load, a concept traditionally quantified using EEG. Using a publicly available multimodal dataset (OpenNeuro) of EEG and ECG recorded during working-memory and listening tasks, features of HRV and Catch22 descriptors are extracted from ECG, and spectral band-power with Catch22 features from EEG. A cross-modal regression framework based on XGBoost was trained to map ECG-derived HRV representations to EEG-derived cognitive features. In order to address data sparsity and model brain-heart interactions, we integrated the PSV-SDG to produce EEG-conditioned synthetic HRV time series.This addresses the challenge of inferring cognitive load solely from ECG-derived features using a combination of multimodal learning, signal processing, and synthetic data generation. These outcomes form a basis for light, interpretable machine learning models that are implemented through wearable biosensors in non-lab environments. Synthetic HRV inclusion enhances robustness, particularly in sparse data situations. Overall, this work is an initiation for building low-cost, explainable, and real-time cognitive monitoring systems for mental health, education, and human-computer interaction, with a focus on ageing and clinical populations.
Authors: Hoang Hiep Trieu
Abstract: Financial fraud is always a major problem in the field of finance, as it can cause significant consequences. As a result, many approaches have been designed to detect it, and lately Graph Neural Networks (GNNs) have been demonstrated as a competent candidate. However, when trained with a large amount of data, they are slow and computationally demanding. In addition, GNNs may need a deep architecture to detect complex fraud patterns, but doing so may make them suffer from problems such as over-fitting or over-smoothing. Over-fitting leads to reduced generalisation of the model on unseen data, while over-smoothing causes all nodes' features to converge to a fixed point due to excessive aggregation of information from neighbouring nodes. In this research, I propose an approach called One-Side Edge Sampling (OES) that can potentially reduce training duration as well as the effects of over-smoothing and over-fitting. The approach leverages predictive confidence in an edge classification task to sample edges from the input graph during a certain number of epochs. To explain why OES can alleviate over-smoothing, I perform a theoretical analysis of the proposed approach. In addition, to validate the effect of OES, I conduct experiments using different GNNs on two datasets. The results show that OES can empirically outperform backbone models in both shallow and deep architectures while also reducing training time.
Authors: Qiangwei Peng, Zihan Wang, Junda Ying, Yuhao Sun, Qing Nie, Lei Zhang, Tiejun Li, Peijie Zhou
Abstract: The Wasserstein-Fisher-Rao (WFR) metric extends dynamic optimal transport (OT) by coupling displacement with change of mass, providing a principled geometry for modeling unbalanced snapshot dynamics. Existing WFR solvers, however, are often unstable, computationally expensive, and difficult to scale. Here we introduce WFR Flow Matching (WFR-FM), a simulation-free training algorithm that unifies flow matching with dynamic unbalanced OT. Unlike classical flow matching which regresses only a transport vector field, WFR-FM simultaneously regresses a vector field for displacement and a scalar growth rate function for birth-death dynamics, yielding continuous flows under the WFR geometry. Theoretically, we show that minimizing the WFR-FM loss exactly recovers WFR geodesics. Empirically, WFR-FM yields more accurate and robust trajectory inference in single-cell biology, reconstructing consistent dynamics with proliferation and apoptosis, estimating time-varying growth fields, and applying to generative dynamics under imbalanced data. It outperforms state-of-the-art baselines in efficiency, stability, and reconstruction accuracy. Overall, WFR-FM establishes a unified and efficient paradigm for learning dynamical systems from unbalanced snapshots, where not only states but also mass evolve over time.
Authors: Toru Yoshinaga, Yasushi Kawase
Abstract: The field of algorithms with predictions aims to improve algorithm performance by integrating machine learning predictions into algorithm design. A central question in this area is how predictions can improve performance, and a key aspect of this analysis is the role of prediction accuracy. In this context, prediction accuracy is defined as a guaranteed probability that an instance drawn from the distribution belongs to the predicted set. As a performance measure that incorporates prediction accuracy, we focus on the distributionally-robust competitive ratio (DRCR), introduced by Sun et al.~(ICML 2024). The DRCR is defined as the expected ratio between the algorithm's cost and the optimal cost, where the expectation is taken over the worst-case instance distribution that satisfies the given prediction and accuracy requirement. A known structural property is that, for any fixed algorithm, the DRCR decreases linearly as prediction accuracy increases. Building on this result, we establish that the optimal DRCR value (i.e., the infimum over all algorithms) is a monotone and concave function of prediction accuracy. We further generalize the DRCR framework to a multiple-prediction setting and show that monotonicity and concavity are preserved in this setting. Finally, we apply our results to the ski rental problem, a benchmark problem in online optimization, to identify the conditions on prediction accuracies required for the optimal DRCR to attain a target value. Moreover, we provide a method for computing the critical accuracy, defined as the minimum accuracy required for the optimal DRCR to strictly improve upon the performance attainable without any accuracy guarantee.
Authors: Ioannis Ziogas, Aamna Al Shehhi, Ahsan H. Khandoker, Leontios J. Hadjileontiadis
Abstract: Understanding the structure of complex, nonstationary, high-dimensional time-evolving signals is a central challenge in scientific data analysis. In many domains, such as speech and biomedical signal processing, the ability to learn disentangled and interpretable representations is critical for uncovering latent generative mechanisms. Traditional approaches to unsupervised representation learning, including variational autoencoders (VAEs), often struggle to capture the temporal and spectral diversity inherent in such data. Here we introduce variational decomposition autoencoding (VDA), a framework that extends VAEs by incorporating a strong structural bias toward signal decomposition. VDA is instantiated through variational decomposition autoencoders (DecVAEs), i.e., encoder-only neural networks that combine a signal decomposition model, a contrastive self-supervised task, and variational prior approximation to learn multiple latent subspaces aligned with time-frequency characteristics. We demonstrate the effectiveness of DecVAEs on simulated data and three publicly available scientific datasets, spanning speech recognition, dysarthria severity evaluation, and emotional speech classification. Our results demonstrate that DecVAEs surpass state-of-the-art VAE-based methods in terms of disentanglement quality, generalization across tasks, and the interpretability of latent encodings. These findings suggest that decomposition-aware architectures can serve as robust tools for extracting structured representations from dynamic signals, with potential applications in clinical diagnostics, human-computer interaction, and adaptive neurotechnologies.
Authors: Xin Ye, Daning Cheng, Boyang Zhang, Yunquan Zhang
Abstract: Training large-scale Mixture-of-Experts (MoE) models typically requires high-memory, high-bandwidth GPUs (e.g., A100), and their high cost has become a major barrier to large-model training. In contrast, affordable hardware is low-cost but constrained by memory capacity and bandwidth, making it unsuitable for direct LLM training. To address this, we propose MoE-DisCo (Mixture-of-Experts with Disentangled Clustering and Coordination), a staged training framework. MoE-DisCo decomposes the MoE model into multiple dense submodels, each consisting of a shared backbone and a single expert, and partitions the training data into subsets using unsupervised clustering. Each submodel is trained independently and in parallel on its assigned data subset using low-cost devices, without any inter-device communication. Subsequently, all experts are integrated into a complete MoE model and fine-tuned globally for a short period on high-memory, high-bandwidth GPUs. Experiments show that our method matches or even surpasses full-parameter training in performance across multiple downstream tasks, loss function, and perplexity (PPL), while reducing training cost by 47.6 percent to 69.5 percent on Qwen1.5-MoE-2.7B and Llama-MoE-3.5B across different datasets.
Authors: Shiyuan Zhang, Yilai Liu, Yuwei Du, Ruoxuan Yang, Dong In Kim, Hongyang Du
Abstract: Personalized mobile artificial intelligence applications are widely deployed, yet they are expected to infer user behavior from sparse and irregular histories under a continuously evolving spatio-temporal context. This setting induces a fundamental tension among three requirements, i.e., immediacy to adapt to recent behavior, stability to resist transient noise, and generalization to support long-horizon prediction and cold-start users. Most existing approaches satisfy at most two of these requirements, resulting in an inherent impossibility triangle in data-scarce, non-stationary personalization. To address this challenge, we model mobile behavior as a partially observed spatio-temporal tensor and unify short-term adaptation, long-horizon forecasting, and cold-start recommendation as a conditional completion problem, where a user- and task-specific mask specifies which coordinates are treated as evidence. We propose U-MASK, a user-adaptive spatio-temporal masking method that allocates evidence budgets based on user reliability and task sensitivity. To enable mask generation under sparse observations, U-MASK learns a compact, task-agnostic user representation from app and location histories via U-SCOPE, which serves as the sole semantic conditioning signal. A shared diffusion transformer then performs mask-guided generative completion while preserving observed evidence, so personalization and task differentiation are governed entirely by the mask and the user representation. Experiments on real-world mobile datasets demonstrate consistent improvements over state-of-the-art methods across short-term prediction, long-horizon forecasting, and cold-start settings, with the largest gains under severe data sparsity. The code and dataset will be available at https://github.com/NICE-HKU/U-MASK.
Authors: Jiazhang Liang, Jianheng Dai, Miaosen Luo, Menghua Jiang, Sijie Mai
Abstract: Multimodal large language models (MLLMs) have demonstrated strong performance on vision-language tasks, yet their effectiveness on multimodal sentiment analysis remains constrained by the scarcity of high-quality training data, which limits accurate multimodal understanding and generalization. To alleviate this bottleneck, we leverage diffusion models to perform semantics-preserving augmentation on the video and audio modalities, expanding the multimodal training distribution. However, increasing data quantity alone is insufficient, as diffusion-generated samples exhibit substantial quality variation and noisy augmentations may degrade performance. We therefore propose DaQ-MSA (Denoising and Qualifying Diffusion Augmentations for Multimodal Sentiment Analysis), which introduces a quality scoring module to evaluate the reliability of augmented samples and assign adaptive training weights. By down-weighting low-quality samples and emphasizing high-fidelity ones, DaQ-MSA enables more stable learning. By integrating the generative capability of diffusion models with the semantic understanding of MLLMs, our approach provides a robust and generalizable automated augmentation strategy for training MLLMs without any human annotation or additional supervision.
Authors: Taehyun Hwang, Dahngoon Kim, Min-hwan Oh
Abstract: We study the multinomial logit (MNL) contextual bandit problem for sequential assortment selection. Although most existing research assumes utility functions to be linear in item features, this linearity assumption restricts the modeling of intricate interactions between items and user preferences. A recent work (Zhang & Luo, 2024) has investigated general utility function classes, yet its method faces fundamental trade-offs between computational tractability and statistical efficiency. To address this limitation, we propose a computationally efficient algorithm for MNL contextual bandits leveraging the upper confidence bound principle, specifically designed for non-linear parametric utility functions, including those modeled by neural networks. Under a realizability assumption and a mild geometric condition on the utility function class, our algorithm achieves a regret bound of $\tilde{O}(\sqrt{T})$, where $T$ denotes the total number of rounds. Our result establishes that sharp $\tilde{O}(\sqrt{T})$-regret is attainable even with neural network-based utilities, without relying on strong assumptions such as neural tangent kernel approximations. To the best of our knowledge, our proposed method is the first computationally tractable algorithm for MNL contextual bandits with non-linear utilities that provably attains $\tilde{O}(\sqrt{T})$ regret. Comprehensive numerical experiments validate the effectiveness of our approach, showing robust performance not only in realizable settings but also in scenarios with model misspecification.
Authors: Mohammed Azeez Khan, Aaron D'Souza, Vijay Choyal
Abstract: Efficient discovery of new materials demands strategies to reduce the number of costly first-principles calculations required to train predictive machine learning models. We develop and validate an active learning framework that iteratively selects informative training structures for machine-learned interatomic potentials (MLIPs) from large, heterogeneous materials databases, specifically the Materials Project and OQMD. Our framework integrates compositional and property-based descriptors with a neural network ensemble model, enabling real-time uncertainty quantification via Query-by-Committee. We systematically compare four selection strategies: random sampling (baseline), uncertainty-based sampling, diversity-based sampling (k-means clustering with farthest-point refinement), and a hybrid approach balancing both objectives. Experiments across four representative material systems (elemental carbon, silicon, iron, and a titanium-oxide compound) with 5 random seeds per configuration demonstrate that diversity sampling consistently achieves competitive or superior performance, with particularly strong advantages on complex systems like titanium-oxide (10.9% improvement, p=0.008). Our results show that intelligent data selection strategies can achieve target accuracy with 5-13% fewer labeled samples compared to random baselines. The entire pipeline executes on Google Colab in under 4 hours per system using less than 8 GB of RAM, thereby democratizing MLIP development for researchers globally with limited computational resources. Our open-source code and detailed experimental configurations are available on GitHub. This multi-system evaluation establishes practical guidelines for data-efficient MLIP training and highlights promising future directions including integration with symmetry-aware neural network architectures.
Authors: Heng Xu, Tianqing Zhu, Dayong Ye, Lefeng Zhang, Le Wang, Wanlei Zhou
Abstract: Machine unlearning, a process enabling pre-trained models to remove the influence of specific training samples, has attracted significant attention in recent years. Although extensive research has focused on developing efficient machine unlearning strategies, we argue that these methods mainly aim at removing samples rather than removing samples' influence on the model, thus overlooking the fundamental definition of machine unlearning. In this paper, we first conduct a comprehensive study to evaluate the effectiveness of existing unlearning schemes when the training dataset includes many samples similar to those targeted for unlearning. Specifically, we evaluate: Do existing unlearning methods truly adhere to the original definition of machine unlearning and effectively eliminate all influence of target samples when similar samples are present in the training dataset? Our extensive experiments, conducted on four carefully constructed datasets with thorough analysis, reveal a notable gap between the expected and actual performance of most existing unlearning methods for image and language models, even for the retraining-from-scratch baseline. Additionally, we also explore potential solutions to enhance current unlearning approaches.
Authors: James Tlhomole, Edoardo Borgomeo, Karthikeyan Matheswaran, Mariangel Garcia Andarcia
Abstract: Robust hydrological simulation is key for sustainable development, water management strategies, and climate change adaptation. In recent years, deep learning methods have been demonstrated to outperform mechanistic models at the task of hydrological discharge simulation. Adoption of these methods has been catalysed by the proliferation of large sample hydrology datasets, consisting of the observed discharge and meteorological drivers, along with geological and topographical catchment descriptors. Deep learning methods infer rainfall-runoff characteristics that have been shown to generalise across catchments, benefitting from the data diversity in large datasets. Despite this, application to catchments in Africa has been limited. The lack of adoption of deep learning methodologies is primarily due to sparsity or lack of the spatiotemporal observational data required to enable downstream model training. We therefore investigate the application of deep learning models, including LSTMs, for hydrological discharge simulation in the transboundary Limpopo River basin, emphasising application to data scarce regions. We conduct a number of computational experiments primarily focused on assessing the impact of varying the LSTM model input data on performance. Results confirm that data constraints remain the largest obstacle to deep learning applications across African river basins. We further outline the impact of human influence on data-driven modelling which is a commonly overlooked aspect of data-driven large-sample hydrology approaches and investigate solutions for model adaptation under smaller datasets. Additionally, we include recommendations for future efforts towards seasonal hydrological discharge prediction and direct comparison or inclusion of SWAT model outputs, as well as architectural improvements.
Authors: Vladimer Khasia
Abstract: Post-training quantization is essential for deploying Large Language Models (LLMs) on resource- constrained devices. However, standard integer quantization (e.g., INT4) fundamentally degrades per- formance by imposing a uniform grid on the heavy-tailed distribution of weight parameters, particularly in smaller-scale models (e.g., <2B parameters). We introduce HAS-VQ (Hessian-Adaptive Sparse Vec- tor Quantization), a compression framework that strictly decouples high-sensitivity outliers from the bulk weight distribution using second-order sensitivity analysis. HAS-VQ employs a Hessian-Masked Decoupling strategy to isolate sensitive parameters, followed by robust Vector Quantization (VQ) of the remaining dense body. Crucially, we introduce a residual sparse feedback mechanism that corrects quan- tization errors in the most sensitive dimensions, ensuring exact reconstruction of outliers. We evaluate HAS-VQ on SmolLM2-1.7B, demonstrating two distinct regimes of superiority: (1) Pareto Dominance over Integer Baselines: At 4.23 effective bits-per-parameter (BPP), we achieve a perplexity of 14.23, significantly outperforming the standard INT4 baseline (20.03 PPL at 4.71 BPP). (2) High-Fidelity Compression: Relative to the FP16 baseline, HAS-VQ achieves a 2.3x reduction in model size (7.03 BPP) while maintaining statistically indistinguishable perplexity (10.12 vs. 10.04), effectively offering a lossless compression alternative for bandwidth-constrained environments. The code is available at https://github.com/VladimerKhasia/HASVQ
Authors: Jinduo Guo, Yinzhi Cao
Abstract: The Newton method has been widely adopted to achieve certified unlearning. A critical assumption in existing approaches is that the data requested for unlearning are selected i.i.d.(independent and identically distributed). However,the problem of certified unlearning under non-i.i.d. deletions remains largely unexplored. In practice, unlearning requests are inherently biased, leading to non-i.i.d. deletions and causing distribution shifts between the original and retained datasets. In this paper, we show that certified unlearning with the Newton method becomes inefficient and ineffective under non-i.i.d. unlearning sets. We then propose a better certified unlearning approach by performing a distribution-aware certified unlearning framework based on iterative Newton updates constrained by a trust region. Our method provides a closer approximation to the retrained model and yields a tighter pre-run bound on the gradient residual, thereby ensuring efficient (epsilon, delta)-certified unlearning. To demonstrate its practical effectiveness under distribution shift, we also conduct extensive experiments across multiple evaluation metrics, providing a comprehensive assessment of our approach.
Authors: Lucas Versini, Paul Mangold, Aymeric Dieuleveut
Abstract: We propose a novel analysis of the Decentralized Stochastic Gradient Descent (DSGD) algorithm with constant step size, interpreting the iterates of the algorithm as a Markov chain. We show that DSGD converges to a stationary distribution, with its bias, to first order, decomposable into two components: one due to decentralization (growing with the graph's spectral gap and clients' heterogeneity) and one due to stochasticity. Remarkably, the variance of local parameters is, at the first-order, inversely proportional to the number of clients, regardless of the network topology and even when clients' iterates are not averaged at the end. As a consequence of our analysis, we obtain non-asymptotic convergence bounds for clients' local iterates, confirming that DSGD has linear speed-up in the number of clients, and that the network topology only impacts higher-order terms.
Authors: Hasan M Jamil
Abstract: Glioblastoma (GBM) is a highly aggressive primary brain tumor with limited therapeutic options and poor prognosis. The methylation status of the O6-methylguanine-DNA methyltransferase (MGMT) gene promoter is a critical molecular biomarker that influences patient response to temozolomide chemotherapy. Traditional methods for determining MGMT status rely on invasive biopsies and are limited by intratumoral heterogeneity and procedural risks. This study presents a radiogenomic molecular imaging analysis framework for the non-invasive prediction of MGMT promoter methylation using multi-parametric magnetic resonance imaging (mpMRI). Our approach integrates radiomics, deep learning, and explainable artificial intelligence (XAI) to analyze MRI-derived imaging phenotypes and correlate them with molecular labels. Radiomic features are extracted from FLAIR, T1-weighted, T1-contrast-enhanced, and T2-weighted MRI sequences, while a 3D convolutional neural network learns deep representations from the same modalities. These complementary features are fused using both early fusion and attention-based strategies and classified to predict MGMT methylation status. To enhance clinical interpretability, we apply XAI methods such as Grad-CAM and SHAP to visualize and explain model decisions. The proposed framework is trained on the RSNA-MICCAI Radiogenomic Classification dataset and externally validated on the BraTS 2021 dataset. This work advances the field of molecular imaging by demonstrating the potential of AI-driven radiogenomics for precision oncology, supporting non-invasive, accurate, and interpretable prediction of clinically actionable molecular biomarkers in GBM.
Authors: Aaron R. Flouro, Shawn P. Chadwick
Abstract: Benchmarks measure whether a model is correct. They do not measure whether a model is reliable. This distinction is largely academic for single-shot inference, but becomes critical for agentic AI systems, where a single rephrased prompt can trigger cascading failures in multi-step execution. Yet this form of instability is not captured by existing evaluations. Hallucinations live in variance: they arise when semantically equivalent prompts activate inconsistent internal pathways, producing divergent outputs. Consistent but incorrect outputs reflect bias or missing knowledge; confident guessing reflects calibration failure. Neither constitutes hallucination under this definition. When error is variance-dominated, reducing redundant pathways improves reliability without adding knowledge. We formalize this through Semantic Stability (SS), measured via Paraphrase Consistency (PC@k): generate k paraphrases, greedy decode each, compute mode agreement. SS is a diagnostic for variance-driven unreliability, not a method for improving correctness. We show that a dense Qwen3-0.6B agrees with itself only 23.8% of the time; at 32% sparsity, agreement jumps to 55.9%. A phase diagram reveals the sweet spot where variance reduction outpaces bias accumulation, and regimes where stability collapses onto wrong answers.
Authors: Dylan Sam, Sachin Goyal, Pratyush Maini, Alexander Robey, J. Zico Kolter
Abstract: Ensuring the safety of language models in high-stakes settings remains a pressing challenge, as aligned behaviors are often brittle and easily undone by adversarial pressure or downstream finetuning. Prior work has shown that interventions applied during pretraining, such as rephrasing harmful content, can substantially improve the safety of the resulting models. In this paper, we study the fundamental question: "When during pretraining should safety interventions be introduced?" We keep the underlying data fixed and vary only the choice of a safety curriculum: the timing of these interventions, i.e., after 0%, 20%, or 60% of the pretraining token budget. We find that introducing interventions earlier generally yields more robust models with no increase in overrefusal rates, with the clearest benefits appearing after downstream, benign finetuning. We also see clear benefits in the steerability of models towards safer generations. Finally, we observe that earlier interventions reshape internal representations: linear probes more cleanly separate safe vs harmful examples. Overall, these results argue for incorporating safety signals early in pretraining, producing models that are more robust to downstream finetuning and jailbreaking, and more reliable under both standard and safety-aware inference procedures.
Authors: Lucas Schott, Elies Gherbi, Hatem Hajri, Sylvain Lamprier
Abstract: Adversarial robustness in RL is difficult because perturbations affect entire trajectories: strong attacks can break learning, while weak attacks yield little robustness, and the appropriate strength varies by state. We propose $\alpha$-reward-preserving attacks, which adapt the strength of the adversary so that an $\alpha$ fraction of the nominal-to-worst-case return gap remains achievable at each state. In deep RL, we use a gradient-based attack direction and learn a state-dependent magnitude $\eta\le \eta_{\mathcal B}$ selected via a critic $Q^{\pi}_\alpha((s,a),\eta)$ trained off-policy over diverse radii. This adaptive tuning calibrates attack strength and, with intermediate $\alpha$, improves robustness across radii while preserving nominal performance, outperforming fixed- and random-radius baselines.
Authors: Sophie Sigfstead, River Jiang, Brianna Davies, Zachary W. M. Laksman, Julia Cadrin-Tourigny, Rafik Tadros, Habib Khan, Joseph Atallah, Christian Steinberg, Shubhayan Sanatani, Mario Talajic, Rahul Krishnan, Andrew D. Krahn, Christopher C. Cheung
Abstract: Arrhythmogenic right ventricular cardiomyopathy (ARVC) and long QT syndrome (LQTS) are inherited arrhythmia syndromes associated with sudden cardiac death. Deep learning shows promise for ECG interpretation, but multi-class inherited arrhythmia classification with clinically grounded interpretability remains underdeveloped. Our objective was to develop and validate a lead-aware deep learning framework for multi-class (ARVC vs LQTS vs control) and binary inherited arrhythmia classification, and to determine optimal strategies for integrating ECG foundation models within arrhythmia screening tools. We assembled a 13-center Canadian cohort (645 patients; 1,344 ECGs). We evaluated four ECG foundation models using three transfer learning approaches: linear probing, fine-tuning, and combined strategies. We developed lead-aware spatial attention networks (LASAN) and assessed integration strategies combining LASAN with foundation models. Performance was compared against the established foundation model baselines. Lead-group masking quantified disease-specific lead dependence. Fine-tuning outperformed linear probing and combined strategies across all foundation models (mean macro-AUROC 0.904 vs 0.825). The best lead-aware integrations achieved near-ceiling performance (HuBERT-ECG hybrid: macro-AUROC 0.990; ARVC vs control AUROC 0.999; LQTS vs control AUROC 0.994). Lead masking demonstrated physiologic plausibility: V1-V3 were most critical for ARVC detection (4.54% AUROC reduction), while lateral leads were preferentially important for LQTS (2.60% drop). Lead-aware architectures achieved state-of-the-art performance for inherited arrhythmia classification, outperforming all existing published models on both binary and multi-class tasks while demonstrating clinically aligned lead dependence. These findings support potential utility for automated ECG screening pending validation.
Authors: Ruhi Sayana, Kate Callon, Jennifer Xu, Jonathan Deutsch, Steven Chu, James Zou, John Janetzko, Rabindra V. Shivnaraine, Kyle Swanson
Abstract: Developing new fluorophores for advanced imaging techniques requires exploring new chemical space. While generative AI approaches have shown promise in designing novel dye scaffolds, prior efforts often produced synthetically intractable candidates due to a lack of reaction constraints. Here, we developed SyntheFluor-RL, a generative AI model that employs known reaction libraries and molecular building blocks to create readily synthesizable fluorescent molecule scaffolds via reinforcement learning. To guide the generation of fluorophores, SyntheFluor-RL employs a scoring function built on multiple graph neural networks (GNNs) that predict key photophysical properties, including photoluminescence quantum yield, absorption, and emission wavelengths. These outputs are dynamically weighted and combined with a computed pi-conjugation score to prioritize candidates with desirable optical characteristics and synthetic feasibility. SyntheFluor-RL generated 11,590 candidate molecules, which were filtered to 19 structures predicted to possess dye-like properties. Of the 19 molecules, 14 were synthesized and 13 were experimentally confirmed. The top three were characterized, with the lead compound featuring a benzothiadiazole chromophore and exhibiting strong fluorescence (PLQY = 0.62), a large Stokes shift (97 nm), and a long excited-state lifetime (11.5 ns). These results demonstrate the effectiveness of SyntheFluor-RL in the identification of synthetically accessible fluorophores for further development.
Authors: Ijun Jang, Jewon Yeom, Juan Yeo, Hyunggu Lim, Taesup Kim
Abstract: Knowledge distillation (KD) is a widely adopted technique for transferring knowledge from large language models to smaller student models; however, conventional supervised KD often suffers from a distribution mismatch between training and inference. While on-policy KD approaches attempt to mitigate this issue by learning directly from student-generated outputs, they frequently encounter training instabilities because the distributional gap between the novice student and the expert teacher is often too wide to bridge directly. These challenges manifest as pathological gradients in forward KL objectives or diversity collapse in reverse KL regimes. To address these limitations, we propose Veto, an objective-level reformulation that constructs a geometric bridge in the logit space. Unlike prior methods that mix data samples, Veto creates an intermediate target distribution that promotes alignment between the teacher and the student. By introducing a tunable parameter beta, Veto serves as an Adaptive Gradient Veto that stabilizes optimization by suppressing harmful gradients on low-confidence tokens, while simultaneously acting as a Decisiveness Knob to balance reward-driven performance with output diversity. Extensive experiments across various reasoning and generation tasks demonstrate that Veto consistently outperforms supervised fine-tuning and existing on-policy baselines.
Authors: Min Wang, Xin Li, Mingzhong Wang, Hasnaa Bennis
Abstract: Offline meta-reinforcement learning (OMRL) combines the strengths of learning from diverse datasets in offline RL with the adaptability to new tasks of meta-RL, promising safe and efficient knowledge acquisition by RL agents. However, OMRL still suffers extrapolation errors due to out-of-distribution (OOD) actions, compromised by broad task distributions and Markov Decision Process (MDP) ambiguity in meta-RL setups. Existing research indicates that the generalization of the $Q$ network affects the extrapolation error in offline RL. This paper investigates this relationship by decomposing the $Q$ value into feature and weight components, observing that while decomposition enhances adaptability and convergence in the case of high-quality data, it often leads to policy degeneration or collapse in complex tasks. We observe that decomposed $Q$ values introduce a large estimation bias when the feature encounters OOD samples, a phenomenon we term ''feature overgeneralization''. To address this issue, we propose FLORA, which identifies OOD samples by modeling feature distributions and estimating their uncertainties. FLORA integrates a return feedback mechanism to adaptively adjust feature components. Furthermore, to learn precise task representations, FLORA explicitly models the complex task distribution using a chain of invertible transformations. We theoretically and empirically demonstrate that FLORA achieves rapid adaptation and meta-policy improvement compared to baselines across various environments.
Authors: Ruiyi Ding, Yongxuan Lv, Xianhui Meng, Jiahe Song, Chao Wang, Chen Jiang, Yuan Cheng
Abstract: Policy optimization for large language models often suffers from sparse reward signals in multi-step reasoning tasks. Critic-free methods like GRPO assign a single normalized outcome reward to all tokens, providing limited guidance for intermediate reasoning . While Process Reward Models (PRMs) offer dense feedback, they risk premature collapse when used alone, as early low-reward tokens can drive policies toward truncated outputs. We introduce Process Relative Policy Optimization (PRPO), which combines outcome reliability with process-level guidance in a critic-free framework. PRPO segments reasoning sequences based on semantic clues, normalizes PRM scores into token-level advantages, and aligns their distribution with outcome advantages through location-parameter shift. On MATH500, PRPO improves Qwen2.5-Math-1.5B accuracy from 61.2% to 64.4% over GRPO using only eight rollouts and no value network, demonstrating efficient fine-grained credit assignment within critic-free optimization.
Authors: Susana Lopez-Moreno, Eric Dolores-Cuenca, Sangil Kim
Abstract: Reproducibility remains a challenge in machine learning research. While code and data availability requirements have become increasingly common, post-publication verification in journals is still limited and unformalized. This position paper argues that it is plausible for journals and conference proceedings to implement post-publication verification. We propose a modification to ACM pre-publication verification badges that allows independent researchers to submit post-publication code replications to the journal, leading to visible verification badges included in the article metadata. Each article may earn up to two badges, each linked to verified code in its corresponding public repository. We describe the motivation, related initiatives, a formal framework, the potential impact, possible limitations, and alternative views.
Authors: Ibne Farabi Shihab, Sanjeda Akter, Anuj Sharma
Abstract: Post-training activation compression is essential for deploying Large Language Models (LLMs) on resource-constrained hardware. However, standard methods like Singular Value Decomposition (SVD) are gradient-blind: they preserve high-variance dimensions regardless of their impact on factual knowledge preservation. We introduce Fisher-Aligned Subspace Compression (FASC), a knowledge-aware compression framework that selects subspaces by directly modeling activation-gradient coupling, minimizing a second-order surrogate of the loss function. FASC leverages the Fisher Information Matrix to identify dimensions critical for factual knowledge, which often reside in low-variance but high-gradient-sensitivity subspaces. We propose the Dependence Violation Score (\r{ho}) as a general-purpose diagnostic metric that quantifies activation-gradient coupling, revealing where factual knowledge is stored within transformer architectures. Extensive experiments on Mistral-7B and Llama-3-8B demonstrate that FASC preserves 6-8% more accuracy on knowledge-intensive benchmarks (MMLU, LAMA) compared to variance-based methods at 50% rank reduction, effectively enabling a 7B model to match the factual recall of a 13B uncompressed model. Our analysis reveals that \r{ho} serves as a fundamental signal of stored knowledge, with high-\r{ho} layers emerging only when models internalize factual associations during training.
Authors: Murtaza Nikzad, Raghuram Ramanujan
Abstract: Large language models exhibit impressive reasoning capabilities yet frequently generate plausible but incorrect solutions, a phenomenon commonly termed hallucination. This paper investigates the effect of training objective composition on reasoning reliability through Direct Preference Optimization. Two complementary training signals are examined: forward chain-of-thought generation, which trains the model to produce correct reasoning traces, and backward verification, which trains the model to verify and acknowledge errors in candidate solutions. Experiments on GSM8K reveal a fundamental trade-off between these objectives. Forward-only DPO training achieves the highest accuracy improvement, increasing from 83.1% to 86.6% (+3.5 percentage points), while backward-only training yields minimal accuracy gains but substantially reduces the false positive rate from 13.4% to 4.3%. Notably, both training variants reduce acknowledgement rate compared to the baseline, suggesting that preference optimization increases model confidence in its outputs. These findings indicate that forward and backward reasoning objectives provide distinct and complementary learning signals: forward training improves problem-solving capability, while backward training improves verification calibration. The complete training and evaluation pipeline, implemented efficiently through Low-Rank Adaptation, is released to facilitate further research.
Authors: Haozhong Wang, Zhuo Li, Yibo Yang, He Zhao, Hongyuan Zha, Dandan Guo
Abstract: The inherent safety alignment of Large Language Models (LLMs) is prone to erosion during fine-tuning, even when using seemingly innocuous datasets. While existing defenses attempt to mitigate this via data selection, they typically rely on heuristic, instance-level assessments that neglect the global geometry of the data distribution and fail to explicitly repel harmful patterns. To address this, we introduce Safety Optimal Transport (SOT), a novel framework that reframes safe fine-tuning from an instance-level filtering challenge to a distribution-level alignment task grounded in Optimal Transport (OT). At its core is a dual-reference ``push-pull'' weight-learning mechanism: SOT optimizes sample importance by actively pulling the downstream distribution towards a trusted safe anchor while simultaneously pushing it away from a general harmful reference. This establishes a robust geometric safety boundary that effectively purifies the training data. Extensive experiments across diverse model families and domains demonstrate that SOT significantly enhances model safety while maintaining competitive downstream performance, achieving a superior safety-utility trade-off compared to baselines.
Authors: Ibne Farabi Shihab, Sanjeda Akter, Anuj Sharma
Abstract: Deep protein structure predictors such as AlphaFold provide confidence estimates (e.g., pLDDT) that are often miscalibrated and degrade under distribution shifts across experimental modalities, temporal changes, and intrinsically disordered regions. We introduce CalPro, a prior-aware evidential-conformal framework for shift-robust uncertainty quantification. CalPro combines (i) a geometric evidential head that outputs Normal-Inverse-Gamma predictive distributions via a graph-based architecture; (ii) a differentiable conformal layer that enables end-to-end training with finite-sample coverage guarantees; and (iii) domain priors (disorder, flexibility) encoded as soft constraints. We derive structure-aware coverage guarantees under distribution shift using PAC-Bayesian bounds over ambiguity sets, and show that CalPro maintains near-nominal coverage while producing tighter intervals than standard conformal methods in regions where priors are informative. Empirically, CalPro exhibits at most 5% coverage degradation across modalities (vs. 15-25% for baselines), reduces calibration error by 30-50%, and improves downstream ligand-docking success by 25%. Beyond proteins, CalPro applies to structured regression tasks in which priors encode local reliability, validated on non-biological benchmarks.
Authors: Yang Zhao, Hepeng Wang, Xiao Ding, Yangou Ouyang, Bibo Cai, Kai Xiong, Jinglong Gao, Zhouhao Sun, Li Du, Bing Qin, Ting Liu
Abstract: Group-Relative Policy Optimization (GRPO) has emerged as an efficient paradigm for aligning Large Language Models (LLMs), yet its efficacy is primarily confined to domains with verifiable ground truths. Extending GRPO to open-domain settings remains a critical challenge, as unconstrained generation entails multi-faceted and often conflicting objectives - such as creativity versus factuality - where rigid, static reward scalarization is inherently suboptimal. To address this, we propose MAESTRO (Meta-learning Adaptive Estimation of Scalarization Trade-offs for Reward Optimization), which introduces a meta-cognitive orchestration layer that treats reward scalarization as a dynamic latent policy, leveraging the model's terminal hidden states as a semantic bottleneck to perceive task-specific priorities. We formulate this as a contextual bandit problem within a bi-level optimization framework, where a lightweight Conductor network co-evolves with the policy by utilizing group-relative advantages as a meta-reward signal. Across seven benchmarks, MAESTRO consistently outperforms single-reward and static multi-objective baselines, while preserving the efficiency advantages of GRPO, and in some settings even reducing redundant generation.
Authors: Mingnan Zhu, Qixuan Zhang, Yixuan Cheng, Fangzhou Gu, Shiming Lin
Abstract: Accurate energy time-series forecasting is crucial for ensuring grid stability and promoting the integration of renewable energy, yet it faces significant challenges from complex temporal dependencies and the heterogeneity of multi-source data. To address these issues, we propose DDT, a novel and robust deep learning framework for high-precision time-series forecasting. At its core, DDT introduces two key innovations. First, we design a dual-masking mechanism that synergistically combines a strict causal mask with a data-driven dynamic mask. This novel design ensures theoretical causal consistency while adaptively focusing on the most salient historical information, overcoming the rigidity of traditional masking techniques. Second, our architecture features a dual-expert system that decouples the modeling of temporal dynamics and cross-variable correlations into parallel, specialized pathways, which are then intelligently integrated through a dynamic gated fusion module. We conducted extensive experiments on 7 challenging energy benchmark datasets, including ETTh, Electricity, and Solar. The results demonstrate that DDT consistently outperforms strong state-of-the-art baselines across all prediction horizons, establishing a new benchmark for the task.
Authors: Anthony M. Polloreno
Abstract: In noisy physical reservoirs, the classical information-processing capacity $C_{\mathrm{ip}}$ quantifies how well a linear readout can realize tasks measurable from the input history, yet $C_{\mathrm{ip}}$ can be far smaller than the observed rank of the readout covariance. We explain this ``missing capacity'' by introducing the innovation capacity $C_{\mathrm{i}}$, the total capacity allocated to readout components orthogonal to the input filtration (Doob innovations, including input-noise mixing). Using a basis-free Hilbert-space formulation of the predictable/innovation decomposition, we prove the conservation law $C_{\mathrm{ip}}+C_{\mathrm{i}}=\mathrm{rank}(\Sigma_{XX})\le d$, so predictable and innovation capacities exactly partition the rank of the observable readout dimension covariance $\Sigma_{XX}\in \mathbb{R}^{\rm d\times d}$. In linear-Gaussian Johnson-Nyquist regimes, $\Sigma_{XX}(T)=S+T N_0$, the split becomes a generalized-eigenvalue shrinkage rule and gives an explicit monotone tradeoff between temperature and predictable capacity. Geometrically, in whitened coordinates the predictable and innovation components correspond to complementary covariance ellipsoids, making $C_{\mathrm{i}}$ a trace-controlled innovation budget. A large $C_{\mathrm{i}}$ forces a high-dimensional innovation subspace with a variance floor and under mild mixing and anti-concentration assumptions this yields extensive innovation-block differential entropy and exponentially many distinguishable histories. Finally, we give an information-theoretic lower bound showing that learning the induced innovation-block law in total variation requires a number of samples that scales with the effective innovation dimension, supporting the generative utility of noisy physical reservoirs.
Authors: Sk Md Ahnaf Akif Alvi, Raymundo Arr\'oyave, Douglas Allaire
Abstract: Bayesian Optimization with multi-objective acquisition functions such as q-Expected Hypervolume Improvement (qEHVI) requires efficient candidate optimization to maximize acquisition function values. Traditional approaches rely on continuous optimization methods like Sequential Least Squares Programming (SLSQP) for candidate selection. However, these gradient-based methods can become trapped in local optima, particularly in complex or high-dimensional objective landscapes. This paper presents a simulated annealing-based approach for candidate optimization in batch acquisition functions as an alternative to conventional continuous optimization methods. We evaluate our simulated annealing approach against SLSQP across four benchmark multi-objective optimization problems: ZDT1 (30D, 2 objectives), DTLZ2 (7D, 3 objectives), Kursawe (3D, 2 objectives), and Latent-Aware (4D, 2 objectives). Our results demonstrate that simulated annealing consistently achieves superior hypervolume performance compared to SLSQP in most test functions. The improvement is particularly pronounced for DTLZ2 and Latent-Aware problems, where simulated annealing reaches significantly higher hypervolume values and maintains better convergence characteristics. The histogram analysis of objective space coverage further reveals that simulated annealing explores more diverse and optimal regions of the Pareto front. These findings suggest that metaheuristic optimization approaches like simulated annealing can provide more robust and effective candidate optimization for multi-objective Bayesian optimization, offering a promising alternative to traditional gradient-based methods for batch acquisition function optimization.
Authors: Haomin Wu, Zhiwei Nie, Hongyu Zhang, Zhixiang Ren
Abstract: Accurate prediction of enzyme kinetic parameters is essential for understanding catalytic mechanisms and guiding enzyme engineering.However, existing deep learning-based enzyme-substrate interaction (ESI) predictors often exhibit performance degradation on sequence-divergent, out-of-distribution (OOD) cases, limiting robustness under biologically relevant perturbations.We propose O$^2$DENet, a lightweight, plug-and-play module that enhances OOD generalization via biologically and chemically informed perturbation augmentation and invariant representation learning.O$^2$DENet introduces enzyme-substrate perturbations and enforces consistency between original and augmented enzyme-substrate-pair representations to encourage invariance to distributional shifts.When integrated with representative ESI models, O$^2$DENet consistently improves predictive performance for both $k_{cat}$ and $K_m$ across stringent sequence-identity-based OOD benchmarks, achieving state-of-the-art results among the evaluated methods in terms of accuracy and robustness metrics.Overall, O$^2$DENet provides a general and effective strategy to enhance the stability and deployability of data-driven enzyme kinetics predictors for real-world enzyme engineering applications.
Authors: Yalan Tan, Yanyong Huang, Zongxin Shen, Dongjie Wang, Fengmao Lv, Tianrui Li
Abstract: Although multi-view unsupervised feature selection (MUFS) has demonstrated success in dimensionality reduction for unlabeled multi-view data, most existing methods reduce feature redundancy by focusing on linear correlations among features but often overlook complex nonlinear dependencies. This limits the effectiveness of feature selection. In addition, existing methods fuse similarity graphs from multiple views by employing sample-invariant weights to preserve local structure. However, this process fails to account for differences in local neighborhood clarity among samples within each view, thereby hindering accurate characterization of the intrinsic local structure of the data. In this paper, we propose a Kernel Alignment-based multi-view unsupervised FeatUre selection with Sample-level adaptive graph lEarning method (KAFUSE) to address these issues. Specifically, we first employ kernel alignment with an orthogonal constraint to reduce feature redundancy in both linear and nonlinear relationships. Then, a cross-view consistent similarity graph is learned by applying sample-level fusion to each slice of a tensor formed by stacking similarity graphs from different views, which automatically adjusts the view weights for each sample during fusion. These two steps are integrated into a unified model for feature selection, enabling mutual enhancement between them. Extensive experiments on real multi-view datasets demonstrate the superiority of KAFUSE over state-of-the-art methods.
Authors: Silvia Ruiz-Espa\~na (ITI, Universitat Polit\`ecnica de Val\`encia, Val\`encia, Spain), Laura Arnal (ITI, Universitat Polit\`ecnica de Val\`encia, Val\`encia, Spain), Fran\c{c}ois Signol (ITI, Universitat Polit\`ecnica de Val\`encia, Val\`encia, Spain), Juan-Carlos Perez-Cortes (ITI, Universitat Polit\`ecnica de Val\`encia, Val\`encia, Spain), Joaquim Arlandis (ITI, Universitat Polit\`ecnica de Val\`encia, Val\`encia, Spain)
Abstract: The rapid adoption of complex Artificial Intelligence (AI) and Machine Learning (ML) models has led to their characterization as black boxes due to the difficulty of explaining their internal decision-making processes. This lack of transparency hinders users' ability to understand, validate and trust model behavior, particularly in high-risk applications. Although explainable AI (XAI) has made significant progress, there remains a need for versatile and effective techniques to address increasingly complex models. This work introduces Multivariate Conditional Expectation (MUCE), a model-agnostic method for local explainability designed to capture prediction changes from feature interactions. MUCE extends Individual Conditional Expectation (ICE) by exploring a multivariate grid of values in the neighborhood of a given observation at inference time, providing graphical explanations that illustrate the local evolution of model predictions. In addition, two quantitative indices, stability and uncertainty, summarize local behavior and assess model reliability. Uncertainty is further decomposed into uncertainty+ and uncertainty- to capture asymmetric effects that global measures may overlook. The proposed method is validated using XGBoost models trained on three datasets: two synthetic (2D and 3D) to evaluate behavior near decision boundaries, and one transformed real-world dataset to test adaptability to heterogeneous feature types. Results show that MUCE effectively captures complex local model behavior, while the stability and uncertainty indices provide meaningful insight into prediction confidence. MUCE, together with the ICE modification and the proposed indices, offers a practical contribution to local explainability, enabling both graphical and quantitative insights that enhance the interpretability of predictive models and support more trustworthy and transparent decision-making.
Authors: Runze Ma, Caizhi Liao
Abstract: Although deep learning has advanced automated electrocardiogram (ECG) diagnosis, prevalent supervised methods typically treat recordings as undifferentiated one-dimensional (1D) signals or two-dimensional (2D) images. This formulation compels models to learn physiological structures implicitly, resulting in data inefficiency and opacity that diverge from medical reasoning. To address these limitations, we propose BEAT-Net, a Biomimetic ECG Analysis with Tokenization framework that reformulates the problem as a language modeling task. Utilizing a QRS tokenization strategy to transform continuous signals into biologically aligned heartbeat sequences, the architecture explicitly decomposes cardiac physiology through specialized encoders that extract local beat morphology while normalizing spatial lead perspectives and modeling temporal rhythm dependencies. Evaluations across three large-scale benchmarks demonstrate that BEAT-Net matches the diagnostic accuracy of dominant convolutional neural network (CNN) architectures while substantially improving robustness. The framework exhibits exceptional data efficiency, recovering fully supervised performance using only 30 to 35 percent of annotated data. Moreover, learned attention mechanisms provide inherent interpretability by spontaneously reproducing clinical heuristics, such as Lead II prioritization for rhythm analysis, without explicit supervision. These findings indicate that integrating biological priors offers a computationally efficient and interpretable alternative to data-intensive large-scale pre-training.
Authors: Xue Gong, Qi Yi, Ziyuan Nan, Guanhua Huang, Kejiao Li, Yuhao Jiang, Ruibin Xiong, Zenan Xu, Jiaming Guo, Shaohui Peng, Bo Zhou
Abstract: Training Large Language Models (LLMs) for reasoning tasks is increasingly driven by Reinforcement Learning with Verifiable Rewards (RLVR), where Proximal Policy Optimization (PPO) provides a principled framework for stable policy updates. However, the practical application of PPO is hindered by unreliable advantage estimation in the sparse-reward RLVR regime. This issue arises because the sparse rewards in RLVR lead to inaccurate intermediate value predictions, which in turn introduce significant bias when aggregated at every token by Generalized Advantage Estimation (GAE). To address this, we introduce Segmental Advantage Estimation (SAE), which mitigates the bias that GAE can incur in RLVR. Our key insight is that aggregating $n$-step advantages at every token(as in GAE) is unnecessary and often introduces excessive bias, since individual tokens carry minimal information. Instead, SAE first partitions the generated sequence into coherent sub-segments using low-probability tokens as heuristic boundaries. It then selectively computes variance-reduced advantage estimates only from these information-rich segment transitions, effectively filtering out noise from intermediate tokens. Our experiments demonstrate that SAE achieves superior performance, with marked improvements in final scores, training stability, and sample efficiency. These gains are shown to be consistent across multiple model sizes, and a correlation analysis confirms that our proposed advantage estimator achieves a higher correlation with an approximate ground-truth advantage, justifying its superior performance.
Authors: Hamda Hmida, Hsiu-Wen Chang Joly, Youssef Mesri
Abstract: Partial differential equations (PDEs) govern a wide range of physical phenomena, but their numerical solution remains computationally demanding, especially when repeated simulations are required across many parameter settings. Recent Scientific Foundation Models (SFMs) aim to alleviate this cost by learning universal surrogates from large collections of simulated systems, yet they typically rely on monolithic architectures with limited interpretability and high pretraining expense. In this work we introduce Compositional Neural Operators (CompNO), a compositional neural operator framework for parametric PDEs. Instead of pretraining a single large model on heterogeneous data, CompNO first learns a library of Foundation Blocks, where each block is a parametric Fourier neural operator specialized to a fundamental differential operator (e.g. convection, diffusion, nonlinear convection). These blocks are then assembled, via lightweight Adaptation Blocks, into task-specific solvers that approximate the temporal evolution operator for target PDEs. A dedicated boundary-condition operator further enforces Dirichlet constraints exactly at inference time. We validate CompNO on one-dimensional convection, diffusion, convection--diffusion and Burgers' equations from the PDEBench suite. The proposed framework achieves lower relative L2 error than strong baselines (PFNO, PDEFormer and in-context learning based models) on linear parametric systems, while remaining competitive on nonlinear Burgers' flows. The model maintains exact boundary satisfaction with zero loss at domain boundaries, and exhibits robust generalization across a broad range of Peclet and Reynolds numbers. These results demonstrate that compositional neural operators provide a scalable and physically interpretable pathway towards foundation models for PDEs.
Authors: Petr Zelina, Marko \v{R}eh\'a\v{c}ek, Jana Hal\'amkov\'a, Lucia Bohovicov\'a, Martin Rusinko, V\'it Nov\'a\v{c}ek
Abstract: Clinical notes hold rich yet unstructured details about diagnoses, treatments, and outcomes that are vital to precision medicine but hard to exploit at scale. We introduce a method that represents each patient as a matrix built from aggregated embeddings of all their notes, enabling robust patient similarity computation based on their latent low-rank representations. Using clinical notes of 4,267 Czech breast-cancer patients and expert similarity labels from Masaryk Memorial Cancer Institute, we evaluate several matrix-based similarity measures and analyze their strengths and limitations across different similarity facets, such as clinical history, treatment, and adverse events. The results demonstrate the usefulness of the presented method for downstream tasks, such as personalized therapy recommendations or toxicity warnings.
Authors: Xueyan Niu, Bo Bai, Wei Han, Weixi Zhang
Abstract: Post-training of large language models routinely interleaves supervised fine-tuning (SFT) with reinforcement learning (RL). These two methods have different objectives: SFT minimizes the cross-entropy loss between model outputs and expert responses, while RL maximizes reward signals derived from human preferences or rule-based verifiers. Modern reasoning models have widely adopted the practice of alternating SFT and RL training. However, there is no theoretical account of whether they can be decoupled. We prove that decoupling is impossible in either order: (1) SFT-then-RL coupling: RL increases SFT loss under SFT optimality and (2) RL-then-SFT coupling: SFT lowers the reward achieved by RL. Experiments on Qwen3-0.6B confirm the predicted degradation, verifying that SFT and RL cannot be separated without loss of prior performance in the post-training
Authors: Alexandre Tuel, Thomas Kerdreux, Quentin Febvre, Alexis Mouche, Antoine Grouazel, Jean-Renaud Miadana, Antoine Audras, Chen Wang, Bertrand Chapron
Abstract: We present OceanSAR-2, the second generation of our foundation model for SAR-based ocean observation. Building on our earlier release, which pioneered self-supervised learning on Sentinel-1 Wave Mode data, OceanSAR-2 relies on improved SSL training and dynamic data curation strategies, which enhances performance while reducing training cost. OceanSAR-2 demonstrates strong transfer performance across downstream tasks, including geophysical pattern classification, ocean surface wind vector and significant wave height estimation, and iceberg detection. We release standardized benchmark datasets, providing a foundation for systematic evaluation and advancement of SAR models for ocean applications.
Authors: Zihao Fu, Xufeng Duan, Zhenguang G. Cai
Abstract: Large language models excel across diverse domains, yet their deployment in healthcare, legal systems, and autonomous decision-making remains limited by incomplete understanding of their internal mechanisms. As these models integrate into high-stakes systems, understanding how they encode capabilities has become fundamental to interpretability research. Traditional approaches identify important modules through gradient attribution or activation analysis, assuming specific capabilities map to specific components. However, this oversimplifies neural computation: modules may contribute to multiple capabilities simultaneously, while single capabilities may distribute across multiple modules. These coarse-grained analyses fail to capture fine-grained, distributed capability encoding. We present SCALPEL (Selective Capability Ablation via Low-rank Parameter Editing for Large language models), a framework representing capabilities as low-rank parameter subspaces rather than discrete modules. Our key insight is that capabilities can be characterized by low-rank modifications distributed across layers and modules, enabling precise capability removal without affecting others. By training LoRA adapters to reduce distinguishing correct from incorrect answers while preserving general language modeling quality, SCALPEL identifies low-rank representations responsible for particular capabilities while remaining disentangled from others. Experiments across diverse capability and linguistic tasks from BLiMP demonstrate that SCALPEL successfully removes target capabilities while preserving general capabilities, providing fine-grained insights into capability distribution across parameter space. Results reveal that capabilities exhibit low-rank structure and can be selectively ablated through targeted parameter-space interventions, offering nuanced understanding of capability encoding in LLMs.
Authors: Junyao Zhang, Jinglai Li, Junqi Tang
Abstract: Agent-Based Models (ABMs) are gaining great popularity in economics and social science because of their strong flexibility to describe the realistic and heterogeneous decisions and interaction rules between individual agents. In this work, we investigate for the first time the practicality of test-time training (TTT) of deep models such as normalizing flows, in the parameters posterior estimations of ABMs. We propose several practical TTT strategies for fine-tuning the normalizing flow against distribution shifts. Our numerical study demonstrates that TTT schemes are remarkably effective, enabling real-time adjustment of flow-based inference for ABM parameters.
Authors: Haotian Gao, Xiangying Zhang, Jingyuan Li, Xinchong Chen, Haojie Wang, Yifei Qi, Renxiao Wang
Abstract: Drug discovery represents a time-consuming and financially intensive process, and virtual screening can accelerate it. Scoring functions, as one of the tools guiding virtual screening, have their precision closely tied to screening efficiency. In our previous study, we developed a graph neural network model called PLANET (Protein-Ligand Affinity prediction NETwork), but it suffers from the defect in representing protein-ligand contact maps. Incorrect binding modes inevitably lead to poor affinity predictions, so accurate prediction of the protein-ligand contact map is desired to improve PLANET. In this study, we have proposed PLANET v2.0 as an upgraded version. The model is trained via multi-objective training strategy and incorporates the Mixture Density Network to predict binding modes. Except for the probability density distributions of non-covalent interactions, we innovatively employ another Gaussian mixture model to describe the relationship between distance and energy of each interaction pair and predict protein-ligand affinity like calculating the mathematical expectation. As on the CASF-2016 benchmark, PLANET v2.0 demonstrates excellent scoring power, ranking power, and docking power. The screening power of PLANET v2.0 gets notably improved compared to PLANET and Glide SP and it demonstrates robust validation on a commercial ultra-large-scale dataset. Given its efficiency and accuracy, PLANET v2.0 can hopefully become one of the practical tools for virtual screening workflows. PLANET v2.0 is freely available at https://www.pdbbind-plus.org.cn/planetv2.
Authors: Fiona Redmen, Ethan Tregidga, James F. Steiner, Cecilia Garraffo
Abstract: Black hole X-ray binaries (BHBs) can be studied with spectral fitting to provide physical constraints on accretion in extreme gravitational environments. Traditional methods of spectral fitting such as Markov Chain Monte Carlo (MCMC) face limitations due to computational times. We introduce a probabilistic model, utilizing a variational autoencoder with a normalizing flow, trained to adopt a physical latent space. This neural network produces predictions for spectral-model parameters as well as their full probability distributions. Our implementations result in a significant improvement in spectral reconstructions over a previous deterministic model while performing three orders of magnitude faster than traditional methods.
Authors: Maaz Ahmad, Iftekhar A. Karimi
Abstract: Global optimization of large-scale, complex systems such as multi-physics black-box simulations and real-world industrial systems is important but challenging. This work presents a novel Surrogate-Based Optimization framework based on Clustering, SBOC for global optimization of such systems, which can be used with any surrogate modeling technique. At each iteration, it uses a single surrogate model for the entire domain, employs k-means clustering to identify unexplored domain, and exploits a local region around the surrogate optimum to potentially add three new sample points in the domain. SBOC has been tested against sixteen promising benchmarking algorithms using 52 analytical test functions of varying input dimensionalities and shape profiles. It successfully identified a global minimum for most test functions with substantially lower computational effort than other algorithms. It worked especially well on test functions with four or more input variables. It was also among the top six algorithms in approaching a global minimum closely. Overall, SBOC is a robust, reliable, and efficient algorithm for global optimization of box-constrained systems.
Authors: Michael J. Clark
Abstract: As models grow more capable, human supervision breaks down: labels don't scale, outputs can be gamed, and training doesn't generalize. Scalable oversight requires steering methods that are internal, self-supervised, and transfer out-of-distribution; existing methods satisfy some but not all three. We introduce AntiPaSTO, which separates representations along an anti-parallel axis ($\alpha=\pm1$ produce opposite shifts), with coherence constraints preventing collapse. Human input is minimal: two contrasting words inserted into template sentences, no preference labels. Using 800 such pairs on Gemma-3-1B, AntiPaSTO beats prompting baselines by $6.9\times$ on DailyDilemmas and maintains bidirectional control where prompting triggers refusal. Code is available at https://github.com/wassname/AntiPaSTO.
Authors: Youngmin Oh, Hyung-Il Kim, Jung Uk Kim
Abstract: Multi-task learning (MTL) is critical in real-world applications such as autonomous driving and robotics, enabling simultaneous handling of diverse tasks. However, obtaining fully annotated data for all tasks is impractical due to labeling costs. Existing methods for partially labeled MTL typically rely on predictions from unlabeled tasks, making it difficult to establish reliable task associations and potentially leading to negative transfer and suboptimal performance. To address these issues, we propose a prototype-based knowledge retrieval framework that achieves robust MTL instead of relying on predictions from unlabeled tasks. Our framework consists of two key components: (1) a task prototype embedding task-specific characteristics and quantifying task associations, and (2) a knowledge retrieval transformer that adaptively refines feature representations based on these associations. To achieve this, we introduce an association knowledge generating (AKG) loss to ensure the task prototype consistently captures task-specific characteristics. Extensive experiments demonstrate the effectiveness of our framework, highlighting its potential for robust multi-task learning, even when only a subset of tasks is annotated.
Authors: Haoqian Meng, Yilun Luo, Yafei Zhao, Wenyuan Liu, Peng Zhang, Xindian Ma
Abstract: The emergence of fine-grained numerical formats like NVFP4 presents new opportunities for efficient Large Language Model (LLM) inference. However, it is difficult to adapt existing Post-Training Quantization (PTQ) strategies to these formats: rotation-based methods compromise fine-grained block isolation; smoothing techniques struggle with significant 4-bit quantization errors; and mixed-precision approaches often conflict with hardware constraints on unified-precision computation. To address these challenges, we propose ARCQuant, a framework that boosts NVFP4 performance via Augmented Residual Channels. Distinct from methods that compromise block isolation or hardware uniformity, ARCQuant maintains a strictly unified NVFP4 format by augmenting the activation matrix with quantized residual channels. This design integrates the error compensation process directly into the matrix reduction dimension, enabling the use of standard, highly optimized GEMM kernels with minimal overhead. Theoretical analysis confirms that the worst-case error bound of our dual-stage NVFP4 quantization is comparable to that of standard 8-bit formats such as MXFP8. Extensive experiments on LLaMA and Qwen models demonstrate that ARCQuant achieves state-of-the-art accuracy, comparable to full-precision baselines in perplexity and downstream tasks. Furthermore, deployment on RTX 5090 and RTX PRO 6000 GPUs confirms practical benefits, achieving up to 3x speedup over FP16. Our code is available at https://github.com/actypedef/ARCQuant .
Authors: Xiaoxiao Deng
Abstract: Automated ICD coding involves assigning standardized diagnostic codes to clinical narratives. The vast label space and extreme class imbalance continue to challenge precise prediction. To address these issues, LabGraph is introduced -- a unified framework that reformulates ICD coding as a graph generation task. By combining adversarial domain adaptation, graph-based reinforcement learning, and perturbation regularization, LabGraph effectively enhances model robustness and generalization. In addition, a label graph discriminator dynamically evaluates each generated code, providing adaptive reward feedback during training. Experiments on benchmark datasets demonstrate that LabGraph consistently outperforms previous approaches on micro-F1, micro-AUC, and P@K.
Authors: Tzu-Hsuan Lin, Chih-Hsuan Kao
Abstract: The rapid advancement of Large Language Models (LLMs) and their integration into autonomous agent systems has created unprecedented opportunities for document analysis, decision support, and knowledge retrieval. However, the complexity of developing, evaluating, and iterating on LLM-based agent workflows presents significant barriers to researchers, particularly those without extensive software engineering expertise. We present FROAV (Framework for RAG Observation and Agent Verification), an open-source research platform that democratizes LLM agent research by providing a plug-and-play architecture combining visual workflow orchestration, a comprehensive evaluation framework, and extensible Python integration. FROAV implements a multi-stage Retrieval-Augmented Generation (RAG) pipeline coupled with a rigorous "LLM-as-a-Judge" evaluation system, all accessible through intuitive graphical interfaces. Our framework integrates n8n for no-code workflow design, PostgreSQL for granular data management, FastAPI for flexible backend logic, and Streamlit for human-in-the-loop interaction. Through this integrated ecosystem, researchers can rapidly prototype RAG strategies, conduct prompt engineering experiments, validate agent performance against human judgments, and collect structured feedback-all without writing infrastructure code. We demonstrate the framework's utility through its application to financial document analysis, while emphasizing its material-agnostic architecture that adapts to any domain requiring semantic analysis. FROAV represents a significant step toward making LLM agent research accessible to a broader scientific community, enabling researchers to focus on hypothesis testing and algorithmic innovation rather than system integration challenges.
Authors: Jingwen Fu, Ming Xiao, Mikael Skoglund, Dong In Kim
Abstract: Due to strict rate and reliability demands, wireless image transmission remains difficult for both classical layered designs and joint source-channel coding (JSCC), especially under low latency. Diffusion-based generative decoders can deliver strong perceptual quality by leveraging learned image priors, but iterative stochastic denoising leads to high decoding delay. To enable low-latency decoding, we propose a flow-matching (FM) generative decoder under a new land-then-transport (LTT) paradigm that tightly integrates the physical wireless channel into a continuous-time probability flow. For AWGN channels, we build a Gaussian smoothing path whose noise schedule indexes effective noise levels, and derive a closed-form teacher velocity field along this path. A neural-network student vector field is trained by conditional flow matching, yielding a deterministic, channel-aware ODE decoder with complexity linear in the number of ODE steps. At inference, it only needs an estimate of the effective noise variance to set the ODE starting time. We further show that Rayleigh fading and MIMO channels can be mapped, via linear MMSE equalization and singular-value-domain processing, to AWGN-equivalent channels with calibrated starting times. Therefore, the same probability path and trained velocity field can be reused for Rayleigh and MIMO without retraining. Experiments on MNIST, Fashion-MNIST, and DIV2K over AWGN, Rayleigh, and MIMO demonstrate consistent gains over JPEG2000+LDPC, DeepJSCC, and diffusion-based baselines, while achieving good perceptual quality with only a few ODE steps. Overall, LTT provides a deterministic, physically interpretable, and computation-efficient framework for generative wireless image decoding across diverse channels.
Authors: Chris Elliott, Einar Urdshals, David Quarel, Matthew Farrugia-Roberts, Daniel Murfet
Abstract: Singular learning theory characterizes Bayesian learning as an evolving tradeoff between accuracy and complexity, with transitions between qualitatively different solutions as sample size increases. We extend this theory to deep reinforcement learning, proving that the concentration of the generalized posterior over policies is governed by the local learning coefficient (LLC), an invariant of the geometry of the regret function. This theory predicts that Bayesian phase transitions in reinforcement learning should proceed from simple policies with high regret to complex policies with low regret. We verify this prediction empirically in a gridworld environment exhibiting stagewise policy development: phase transitions over SGD training manifest as "opposing staircases" where regret decreases sharply while the LLC increases. Notably, the LLC detects phase transitions even when estimated on a subset of states where the policies appear identical in terms of regret, suggesting it captures changes in the underlying algorithm rather than just performance.
Authors: Omri Lev, Moshe Shenfeld, Vishwak Srinivasan, Katrina Ligett, Ashia C. Wilson
Abstract: We study differentially private ordinary least squares (DP-OLS) with bounded data. The dominant approach, adaptive sufficient-statistics perturbation (AdaSSP), adds an adaptively chosen perturbation to the sufficient statistics, namely, the matrix $X^{\top}X$ and the vector $X^{\top}Y$, and is known to achieve near-optimal accuracy and to have strong empirical performance. In contrast, methods that rely on Gaussian-sketching, which ensure differential privacy by pre-multiplying the data with a random Gaussian matrix, are widely used in federated and distributed regression, yet remain relatively uncommon for DP-OLS. In this work, we introduce the iterative Hessian mixing, a novel DP-OLS algorithm that relies on Gaussian sketches and is inspired by the iterative Hessian sketch algorithm. We provide utility analysis for the iterative Hessian mixing as well as a new analysis for the previous methods that rely on Gaussian sketches. Then, we show that our new approach circumvents the intrinsic limitations of the prior methods and provides non-trivial improvements over AdaSSP. We conclude by running an extensive set of experiments across standard benchmarks to demonstrate further that our approach consistently outperforms these prior baselines.
Authors: Kaito Tanaka, Aya Nakayama, Masato Ito, Yuji Nishimura, Keisuke Matsuda
Abstract: Medical time series data, such as EEG and ECG, are vital for diagnosing neurological and cardiovascular diseases. However, their precise interpretation faces significant challenges due to high annotation costs, leading to data scarcity, and the limitations of traditional contrastive learning in capturing complex temporal patterns. To address these issues, we propose CoDAC (Contextual Discrepancy-Aware Contrastive learning), a novel framework that enhances diagnostic accuracy and generalization, particularly in small-sample settings. CoDAC leverages external healthy data and introduces a Contextual Discrepancy Estimator (CDE), built upon a Transformer-based Autoencoder, to precisely quantify abnormal signals through context-aware anomaly scores. These scores dynamically inform a Dynamic Multi-views Contrastive Framework (DMCF), which adaptively weights different temporal views to focus contrastive learning on diagnostically relevant, discrepant regions. Our encoder combines dilated convolutions with multi-head attention for robust feature extraction. Comprehensive experiments on Alzheimer's Disease EEG, Parkinson's Disease EEG, and Myocardial Infarction ECG datasets demonstrate CoDAC's superior performance across all metrics, consistently outperforming state-of-the-art baselines, especially under low label availability. Ablation studies further validate the critical contributions of CDE and DMCF. CoDAC offers a robust and interpretable solution for medical time series diagnosis, effectively mitigating data scarcity challenges.
Authors: Zexi Tan, Tao Xie, Haoyi Xiao, Baoyao Yang, Yuzhu Ji, An Zeng, Xiang Zhang, Yiqun Zhang
Abstract: Multivariate Time-Series (MTS) clustering is crucial for signal processing and data analysis. Although deep learning approaches, particularly those leveraging Contrastive Learning (CL), are prominent for MTS representation, existing CL-based models face two key limitations: 1) neglecting clustering information during positive/negative sample pair construction, and 2) introducing unreasonable inductive biases, e.g., destroying time dependence and periodicity through augmentation strategies, compromising representation quality. This paper, therefore, proposes a Temporal-Frequency Enhanced Contrastive (TFEC) learning framework. To preserve temporal structure while generating low-distortion representations, a temporal-frequency Co-EnHancement (CoEH) mechanism is introduced. Accordingly, a synergistic dual-path representation and cluster distribution learning framework is designed to jointly optimize cluster structure and representation fidelity. Experiments on six real-world benchmark datasets demonstrate TFEC's superiority, achieving 4.48% average NMI gains over SOTA methods, with ablation studies validating the design. The code of the paper is available at: https://github.com/yueliangy/TFEC.
Authors: Yu-Yang Qian, Junda Su, Lanxiang Hu, Peiyuan Zhang, Zhijie Deng, Peng Zhao, Hao Zhang
Abstract: Diffusion large language models (dLLMs) offer capabilities beyond those of autoregressive (AR) LLMs, such as parallel decoding and random-order generation. However, realizing these benefits in practice is non-trivial, as dLLMs inherently face an accuracy-parallelism trade-off. Despite increasing interest, existing methods typically focus on only one-side of the coin, targeting either efficiency or performance. To address this limitation, we propose d3LLM (Pseudo-Distilled Diffusion Large Language Model), striking a balance between accuracy and parallelism: (i) during training, we introduce pseudo-trajectory distillation to teach the model which tokens can be decoded confidently at early steps, thereby improving parallelism; (ii) during inference, we employ entropy-based multi-block decoding with a KV-cache refresh mechanism to achieve high parallelism while maintaining accuracy. To better evaluate dLLMs, we also introduce AUP (Accuracy Under Parallelism), a new metric that jointly measures accuracy and parallelism. Experiments demonstrate that our d3LLM achieves up to 10$\times$ speedup over vanilla LLaDA/Dream and 5$\times$ speedup over AR models without much accuracy drop. Our code is available at https://github.com/hao-ai-lab/d3LLM.
Authors: Yulu Wang, Ziqian Zeng, Jianjun Wu, Zhifeng Tang
Abstract: In an ideal medical environment, real-time coagulation monitoring can enable early detection and prompt remediation of risks. However, traditional Thromboelastography (TEG), a widely employed diagnostic modality, can only provide such outputs after nearly 1 hour of measurement. The delay might lead to elevated mortality rates. These issues clearly point out one of the key challenges for medical AI development: Mak-ing reasonable predictions based on very small data sets and accounting for variation between different patient populations, a task where conventional deep learning methods typically perform poorly. We present Physiological State Reconstruc-tion (PSR), a new algorithm specifically designed to take ad-vantage of dynamic changes between individuals and to max-imize useful information produced by small amounts of clini-cal data through mapping to reliable predictions and diagnosis. We develop MDFE to facilitate integration of varied temporal signals using multi-domain learning, and jointly learn high-level temporal interactions together with attentions via HLA; furthermore, the parameterized DAM we designed maintains the stability of the computed vital signs. PSR evaluates with 4 TEG-specialized data sets and establishes remarkable perfor-mance -- predictions of R2 > 0.98 for coagulation traits and error reduction around half compared to the state-of-the-art methods, and halving the inferencing time too. Drift-aware learning suggests a new future, with potential uses well be-yond thrombophilia discovery towards medical AI applica-tions with data scarcity.
Authors: Yanan Chen, Tieliang Gong, Yunjiao Zhang, Wen Wen
Abstract: Continual Learning (CL) aims to enable models to sequentially learn multiple tasks without forgetting previous knowledge. Recent studies have shown that optimizing towards flatter loss minima can improve model generalization. However, existing sharpness-aware methods for CL suffer from two key limitations: (1) they treat sharpness regularization as a unified signal without distinguishing the contributions of its components. and (2) they introduce substantial computational overhead that impedes practical deployment. To address these challenges, we propose FLAD, a novel optimization framework that decomposes sharpness-aware perturbations into gradient-aligned and stochastic-noise components, and show that retaining only the noise component promotes generalization. We further introduce a lightweight scheduling scheme that enables FLAD to maintain significant performance gains even under constrained training time. FLAD can be seamlessly integrated into various CL paradigms and consistently outperforms standard and sharpness-aware optimizers in diverse experimental settings, demonstrating its effectiveness and practicality in CL.
Authors: Kishan Padayachy, Ronald Richman, Mario V. W\"uthrich
Abstract: We introduce Tab-TRM (Tabular-Tiny Recursive Model), a network architecture that adapts the recursive latent reasoning paradigm of Tiny Recursive Models (TRMs) to insurance modeling. Drawing inspiration from both the Hierarchical Reasoning Model (HRM) and its simplified successor TRM, the Tab-TRM model makes predictions by reasoning over the input features. It maintains two learnable latent tokens - an answer token and a reasoning state - that are iteratively refined by a compact, parameter-efficient recursive network. The recursive processing layer repeatedly updates the reasoning state given the full token sequence and then refines the answer token, in close analogy with iterative insurance pricing schemes. Conceptually, Tab-TRM bridges classical actuarial workflows - iterative generalized linear model fitting and minimum-bias calibration - on the one hand, and modern machine learning, in terms of Gradient Boosting Machines, on the other.
Authors: Robert Lewis, Katie Matton, Rosalind W. Picard, John Guttag
Abstract: Self-supervised pre-training with contrastive learning is a powerful method for learning from sparsely labeled data. However, performance can drop considerably when there is a shift in the distribution of data from training to test time. We study this phenomenon in a setting in which the training data come from multiple domains, and the test data come from a domain not seen at training that is subject to significant covariate shift. We present a new method for contrastive learning that incorporates domain labels to increase the domain invariance of learned representations, leading to improved out-of-distribution generalization. Our method adjusts the temperature parameter in the InfoNCE loss -- which controls the relative weighting of negative pairs -- using the probability that a negative sample comes from the same domain as the anchor. This upweights pairs from more similar domains, encouraging the model to discriminate samples based on domain-invariant attributes. Through experiments on a variant of the MNIST dataset, we demonstrate that our method yields better out-of-distribution performance than domain generalization baselines. Furthermore, our method maintains strong in-distribution task performance, substantially outperforming baselines on this measure.
Authors: Shao-Ting Chiu, Siu Wun Cheung, Ulisses Braga-Neto, Chak Shing Lee, Rui Peng Li
Abstract: Kolmogorov-Arnold Networks (KANs) have shown strong potential for efficiently approximating complex nonlinear functions. However, the original KAN formulation relies on B-spline basis functions, which incur substantial computational overhead due to De Boor's algorithm. To address this limitation, recent work has explored alternative basis functions such as radial basis functions (RBFs) that can improve computational efficiency and flexibility. Yet, standard RBF-KANs often sacrifice accuracy relative to the original KAN design. In this work, we propose Free-RBF-KAN, a RBF-based KAN architecture that incorporates adaptive learning grids and trainable smoothness to close this performance gap. Our method employs freely learnable RBF shapes that dynamically align grid representations with activation patterns, enabling expressive and adaptive function approximation. Additionally, we treat smoothness as a kernel parameter optimized jointly with network weights, without increasing computational complexity. We provide a general universality proof for RBF-KANs, which encompasses our Free-RBF-KAN formulation. Through a broad set of experiments, including multiscale function approximation, physics-informed machine learning, and PDE solution operator learning, Free-RBF-KAN achieves accuracy comparable to the original B-spline-based KAN while delivering faster training and inference. These results highlight Free-RBF-KAN as a compelling balance between computational efficiency and adaptive resolution, particularly for high-dimensional structured modeling tasks.
Authors: Jiawei Wang, Yanfei Zhou, Siddartha Devic, Deqing Fu
Abstract: Large Language Models (LLMs) can produce surprisingly sophisticated estimates of their own uncertainty. However, it remains unclear to what extent this expressed confidence is tied to the reasoning, knowledge, or decision making of the model. To test this, we introduce $\textbf{RiskEval}$: a framework designed to evaluate whether models adjust their abstention policies in response to varying error penalties. Our evaluation of several frontier models reveals a critical dissociation: models are neither cost-aware when articulating their verbal confidence, nor strategically responsive when deciding whether to engage or abstain under high-penalty conditions. Even when extreme penalties render frequent abstention the mathematically optimal strategy, models almost never abstain, resulting in utility collapse. This indicates that calibrated verbal confidence scores may not be sufficient to create trustworthy and interpretable AI systems, as current models lack the strategic agency to convert uncertainty signals into optimal and risk-sensitive decisions.
Authors: Wen Guo
Abstract: We introduce DT-ICU, a multimodal digital twin framework for continuous risk estimation in intensive care. DT-ICU integrates variable-length clinical time series with static patient information in a unified multitask architecture, enabling predictions to be updated as new observations accumulate over the ICU stay. We evaluate DT-ICU on the large, publicly available MIMIC-IV dataset, where it consistently outperforms established baseline models under different evaluation settings. Our test-length analysis shows that meaningful discrimination is achieved shortly after admission, while longer observation windows further improve the ranking of high-risk patients in highly imbalanced cohorts. To examine how the model leverages heterogeneous data sources, we perform systematic modality ablations, revealing that the model learnt a reasonable structured reliance on interventions, physiological response observations, and contextual information. These analyses provide interpretable insights into how multimodal signals are combined and how trade-offs between sensitivity and precision emerge. Together, these results demonstrate that DT-ICU delivers accurate, temporally robust, and interpretable predictions, supporting its potential as a practical digital twin framework for continuous patient monitoring in critical care. The source code and trained model weights for DT-ICU are publicly available at https://github.com/GUO-W/DT-ICU-release.
Authors: Valentina Njaradi, Rodrigo Carrasco-Davis, Peter E. Latham, Andrew Saxe
Abstract: Learning how to learn efficiently is a fundamental challenge for biological agents and a growing concern for artificial ones. To learn effectively, an agent must regulate its learning speed, balancing the benefits of rapid improvement against the costs of effort, instability, or resource use. We introduce a normative framework that formalizes this problem as an optimal control process in which the agent maximizes cumulative performance while incurring a cost of learning. From this objective, we derive a closed-form solution for the optimal learning rate, which has the form of a closed-loop controller that depends only on the agent's current and expected future performance. Under mild assumptions, this solution generalizes across tasks and architectures and reproduces numerically optimized schedules in simulations. In simple learning models, we can mathematically analyze how agent and task parameters shape learning-rate scheduling as an open-loop control solution. Because the optimal policy depends on expectations of future performance, the framework predicts how overconfidence or underconfidence influence engagement and persistence, linking the control of learning speed to theories of self-regulated learning. We further show how a simple episodic memory mechanism can approximate the required performance expectations by recalling similar past learning experiences, providing a biologically plausible route to near-optimal behaviour. Together, these results provide a normative and biologically plausible account of learning speed control, linking self-regulated learning, effort allocation, and episodic memory estimation within a unified and tractable mathematical framework.
Authors: Mohammadreza Behboodi, Eli Kinney-Lang, Ali Etemad, Adam Kirton, Hatem Abou-Zeid
Abstract: Foundation Models (FMs) have surged in popularity over the past five years, with applications spanning fields from computer vision to natural language processing. Brain-Computer Interfaces (BCIs) have also gained momentum due to their potential to support individuals with complex disabilities. Among BCI paradigms, code-modulated Visual Evoked Potentials (c-VEPs) remain relatively understudied, despite offering high information transfer rates and large selection target capacities. However, c-VEP systems require lengthy calibration sessions, limiting their practicality outside of laboratory settings. In this study, we use a FM for the first time to eliminate the need for lengthy calibration in c-VEP BCI systems. We evaluated two approaches: (1) a truly calibration-free approach requiring no subject-specific data, and (2) a limited calibration approach, where we assessed the benefit of incorporating incremental amounts of calibration data. In both cases, a classification head is trained on data from other subjects. For a new subject, no calibration data is required in the calibration-free setup, making the c-VEP system effectively plug-and-play. The proposed method was tested on two c-VEP datasets. For the calibration-free approach, the average accuracy on the first dataset (n = 17) was 68.8% +/- 17.6%, comparable to the full-calibration performance reported in the original study (66.2% +/- 13.8%), which required approximately 11 minutes of calibration. On the second dataset (n = 12), the calibration-free accuracy was 71.8% +/- 20.2%, versus 93.7% +/- 5.5% from the original study, which required around 3.5 minutes. A limited-calibration approach using only 20% of the subject's data (approximately 43 seconds) yielded 92% +/- 5.2% accuracy. These results indicate that our FM-based approach can effectively eliminate or significantly reduce the need for lengthy calibration in c-VEP BCIs.
Authors: Hanyu Li, Jiangshan Duo, Bofei Gao, Hailin Zhang, Sujian Li, Xiaotie Deng, Liang Zhao
Abstract: Chain-of-thought reasoning in large language models often creates an "overthinking trap," leading to excessive computational cost and latency for unreliable accuracy gains. Prior work has typically relied on global, static controls that risk penalizing necessary reasoning. We introduce a sample-level, soft reinforcement learning compression method that penalizes inefficiently long rollouts, but only on problems where the model has already mastered and already produced a more concise rollout. Our experiments show that this method reduces average response length by 20-40% with comparable or higher accuracy. Crucially, the compression exhibits strong cross-domain generalization; a model trained on math spontaneously shortens responses on unseen tasks like code, instruction following, and general knowledge QA, with stable or improved accuracy. We demonstrate a stable post-training curriculum (accuracy-compression-accuracy) that can ultimately produce models that are more accurate and reason more concisely, arguing that such compression method should be a standard phase in developing efficient reasoning models.
Authors: Aolin Xu
Abstract: This work aims to rigorously define the values of perception, prediction, communication, and common sense in decision making. The defined quantities are decision-theoretic, but have information-theoretic analogues, e.g., they share some simple but key mathematical properties with Shannon entropy and mutual information, and can reduce to these quantities in particular settings. One interesting observation is that, the value of perception without prediction can be negative, while the value of perception together with prediction and the value of prediction alone are always nonnegative. The defined quantities suggest answers to practical questions arising in the design of autonomous decision-making systems. Example questions include: Do we need to observe and predict the behavior of a particular agent? How important is it? What is the best order to observe and predict the agents? The defined quantities may also provide insights to cognitive science and neural science, toward the understanding of how natural decision makers make use of information gained from different sources and operations.
Authors: Bohan Liang, Zijian Chen, Qi Jia, Kaiwei Zhang, Kaiyuan Ji, Guangtao Zhai
Abstract: Stock prediction, a subject closely related to people's investment activities in fully dynamic and live environments, has been widely studied. Current large language models (LLMs) have shown remarkable potential in various domains, exhibiting expert-level performance through advanced reasoning and contextual understanding. In this paper, we introduce PriceSeer, a live, dynamic, and data-uncontaminated benchmark specifically designed for LLMs performing stock prediction tasks. Specifically, PriceSeer includes 110 U.S. stocks from 11 industrial sectors, with each containing 249 historical data points. Our benchmark implements both internal and external information expansion, where LLMs receive extra financial indicators, news, and fake news to perform stock price prediction. We evaluate six cutting-edge LLMs under different prediction horizons, demonstrating their potential in generating investment strategies after obtaining accurate price predictions for different sectors. Additionally, we provide analyses of LLMs' suboptimal performance in long-term predictions, including the vulnerability to fake news and specific industries. The code and evaluation data will be open-sourced at https://github.com/BobLiang2113/PriceSeer.
Authors: Truong Xuan Khanh, Truong Quynh Hoa
Abstract: Recent advances in artificial intelligence have produced systems capable of remarkable performance across a wide range of tasks. These gains, however, are increasingly accompanied by concerns regarding long-horizon developmental behavior, as many systems converge toward repetitive solution patterns rather than sustained growth. We argue that a central limitation of contemporary AI systems lies not in capability per se, but in the premature fixation of their performance frontier. To address this issue, we introduce the concept of a \emph{Dynamic Intelligence Ceiling} (DIC), defined as the highest level of effective intelligence attainable by a system at a given time under its current resources, internal intent, and structural configuration. To make this notion empirically tractable, we propose a trajectory-centric evaluation framework that measures intelligence as a moving frontier rather than a static snapshot. We operationalize DIC using two estimators: the \emph{Progressive Difficulty Ceiling} (PDC), which captures the maximal reliably solvable difficulty under constrained resources, and the \emph{Ceiling Drift Rate} (CDR), which quantifies the temporal evolution of this frontier. These estimators are instantiated through a procedurally generated benchmark that jointly evaluates long-horizon planning and structural creativity within a single controlled environment. Our results reveal a qualitative distinction between systems that deepen exploitation within a fixed solution manifold and those that sustain frontier expansion over time. Importantly, our framework does not posit unbounded intelligence, but reframes limits as dynamic and trajectory-dependent rather than static and prematurely fixed. \vspace{0.5em} \noindent\textbf{Keywords:} AI evaluation, planning and creativity, developmental intelligence, dynamic intelligence ceilings, complex adaptive systems
Authors: Ahmed H. Ismail, Anthony Kuang, Ayo Akinkugbe, Kevin Zhu, Sean O'Brien
Abstract: Large language models (LLMs) often encode cognitive behaviors unpredictably across prompts, layers, and contexts, making them difficult to diagnose and control. We present CBMAS, a diagnostic framework for continuous activation steering, which extends cognitive bias analysis from discrete before/after interventions to interpretable trajectories. By combining steering vector construction with dense {\alpha}-sweeps, logit lens-based bias curves, and layer-site sensitivity analysis, our approach can reveal tipping points where small intervention strengths flip model behavior and show how steering effects evolve across layer depth. We argue that these continuous diagnostics offer a bridge between high-level behavioral evaluation and low-level representational dynamics, contributing to the cognitive interpretability of LLMs. Lastly, we provide a CLI and datasets for various cognitive behaviors at the project repository, https://github.com/shimamooo/CBMAS.
Authors: Canming Xia, Peixi Peng, Guang Tan, Zhan Su, Haoran Xu, Zhenxian Liu, Luntong Li
Abstract: Visual reinforcement learning (RL) suffers from poor sample efficiency due to high-dimensional observations in complex tasks. While existing works have shown that vision-language models (VLMs) can assist RL, they often focus on knowledge distillation from the VLM to RL, overlooking the potential of RL-generated interaction data to enhance the VLM. To address this, we propose COVR, a collaborative optimization framework that enables the mutual enhancement of the VLM and RL policies. Specifically, COVR fine-tunes the VLM with RL-generated data to enhance the semantic reasoning ability consistent with the target task, and uses the enhanced VLM to further guide policy learning via action priors. To improve fine-tuning efficiency, we introduce two key modules: (1) an Exploration-Driven Dynamic Filter module that preserves valuable exploration samples using adaptive thresholds based on the degree of exploration, and (2) a Return-Aware Adaptive Loss Weight module that improves the stability of training by quantifying the inconsistency of sampling actions via return signals of RL. We further design a progressive fine-tuning strategy to reduce resource consumption. Extensive experiments show that COVR achieves strong performance across various challenging visual control tasks.
Authors: Anshul Kumar
Abstract: Tokens are the basic units of Large Language Models (LLMs). LLMs rely on tokenizers to segment text into these tokens, and tokenization is the primary determinant of computational and inference cost. Sanskrit, one of the oldest languages, is hypothesized to express more meaning per token due to its morphology and grammar rules; however, no prior work has quantified this. We use a dataset of 701 parallel verses of the Bhagavad Gita, which comprises three languages-Sanskrit, English, and Hindi along with transliteration of Sanskrit into English. We test tokenizers including SentencePiece (SPM), older GPT models, and the latest generation tokenizers from Gemini and GPT. We use metrics of token count, characters per token (token efficiency), and tokens per character (token cost). Results show a ~2x difference in token counts between Sanskrit and English/Hindi under the unbiased SPM baseline. English/Hindi translations of Sanskrit commentary resulted in an approximately 20x increase in token count. GPT o200k base (latest, used by GPT-4o) and Gemini (latest) reduce bias by a significant degree compared to GPT cl100k base (used until GPT-4), but still fail to fully capture Sanskrit's compactness. This matters because there might be a penalty bias for non-English users, which inflates the token count. This research provides a foundation for improving future tokenizer design and shows the potential of Sanskrit for highly compact encoding, saving on cost while speeding up training and inference. The code and dataset are available at https://github.com/anshulkr713/sanskrit-token-efficiency
URLs: https://github.com/anshulkr713/sanskrit-token-efficiency
Authors: Kaiyuan Deng, Bo Hui, Gen Li, Jie Ji, Minghai Qin, Geng Yuan, Xiaolong Ma
Abstract: The widespread adoption of text-to-image (T2I) diffusion models has raised concerns about their potential to generate copyrighted, inappropriate, or sensitive imagery learned from massive training corpora. As a practical solution, machine unlearning aims to selectively erase unwanted concepts from a pre-trained model without retraining from scratch. While most existing methods are effective for single-concept unlearning, they often struggle in real-world scenarios that require removing multiple concepts, since extending them to this setting is both non-trivial and problematic, causing significant challenges in unlearning effectiveness, generation quality, and sensitivity to hyperparameters and datasets. In this paper, we take a unique perspective on multi-concept unlearning by leveraging model sparsity and propose the Forget It All (FIA) framework. FIA first introduces Contrastive Concept Saliency to quantify each weight connection's contribution to a target concept. It then identifies Concept-Sensitive Neurons by combining temporal and spatial information, ensuring that only neurons consistently responsive to the target concept are selected. Finally, FIA constructs masks from the identified neurons and fuses them into a unified multi-concept mask, where Concept-Agnostic Neurons that broadly support general content generation are preserved while concept-specific neurons are pruned to remove the targets. FIA is training-free and requires only minimal hyperparameter tuning for new tasks, thereby promoting a plug-and-play paradigm. Extensive experiments across three distinct unlearning tasks demonstrate that FIA achieves more reliable multi-concept unlearning, improving forgetting effectiveness while maintaining semantic fidelity and image quality.
Authors: Jonas Klingwort, Nina M. Leach, Joep Burger
Abstract: Machine learning (ML) is a tool to exploit remote sensing data for the monitoring and implementation of the United Nations' Sustainable Development Goals (SDGs). In this paper, we report on a meta-analysis to evaluate the performance of ML applied to remote sensing data to monitor SDGs. Specifically, we aim to 1) estimate the average performance; 2) determine the degree of heterogeneity between and within studies; and 3) assess how study features influence model performance. Using PRISMA guidelines, a search was performed across multiple academic databases to identify potentially relevant studies. A random sample of 200 was screened by three reviewers, resulting in 86 trials within 20 studies with 14 study features. Overall accuracy was the most reported performance metric. It was analyzed using double arcsine transformation and a three-level random effects model. The average overall accuracy of the best model was 0.90 [0.86, 0.92]. There was considerable heterogeneity in model performance, 64% of which was between studies. The only significant feature was the prevalence of the majority class, which explained 61% of the between-study heterogeneity. None of the other thirteen features added value to the model. The most important contributions of this paper are the following two insights. 1) Overall accuracy is the most popular performance metric, yet arguably the least insightful. Its sensitivity to class imbalance makes it necessary to normalize it, which is far from common practice. 2) The field needs to standardize the reporting. Reporting of the confusion matrix for independent test sets is the most important ingredient for between-study comparisons of ML classifiers. These findings underscore the need for robust and comparable evaluation metrics in machine learning applications to ensure reliable and actionable insights for effective SDG monitoring and policy formulation.
Authors: Yung-Shen Hsia, Fang Yu, Jie-Hong Roland Jiang
Abstract: Financial regulations are increasingly complex, hindering automated compliance-especially the maintenance of logical consistency with minimal human oversight. We introduce a Neuro-Symbolic Compliance Framework that integrates Large Language Models (LLMs) with Satisfiability Modulo Theories (SMT) solvers to enable formal verifiability and optimization-based compliance correction. The LLM interprets statutes and enforcement cases to generate SMT constraints, while the solver enforces consistency and computes the minimal factual modification required to restore legality when penalties arise. Unlike transparency-oriented methods, our approach emphasizes logic-driven optimization, delivering verifiable, legally consistent reasoning rather than post-hoc explanation. Evaluated on 87 enforcement cases from Taiwan's Financial Supervisory Commission (FSC), the system attains 86.2% correctness in SMT code generation, improves reasoning efficiency by over 100x, and consistently corrects violations-establishing a preliminary foundation for optimization-based compliance applications.
Authors: Atharv Naphade
Abstract: Retrieval-Augmented Generation (RAG) is the prevailing paradigm for grounding Large Language Models (LLMs), yet the mechanisms governing how models integrate groups of conflicting retrieved evidence remain opaque. Does an LLM answer a certain way because the evidence is factually strong, because of a prior belief, or merely because it is repeated frequently? To answer this, we introduce GroupQA, a curated dataset of 1,635 controversial questions paired with 15,058 diversely-sourced evidence documents, annotated for stance and qualitative strength. Through controlled experiments, we characterize group-level evidence aggregation dynamics: Paraphrasing an argument can be more persuasive than providing distinct independent support; Models favor evidence presented first rather than last, and Larger models are increasingly resistant to adapt to presented evidence. Additionally, we find that LLM explanations to group-based answers are unfaithful. Together, we show that LLMs behave consistently as vulnerable heuristic followers, with direct implications for improving RAG system design.
Authors: Zhanyu Liu, Yang Yu
Abstract: Causal relationships play a pivotal role in research within the field of public administration. Ensuring reliable causal inference requires validating the predictability of these relationships, which is a crucial precondition. However, prediction has not garnered adequate attention within the realm of quantitative research in public administration and the broader social sciences. The advent of interpretable machine learning presents a significant opportunity to integrate prediction into quantitative research conducted in public administration. This article delves into the fundamental principles of interpretable machine learning while also examining its current applications in social science research. Building upon this foundation, the article further expounds upon the implementation process of interpretable machine learning, encompassing key aspects such as dataset construction, model training, model evaluation, and model interpretation. Lastly, the article explores the disciplinary value of interpretable machine learning within the field of public administration, highlighting its potential to enhance the generalization of inference, facilitate the selection of optimal explanations for phenomena, stimulate the construction of theoretical hypotheses, and provide a platform for the translation of knowledge. As a complement to traditional causal inference methods, interpretable machine learning ushers in a new era of credibility in quantitative research within the realm of public administration.
Authors: Keerthi Kumar. M, Swarun Kumar Joginpelly, Sunil Khemka, Lakshmi. S R, Navin Chhibber
Abstract: Background: Cyber-attacks have evolved rapidly in recent years, many individuals and business owners have been affected by cyber-attacks in various ways. Cyber-attacks include various threats such as ransomware, malware, phishing, and Denial of Service (DoS)-related attacks. Challenges: Traditional models such as Generative Artificial Intelligence (AI) and Security Bidirectional Encoder Representations from Transformers (BERT) were implemented to detect cyber threats. However, the existing Security BERT model has a limited contextual understanding of text data, which has less impact on detecting cyber-attacks. Proposed Methodology: To overcome the above-mentioned challenges, Robustly Optimized Bidirectional Encoder Representations from Transformers Pretraining Approach (RoBERTa) model is proposed which consists of diverse words of vocabulary understanding. Initially, data are extracted from a Packet Capture (PCAP) file and encrypted using Fully Harmonic Encryption (FHE). Subsequently, a Byte-level and Byte Pair Encoding (BBPE) tokenizer was used to generate tokens and help maintain the vocabulary for the encrypted values. Then, these values are applied to the RoBERTa model of the transformer with extensive training. Finally, Softmax is used for the detection and classification of attacks. The proposed RoBERTa model achieved better results than the existing BERT model in terms of accuracy (0.99), recall (0.91), and precision (0.89) respectively.
Authors: Miranda J. S. Horne (University of Leeds), Peter K. Jimack (University of Leeds), Amirul Khan (University of Leeds), He Wang (University College London)
Abstract: In this work, we embed hard constraints in a physics informed neural network (PINN) which predicts solutions to the 2D incompressible Navier Stokes equations. We extend the hard constraint method introduced by Chen et al. (arXiv:2012.06148) from a linear PDE to a strongly non-linear PDE. The PINN is used to estimate the stream function and pressure of the fluid, and by differentiating the stream function we can recover an incompressible velocity field. An unlearnable hard constraint projection (HCP) layer projects the predicted velocity and pressure to a hyperplane that admits only exact solutions to a discretised form of the governing equations.
Authors: Yufeng Wang, Lu Wei, Lin Liu, Hao Xu, Haibin Ling
Abstract: Mass spectrometry (MS) is a powerful analytical technique for identifying small molecules, yet determining complete molecular structures directly from tandem mass spectra (MS/MS) remains a long-standing challenge due to complex fragmentation patterns and the vast diversity of chemical space. Recent progress in large language models (LLMs) has shown promise for reasoning-intensive scientific tasks, but their capability for chemical interpretation is still unclear. In this work, we introduce a Chain-of-Thought (CoT) prompting framework and benchmark that evaluate how LLMs reason about mass spectral data to predict molecular structures. We formalize expert chemists' reasoning steps-such as double bond equivalent (DBE) analysis, neutral loss identification, and fragment assembly-into structured prompts and assess multiple state-of-the-art LLMs (Claude-3.5-Sonnet, GPT-4o-mini, and Llama-3 series) in a zero-shot setting using the MassSpecGym dataset. Our evaluation across metrics of SMILES validity, formula consistency, and structural similarity reveals that while LLMs can produce syntactically valid and partially plausible structures, they fail to achieve chemical accuracy or link reasoning to correct molecular predictions. These findings highlight both the interpretive potential and the current limitations of LLM-based reasoning for molecular elucidation, providing a foundation for future work that combines domain knowledge and reinforcement learning to achieve chemically grounded AI reasoning.
Authors: Trisha Das, Mandis Beigi, Jacob Aptekar, Jimeng Sun
Abstract: Clinical trial amendments frequently introduce delays, increased costs, and administrative burden, with eligibility criteria being the most commonly amended component. We introduce \textit{eligibility criteria amendment prediction}, a novel NLP task that aims to forecast whether the eligibility criteria of an initial trial protocol will undergo future amendments. To support this task, we release $\texttt{AMEND++}$, a benchmark suite comprising two datasets: $\texttt{AMEND}$, which captures eligibility-criteria version histories and amendment labels from public clinical trials, and $\verb|AMEND_LLM|$, a refined subset curated using an LLM-based denoising pipeline to isolate substantive changes. We further propose $\textit{Change-Aware Masked Language Modeling}$ (CAMLM), a revision-aware pretraining strategy that leverages historical edits to learn amendment-sensitive representations. Experiments across diverse baselines show that CAMLM consistently improves amendment prediction, enabling more robust and cost-effective clinical trial design.
Authors: Binxu Wang, Jingxuan Fan, Xu Pan
Abstract: Diffusion Transformers (DiTs) have greatly advanced text-to-image generation, but models still struggle to generate the correct spatial relations between objects as specified in the text prompt. In this study, we adopt a mechanistic interpretability approach to investigate how a DiT can generate correct spatial relations between objects. We train, from scratch, DiTs of different sizes with different text encoders to learn to generate images containing two objects whose attributes and spatial relations are specified in the text prompt. We find that, although all the models can learn this task to near-perfect accuracy, the underlying mechanisms differ drastically depending on the choice of text encoder. When using random text embeddings, we find that the spatial-relation information is passed to image tokens through a two-stage circuit, involving two cross-attention heads that separately read the spatial relation and single-object attributes in the text prompt. When using a pretrained text encoder (T5), we find that the DiT uses a different circuit that leverages information fusion in the text tokens, reading spatial-relation and single-object information together from a single text token. We further show that, although the in-domain performance is similar for the two settings, their robustness to out-of-domain perturbations differs, potentially suggesting the difficulty of generating correct relations in real-world scenarios.
Authors: Mara Pleasure, Ekaterina Redekop, Dhakshina Ilango, Zichen Wang, Vedrana Ivezic, Kimberly Flores, Israa Laklouk, Jitin Makker, Gregory Fishbein, Anthony Sisk, William Speier, Corey W. Arnold
Abstract: Current histopathological grading of prostate cancer relies primarily on glandular architecture, largely overlooking the tumor microenvironment. Here, we present PROTAS, a deep learning framework that quantifies reactive stroma (RS) in routine hematoxylin and eosin (H&E) slides and links stromal morphology to underlying biology. PROTAS-defined RS is characterized by nuclear enlargement, collagen disorganization, and transcriptomic enrichment of contractile pathways. PROTAS detects RS robustly in the external Prostate, Lung, Colorectal, and Ovarian (PLCO) dataset and, using domain-adversarial training, generalizes to diagnostic biopsies. In head-to-head comparisons, PROTAS outperforms pathologists for RS detection, and spatial RS features predict biochemical recurrence independently of established prognostic variables (c-index 0.80). By capturing subtle stromal phenotypes associated with tumor progression, PROTAS provides an interpretable, scalable biomarker to refine risk stratification.
Authors: Zakaria Baba, Alexandre M. Bayen, Alexi Canesse, Maria Laura Delle Monache, Martin Drieux, Zhe Fu, Nathan Lichtl\'e, Zihe Liu, Hossein Nick Zinat Matin, Benedetto Piccoli
Abstract: We present a neural network-based method for learning scalar hyperbolic conservation laws. Our method replaces the traditional numerical flux in finite volume schemes with a trainable neural network while preserving the conservative structure of the scheme. The model can be trained both in a supervised setting with efficiently generated synthetic data or in an unsupervised manner, leveraging the weak formulation of the partial differential equation. We provide theoretical results that our model can perform arbitrarily well, and provide associated upper bounds on neural network size. Extensive experiments demonstrate that our method often outperforms efficient schemes such as Godunov's scheme, WENO, and Discontinuous Galerkin for comparable computational budgets. Finally, we demonstrate the effectiveness of our method on a traffic prediction task, leveraging field experimental highway data from the Berkeley DeepDrive drone dataset.
Authors: Jun Qi, Chao-Han Huck Yang, Pin-Yu Chen, Javier Tejedor, Ling Li, Min-Hsiu Hsieh
Abstract: We introduce CL-QAS, a continual quantum architecture search framework that mitigates the challenges of costly amplitude encoding and catastrophic forgetting in variational quantum circuits. The method uses Tensor-Train encoding to efficiently compress high-dimensional stochastic signals into low-rank quantum feature representations. A bi-loop learning strategy separates circuit parameter optimization from architecture exploration, while an Elastic Weight Consolidation regularization ensures stability across sequential tasks. We derive theoretical upper bounds on approximation, generalization, and robustness under quantum noise, demonstrating that CL-QAS achieves controllable expressivity, sample-efficient generalization, and smooth convergence without barren plateaus. Empirical evaluations on electrocardiogram (ECG)-based signal classification and financial time-series forecasting confirm substantial improvements in accuracy, balanced accuracy, F1 score, and reward. CL-QAS maintains strong forward and backward transfer and exhibits bounded degradation under depolarizing and readout noise, highlighting its potential for adaptive, noise-resilient quantum learning on near-term devices.
Authors: Abhinav Raghuvanshi, Mayank Baranwal, Debasish Chatterjee
Abstract: We introduce $\textsf{gradOL}$, the first gradient-based optimization framework for solving Chebyshev center problems, a fundamental challenge in optimal function learning and geometric optimization. $\textsf{gradOL}$ hinges on reformulating the semi-infinite problem as a finitary max-min optimization, making it amenable to gradient-based techniques. By leveraging automatic differentiation for precise numerical gradient computation, $\textsf{gradOL}$ ensures numerical stability and scalability, making it suitable for large-scale settings. Under strong convexity of the ambient norm, $\textsf{gradOL}$ provably recovers optimal Chebyshev centers while directly computing the associated radius. This addresses a key bottleneck in constructing stable optimal interpolants. Empirically, $\textsf{gradOL}$ achieves significant improvements in accuracy and efficiency on 34 benchmark Chebyshev center problems from a benchmark $\textsf{CSIP}$ library. Moreover, we extend $\textsf{gradOL}$ to general convex semi-infinite programming (CSIP), attaining up to $4000\times$ speedups over the state-of-the-art $\texttt{SIPAMPL}$ solver tested on the indicated $\textsf{CSIP}$ library containing 67 benchmark problems. Furthermore, we provide the first theoretical foundation for applying gradient-based methods to Chebyshev center problems, bridging rigorous analysis with practical algorithms. $\textsf{gradOL}$ thus offers a unified solution framework for Chebyshev centers and broader CSIPs.
Authors: Sayak Chakrabarty, Souradip Pal
Abstract: Large Language Models (LLMs) have recently shown strong potential for usage in sequential recommendation tasks through text-only models, which combine advanced prompt design, contrastive alignment, and fine-tuning on downstream domain-specific data. While effective, these approaches overlook the rich visual information present in many real-world recommendation scenarios, particularly in e-commerce. This paper proposes PixRec - a vision-language framework that incorporates both textual attributes and product images into the recommendation pipeline. Our architecture leverages a vision-language model backbone capable of jointly processing image-text sequences, maintaining a dual-tower structure and mixed training objective while aligning multi-modal feature projections for both item-item and user-item interactions. Using the Amazon Reviews dataset augmented with product images, our experiments demonstrate $3\times$ and 40% improvements in top-rank and top-10 rank accuracy over text-only recommenders respectively, indicating that visual features can help distinguish items with similar textual descriptions. Our work outlines future directions for scaling multi-modal recommenders training, enhancing visual-text feature fusion, and evaluating inference-time performance. This work takes a step toward building software systems utilizing visual information in sequential recommendation for real-world applications like e-commerce.
Authors: Tianming Bai, Jiannan Yang
Abstract: Applying Physics-Informed Gaussian Process Regression to the eigenvalue problem $(\mathcal{L}-\lambda)u = 0$ poses a fundamental challenge, where the null source term results in a trivial predictive mean and a degenerate marginal likelihood. Drawing inspiration from system identification, we construct a transfer function-type indicator for the unknown eigenvalue/eigenfunction using the physics-informed Gaussian Process posterior. We demonstrate that the posterior covariance is only non-trivial when $\lambda$ corresponds to an eigenvalue of the partial differential operator $\mathcal{L}$, reflecting the existence of a non-trivial eigenspace, and any sample from the posterior lies in the eigenspace of the linear operator. We demonstrate the effectiveness of the proposed approach through several numerical examples with both linear and non-linear eigenvalue problems.
Authors: Junho Park, Dohoon Kim, Taesup Moon
Abstract: Large language model (LLM) personalization aims to adapt general-purpose models to individual users. Most existing methods, however, are developed under data-rich and resource-abundant settings, often incurring privacy risks. In contrast, realistic personalization typically occurs after deployment under (i) extremely limited user data, (ii) constrained computational resources, and (iii) strict privacy requirements. We propose PRISP, a lightweight and privacy-safe personalization framework tailored to these constraints. PRISP leverages a Text-to-LoRA hypernetwork to generate task-aware LoRA parameters from task descriptions, and enables efficient user personalization by optimizing a small subset of task-aware LoRA parameters together with minimal additional modules using few-shot user data. Experiments on a few-shot variant of the LaMP benchmark demonstrate that PRISP achieves strong overall performance compared to prior approaches, while reducing computational overhead and eliminating privacy risks.
Authors: Mundla Narasimhappa, Praveen Kumar
Abstract: Accurate prediction of joint kinematics and kinetics is essential for advancing gait analysis and developing intelligent assistive systems such as prosthetics and exoskeletons. This study presents a hybrid LSTM-UKF framework for estimating ankle angle and ground reaction force (GRF) across varying walking speeds. A multimodal sensor fusion strategy integrates force plate data, knee angle, and GRF signals to enrich biomechanical context. Model performance was evaluated using RMSE and $R^2$ under subject-specific validation. The LSTM-UKF consistently outperformed standalone LSTM and UKF models, achieving up to 18.6\% lower RMSE for GRF prediction at 3 km/h. Additionally, UKF integration improved robustness, reducing ankle angle RMSE by up to 22.4\% compared to UKF alone at 1 km/h. These results underscore the effectiveness of hybrid architectures for reliable gait prediction across subjects and walking conditions.
Authors: Jinyuan Chang, Chenguang Duan, Yuling Jiao, Yi Xu, Jerry Zhijian Yang
Abstract: Inference-time alignment for diffusion models aims to adapt a pre-trained diffusion model toward a target distribution without retraining the base score network, thereby preserving the generative capacity of the base model while enforcing desired properties at the inference time. A central mechanism for achieving such alignment is guidance, which modifies the sampling dynamics through an additional drift term. In this work, we introduce Doob's matching, a novel framework for guidance estimation grounded in Doob's $h$-transform. Our approach formulates guidance as the gradient of logarithm of an underlying Doob's $h$-function and employs gradient-penalized regression to simultaneously estimate both the $h$-function and its gradient, resulting in a consistent estimator of the guidance. Theoretically, we establish non-asymptotic convergence rates for the estimated guidance. Moreover, we analyze the resulting controllable diffusion processes and prove non-asymptotic convergence guarantees for the generated distributions in the 2-Wasserstein distance.
Authors: Carl Vincent Ladres Kho
Abstract: Consumer-grade biosensors offer a cost-effective alternative to medical-grade electromyography (EMG) systems, reducing hardware costs from thousands of dollars to approximately $13. However, these low-cost sensors introduce significant signal instability and motion artifacts. Deploying machine learning models on resource-constrained edge devices like the ESP32 presents a challenge: balancing classification accuracy with strict latency (<100ms) and memory (<320KB) constraints. Using a single-subject dataset comprising 1,540 seconds of raw data (1.54M data points, segmented into ~1,300 one-second windows), I evaluate 18 model architectures, ranging from statistical heuristics to deep transfer learning (ResNet50) and custom hybrid networks (MaxCRNN). While my custom "MaxCRNN" (Inception + Bi-LSTM + Attention) achieved the highest safety (99% Precision) and robustness, I identify Random Forest (74% accuracy) as the Pareto-optimal solution for embedded control on legacy microcontrollers. I demonstrate that reliable, low-latency EMG control is feasible on commodity hardware, with Deep Learning offering a path to near-perfect reliability on modern Edge AI accelerators.
Authors: Jun-Qi Chen, Kun Zhang, Rui Zheng, Ying Zhong
Abstract: The Python package SimPy is widely used for modeling queueing systems due to its flexibility, simplicity, and smooth integration with modern data analysis and optimization frameworks. Recent advances in large language models (LLMs) have shown strong ability in generating clear and executable code, making them powerful and suitable tools for writing SimPy queueing simulation code. However, directly employing closed-source models like GPT-4o to generate such code may lead to high computational costs and raise data privacy concerns. To address this, we fine-tune two open-source LLMs, Qwen-Coder-7B and DeepSeek-Coder-6.7B, on curated SimPy queueing data, which enhances their code-generating performance in executability, output-format compliance, and instruction-code consistency. Particularly, we proposed a multi-stage fine-tuning framework comprising two stages of supervised fine-tuning (SFT) and one stage of direct preference optimization (DPO), progressively enhancing the model's ability in SimPy-based queueing simulation code generation. Extensive evaluations demonstrate that both fine-tuned models achieve substantial improvements in executability, output-format compliance, and instruct consistency. These results confirm that domain-specific fine-tuning can effectively transform compact open-source code models into reliable SimPy simulation generators which provide a practical alternative to closed-source LLMs for education, research, and operational decision support.
Authors: Hongyi Zhou, Jin Zhu, Ying Yang, Chengchun Shi
Abstract: Large language models (LLMs) such as GPT, Claude, Gemini, and Grok have been deeply integrated into our daily life. They now support a wide range of tasks -- from dialogue and email drafting to assisting with teaching and coding, serving as search engines, and much more. However, their ability to produce highly human-like text raises serious concerns, including the spread of fake news, the generation of misleading governmental reports, and academic misconduct. To address this practical problem, we train a classifier to determine whether a piece of text is authored by an LLM or a human. Our detector is deployed on an online CPU-based platform https://huggingface.co/spaces/stats-powered-ai/StatDetectLLM, and contains three novelties over existing detectors: (i) it does not rely on auxiliary information, such as watermarks or knowledge of the specific LLM used to generate the text; (ii) it more effectively distinguishes between human- and LLM-authored text; and (iii) it enables statistical inference, which is largely absent in the current literature. Empirically, our classifier achieves higher classification accuracy compared to existing detectors, while maintaining type-I error control, high statistical power, and computational efficiency.
URLs: https://huggingface.co/spaces/stats-powered-ai/StatDetectLLM,
Authors: Mohammed S. Alharbi, Shinkyu Park
Abstract: Inertial localization is particularly valuable in GPS-denied environments such as indoors. However, localization using only Inertial Measurement Units (IMUs) suffers from drift caused by motion-process noise and sensor biases. This paper introduces Uncertainty-aware Map-constrained Inertial Localization (UMLoc), an end-to-end framework that jointly models IMU uncertainty and map constraints to achieve drift-resilient positioning. UMLoc integrates two coupled modules: (1) a Long Short-Term Memory (LSTM) quantile regressor, which estimates the specific quantiles needed to define 68%, 90%, and 95% prediction intervals serving as a measure of localization uncertainty and (2) a Conditioned Generative Adversarial Network (CGAN) with cross-attention that fuses IMU dynamic data with distance-based floor-plan maps to generate geometrically feasible trajectories. The modules are trained jointly, allowing uncertainty estimates to propagate through the CGAN during trajectory generation. UMLoc was evaluated on three datasets, including a newly collected 2-hour indoor benchmark with time-aligned IMU data, ground-truth poses and floor-plan maps. Results show that the method achieves a mean drift ratio of 5.9% over a 70 m travel distance and an average Absolute Trajectory Error (ATE) of 1.36 m, while maintaining calibrated prediction bounds.
Authors: Rodion Vakhitov, Leonid Ugadiarov, Aleksandr Panov
Abstract: In this paper, we introduce ObjectZero, a novel reinforcement learning (RL) algorithm that leverages the power of object-level representations to model dynamic environments more effectively. Unlike traditional approaches that process the world as a single undifferentiated input, our method employs Graph Neural Networks (GNNs) to capture intricate interactions among multiple objects. These objects, which can be manipulated and interact with each other, serve as the foundation for our model's understanding of the environment. We trained the algorithm in a complex setting teeming with diverse, interactive objects, demonstrating its ability to effectively learn and predict object dynamics. Our results highlight that a structured world model operating on object-centric representations can be successfully integrated into a model-based RL algorithm utilizing Monte Carlo Tree Search as a planning module.
Authors: Tanisha Raorane, Prasenjit Kole
Abstract: Sanskrit Subhasitas encapsulate centuries of cultural and philosophical wisdom, yet remain underutilized in the digital age due to linguistic and contextual barriers. In this work, we present Pragya, a retrieval-augmented generation (RAG) framework for semantic recommendation of Subhasitas. We curate a dataset of 200 verses annotated with thematic tags such as motivation, friendship, and compassion. Using sentence embeddings (IndicBERT), the system retrieves top-k verses relevant to user queries. The retrieved results are then passed to a generative model (Mistral LLM) to produce transliterations, translations, and contextual explanations. Experimental evaluation demonstrates that semantic retrieval significantly outperforms keyword matching in precision and relevance, while user studies highlight improved accessibility through generated summaries. To our knowledge, this is the first attempt at integrating retrieval and generation for Sanskrit Subhasitas, bridging cultural heritage with modern applied AI.
Authors: Chalitha Handapangoda
Abstract: The reliance of Large Language Models and Internet of Things systems on massive, globally distributed data flows creates systemic security and privacy challenges. When data traverses borders, it becomes subject to conflicting legal regimes, such as the EU's General Data Protection Regulation and China's Personal Information Protection Law, compounded by technical vulnerabilities like model memorization. Current static encryption and data localization methods are fragmented and reactive, failing to provide adequate, policy-aligned safeguards. This research proposes a Jurisdiction-Aware, Privacy-by-Design architecture that dynamically integrates localized encryption, adaptive differential privacy, and real-time compliance assertion via cryptographic proofs. Empirical validation in a multi-jurisdictional simulation demonstrates this architecture reduced unauthorized data exposure to below five percent and achieved zero compliance violations. These security gains were realized while maintaining model utility retention above ninety percent and limiting computational overhead. This establishes that proactive, integrated controls are feasible for secure and globally compliant AI deployment.
Authors: Luis Alberto Croquevielle, Roman Sokolovskii, Thomas Heinis
Abstract: Learned index structures aim to accelerate queries by training machine learning models to approximate the rank function associated with a database attribute. While effective in practice, their theoretical limitations are not fully understood. We present a general framework for proving lower bounds on query time for learned indexes, expressed in terms of their space overhead and parameterized by the model class used for approximation. Our formulation captures a broad family of learned indexes, including most existing designs, as piecewise model-based predictors. We solve the problem of lower bounding query time in two steps: first, we use probabilistic tools to control the effect of sampling when the database attribute is drawn from a probability distribution. Then, we analyze the approximation-theoretic problem of how to optimally represent a cumulative distribution function with approximators from a given model class. Within this framework, we derive lower bounds under a range of modeling and distributional assumptions, paying particular attention to the case of piecewise linear and piecewise constant model classes, which are common in practical implementations. Our analysis shows how tools from approximation theory, such as quantization and Kolmogorov widths, can be leveraged to formalize the space-time tradeoffs inherent to learned index structures. The resulting bounds illuminate core limitations of these methods.
Authors: Juan Miguel L\'opez Alcaraz, Xicot\'encatl L\'opez Moran, Erick D\'avila Zaragoza, Claas H\"andel, Richard Koebe, Wilhelm Haverkamp, Nils Strodthoff
Abstract: Artificial intelligence holds strong potential to support clinical decision making in intensive care units where timely and accurate risk assessment is critical. However, many existing models focus on isolated outcomes or limited data types, while clinicians integrate longitudinal history, real time physiology, and heterogeneous clinical information. To address this gap, we developed MDS ICU, a unified multimodal machine learning framework that fuses routinely collected data including demographics, biometrics, vital signs, laboratory values, ECG waveforms, surgical procedures, and medical device usage to provide continuous predictive support during ICU stays. Using 63001 samples from 27062 patients in MIMIC IV, we trained a deep learning architecture that combines structured state space S4 encoders for ECG waveforms with multilayer perceptron RealMLP encoders for tabular data to jointly predict 33 clinically relevant outcomes spanning mortality, organ dysfunction, medication needs, and acute deterioration. The model achieved strong discrimination with AUROCs of 0.90 for 24 hour mortality, 0.92 for sedative administration, 0.97 for invasive mechanical ventilation, and 0.93 for coagulation dysfunction. Calibration analysis showed close agreement between predicted and observed risks, with consistent gains from ECG waveform integration. Comparisons with clinicians and large language models showed that model predictions alone outperformed both, and that providing model outputs as decision support further improved their performance. These results demonstrate that multimodal AI can deliver clinically meaningful risk stratification across diverse ICU outcomes while augmenting rather than replacing clinical expertise, establishing a scalable foundation for precision critical care decision support.
Authors: Yifeng Yu, Lu Yu
Abstract: Score-based diffusion models have become a powerful framework for generative modeling, with score estimation as a central statistical bottleneck. Existing guarantees for score estimation largely focus on light-tailed targets or rely on restrictive assumptions such as compact support, which are often violated by heavy-tailed data in practice. In this work, we study conventional (Gaussian) score-based diffusion models when the target distribution is heavy-tailed and belongs to a Sobolev class with smoothness parameter $\beta>0$. We consider both exponential and polynomial tail decay, indexed by a tail parameter $\gamma$. Using kernel density estimation, we derive sharp minimax rates for score estimation, revealing a qualitative dichotomy: under exponential tails, the rate matches the light-tailed case up to polylogarithmic factors, whereas under polynomial tails the rate depends explicitly on $\gamma$. We further provide sampling guarantees for the associated continuous reverse dynamics. In total variation, the generated distribution converges at the minimax optimal rate $n^{-\beta/(2\beta+d)}$ under exponential tails (up to logarithmic factors), and at a $\gamma$-dependent rate under polynomial tails. Whether the latter sampling rate is minimax optimal remains an open question. These results characterize the statistical limits of score estimation and the resulting sampling accuracy for heavy-tailed targets, extending diffusion theory beyond the light-tailed setting.
Authors: Kunming Shao, Liang Zhao, Jiangnan Yu, Zhipeng Liao, Xiaomeng Wang, Yi Zou, Tim Kwang-Ting Cheng, Chi-Ying Tsui
Abstract: Stochastic computing (SC) offers hardware simplicity but suffers from low throughput, while high-throughput Digital Computing-in-Memory (DCIM) is bottlenecked by costly adder logic for matrix-vector multiplication (MVM). To address this trade-off, this paper introduces a digital stochastic CIM (DS-CIM) architecture that achieves both high accuracy and efficiency. We implement signed multiply-accumulation (MAC) in a compact, unsigned OR-based circuit by modifying the data representation. Throughput is enhanced by replicating this low-cost circuit 64 times with only a 1x area increase. Our core strategy, a shared Pseudo Random Number Generator (PRNG) with 2D partitioning, enables single-cycle mutually exclusive activation to eliminate OR-gate collisions. We also resolve the 1s saturation issue via stochastic process analysis and data remapping, significantly improving accuracy and resilience to input sparsity. Our high-accuracy DS-CIM1 variant achieves 94.45% accuracy for INT8 ResNet18 on CIFAR-10 with a root-mean-squared error (RMSE) of just 0.74%. Meanwhile, our high-efficiency DS-CIM2 variant attains an energy efficiency of 3566.1 TOPS/W and an area efficiency of 363.7 TOPS/mm^2, while maintaining a low RMSE of 3.81%. The DS-CIM capability with larger models is further demonstrated through experiments with INT8 ResNet50 on ImageNet and the FP8 LLaMA-7B model.
Authors: Tamara Alshammari, Mehdi Bennis
Abstract: The advent of 6G networks is accelerating autonomy and intelligence in large-scale, decentralized multi-agent systems (MAS). While this evolution enables adaptive behavior, it also heightens vulnerability to stressors such as environmental changes and adversarial behavior. Existing literature on resilience in decentralized MAS largely focuses on isolated aspects, such as fault tolerance, without offering a principled unified definition of multi-agent resilience. This gap limits the ability to design systems that can continuously sense, adapt, and recover under dynamic conditions. This article proposes a formal definition of MAS resilience grounded in two complementary dimensions: epistemic resilience, wherein agents recover and sustain accurate knowledge of the environment, and action resilience, wherein agents leverage that knowledge to coordinate and sustain goals under disruptions. We formalize resilience via temporal epistemic logic and quantify it using recoverability time (how quickly desired properties are re-established after a disturbance) and durability time (how long accurate beliefs and goal-directed behavior are sustained after recovery). We design an agent architecture and develop decentralized algorithms to achieve both epistemic and action resilience. We provide formal verification guarantees, showing that our specifications are sound with respect to the metric bounds and admit finite-horizon verification, enabling design-time certification and lightweight runtime monitoring. Through a case study on distributed multi-agent decision-making under stressors, we show that our approach outperforms baseline methods. Our formal verification analysis and simulation results highlight that the proposed framework enables resilient, knowledge-driven decision-making and sustained operation, laying the groundwork for resilient decentralized MAS in next-generation communication systems.
Authors: Josh Li
Abstract: Feedback-rich neural architectures can regenerate earlier representations and inject temporal context, making them a natural setting for strictly local synaptic plasticity. We ask whether a minimal, backpropagation-free feedback--Hebbian system can already express interpretable continual-learning--relevant behaviors under controlled training schedules. We introduce a compact prediction--reconstruction architecture with two feedforward layers for supervised association learning and two dedicated feedback layers trained to reconstruct earlier activity and re-inject it as additive temporal context. All synapses are updated by a unified local rule combining centered Hebbian covariance, Oja-style stabilization, and a local supervised drive where targets are available, requiring no weight transport or global error backpropagation. On a small two-pair association task, we characterize learning through layer-wise activity snapshots, connectivity trajectories (row/column means of learned weights), and a normalized retention index across phases. Under sequential A->B training, forward output connectivity exhibits a long-term depression (LTD)-like suppression of the earlier association while feedback connectivity preserves an A-related trace during acquisition of B. Under deterministic interleaving A,B,A,B,..., both associations are concurrently maintained rather than sequentially suppressed. Architectural controls and rule-term ablations isolate the role of dedicated feedback in regeneration and co-maintenance, and the role of the local supervised term in output selectivity and unlearning. Together, the results show that a compact feedback pathway trained with local plasticity can support regeneration and continual-learning--relevant dynamics in a minimal, mechanistically transparent setting.
Authors: Xiaoyin Xi, Neeku Capak, Kate Stockwell, Zhe Yu
Abstract: This research seeks to benefit the software engineering society by proposing comparative separation, a novel group fairness notion to evaluate the fairness of machine learning software on comparative judgment test data. Fairness issues have attracted increasing attention since machine learning software is increasingly used for high-stakes and high-risk decisions. It is the responsibility of all software developers to make their software accountable by ensuring that the machine learning software do not perform differently on different sensitive groups -- satisfying the separation criterion. However, evaluation of separation requires ground truth labels for each test data point. This motivates our work on analyzing whether separation can be evaluated on comparative judgment test data. Instead of asking humans to provide the ratings or categorical labels on each test data point, comparative judgments are made between pairs of data points such as A is better than B. According to the law of comparative judgment, providing such comparative judgments yields a lower cognitive burden for humans than providing ratings or categorical labels. This work first defines the novel fairness notion comparative separation on comparative judgment test data, and the metrics to evaluate comparative separation. Then, both theoretically and empirically, we show that in binary classification problems, comparative separation is equivalent to separation. Lastly, we analyze the number of test data points and test data pairs required to achieve the same level of statistical power in the evaluation of separation and comparative separation, respectively. This work is the first to explore fairness evaluation on comparative judgment test data. It shows the feasibility and the practical benefits of using comparative judgment test data for model evaluations.
Authors: Shubhashis Roy Dipta, Khairul Mahbub, Nadia Najjar
Abstract: We present a Bengali mathematical reasoning model called GanitLLM (named after the Bangla word for mathematics, "Ganit"), together with a new difficulty-aware Bengali math corpus and a curriculum-based GRPO pipeline. Bengali is one of the world's most widely spoken languages, yet existing LLMs either reason in English and then translate, or simply fail on multi-step Bengali math, in part because reinforcement learning recipes are tuned for high-resource languages and collapse under reward sparsity in low-resource settings. To address this, we construct Ganit, a rigorously filtered and decontaminated Bengali math dataset with automatic difficulty tags derived from the pass@k of a strong evaluator model. Building on this dataset, we propose Curriculum-GRPO, which combines multi-stage training (SFT + GRPO) with difficulty-aware sampling and verifiable rewards for format, numerical correctness, and Bengali reasoning. On Bn-MGSM and Bn-MSVAMP, GanitLLM-4B improves over its Qwen3-4B base by +8 and +7 accuracy points, respectively, while increasing the percentage of Bengali reasoning tokens from 14% to over 88% and reducing average solution length from 943 to 193 words.
Authors: Muhammad Wahid Akram, Keshav Sood, Muneeb Ul Hassan, Dhananjay Thiruvady
Abstract: Phishing with Quick Response (QR) codes is termed as Quishing. The attackers exploit this method to manipulate individuals into revealing their confidential data. Recently, we see the colorful and fancy representations of QR codes, the 2D matrix of QR codes which does not reflect a typical mixture of black-white modules anymore. Instead, they become more tempting as an attack vector for adversaries which can evade the state-of-the-art deep learning visual-based and other prevailing countermeasures. We introduce "ALFA", a safe-by-design approach, to mitigate Quishing and prevent everyone from accessing the post-scan harmful payload of fancy QR codes. Our method first converts a fancy QR code into the replica of binary grid and then identify the erroneous representation of modules in that grid. Following that, we present "FAST" method which can conveniently recover erroneous modules from that binary grid. Afterwards, using this binary grid, our solution extracts the structural features of fancy QR code and predicts its legitimacy using a pre-trained model. The effectiveness of our proposal is demonstrated by the experimental evaluation on a synthetic dataset (containing diverse variations of fancy QR codes) and achieve a FNR of 0.06% only. We also develop the mobile app to test the practical feasibility of our solution and provide a performance comparison of the app with the real-world QR readers. This comparison further highlights the classification reliability and detection accuracy of this solution in real-world environments.
Authors: Sungtaek Son, Eardi Lila, Kwun Chuen Gary Chan
Abstract: Individualized treatment regimes (ITRs) aim to improve clinical outcomes by assigning treatment based on patient-specific characteristics. However, existing methods often struggle with high-dimensional covariates, limiting accuracy, interpretability, and real-world applicability. We propose a novel sufficient dimension reduction approach that directly targets the contrast between potential outcomes and identifies a low-dimensional subspace of the covariates capturing treatment effect heterogeneity. This reduced representation enables more accurate estimation of optimal ITRs through outcome-weighted learning. To accommodate observational data, our method incorporates kernel-based covariate balancing, allowing treatment assignment to depend on the full covariate set and avoiding the restrictive assumption that the subspace sufficient for modeling heterogeneous treatment effects is also sufficient for confounding adjustment. We show that the proposed method achieves universal consistency, i.e., its risk converges to the Bayes risk, under mild regularity conditions. We demonstrate its finite sample performance through simulations and an analysis of intensive care unit sepsis patient data to determine who should receive transthoracic echocardiography.
Authors: Zhongping Ji
Abstract: Modern computer vision architectures, from CNNs to Transformers, predominantly rely on the stacking of heuristic modules: spatial mixers (Attention/Conv) followed by channel mixers (FFNs). In this work, we challenge this paradigm by returning to mathematical first principles. We propose the \textbf{Clifford Algebra Network (CAN)}, also referred to as CliffordNet, a vision backbone grounded purely in Geometric Algebra. Instead of engineering separate modules for mixing and memory, we derive a unified interaction mechanism based on the \textbf{Clifford Geometric Product} ($uv = u \cdot v + u \wedge v$). This operation ensures algebraic completeness regarding the Geometric Product by simultaneously capturing feature coherence (via the generalized inner product) and structural variation (via the exterior wedge product). Implemented via an efficient sparse rolling mechanism with \textbf{strict linear complexity $\mathcal{O}(N)$}, our model reveals a surprising emergent property: the geometric interaction is so representationally dense that standard Feed-Forward Networks (FFNs) become redundant. Empirically, CliffordNet establishes a new Pareto frontier: our \textbf{Nano} variant achieves \textbf{76.41\%} accuracy on CIFAR-100 with only \textbf{1.4M} parameters, effectively matching the heavy-weight ResNet-18 (11.2M) with \textbf{$8\times$ fewer parameters}, while our \textbf{Base} variant sets a new SOTA for tiny models at \textbf{78.05\%}. Our results suggest that global understanding can emerge solely from rigorous, algebraically complete local interactions, potentially signaling a shift where \textit{geometry is all you need}. Code is available at https://github.com/ParaMind2025/CAN.
Authors: Shujian Gao, Yuan Wang, Jiangtao Yan, Zuxuan Wu, Yu-Gang Jiang
Abstract: Reinforcement Learning with Verifiable Rewards (RLVR) has significantly advanced reasoning capabilities in Large Language Models. However, adapting RLVR to multimodal domains suffers from a critical \textit{perception-reasoning decoupling}. Existing paradigms, driven by text-centric outcome rewards, reasoning in language medium, inadvertently encourage models to bypass visual perception. We empirically validate this through blind experiments: state-of-the-art policies maintain or surprisingly improve performance even when visual inputs are entirely removed. This reveals that these models degenerate into \textit{blind reasoners}, exploiting linguistic priors to generate plausible answers instead of attending to visual evidence. In response, we propose \textbf{Thinking with Deltas}, a framework driven by a \textbf{Differential Visual Reasoning Policy (DVRP)}. DVRP introduces intrinsic supervision via visual triplets, comprising original, masked, and perturbed inputs. It optimizes the model to maximize reasoning divergence from masked inputs (enforcing \textit{visual sensitivity}) while minimizing divergence from perturbed inputs (ensuring \textit{visual robustness}). By aligning reasoning variations strictly with the \textit{Delta} of visual information, DVRP inherently bolsters visual understanding capabilities and significantly outperforms state-of-the-art methods on both general and medical benchmarks, without requiring external annotations or auxiliary tools.
Authors: Yinan Hu, Estaban Tabak
Abstract: A novel framework for density estimation under expectation constraints is proposed. The framework minimizes the Wasserstein distance between the estimated density and a prior, subject to the constraints that the expected value of a set of functions adopts or exceeds given values. The framework is generalized to include regularization inequalities to mitigate the artifacts in the target measure. An annealing-like algorithm is developed to address non-smooth constraints, with its effectiveness demonstrated through both synthetic and proof-of-concept real world examples in finance.
Authors: Zabir Al Nazi, Shubhashis Roy Dipta, Sudipta Kar
Abstract: Chain-of-Thought (CoT) prompting is widely adopted for mathematical problem solving, including in low-resource languages, yet its behavior under irrelevant context remains underexplored. To systematically study this challenge, we introduce DISTRACTMATH-BN, a Bangla benchmark that augments MGSM and MSVAMP with semantically coherent but computationally irrelevant information. Evaluating seven models ranging from 3B to 12B parameters, we observe substantial performance degradation under distractors: standard models drop by up to 41 points, while reasoning-specialized models decline by 14 to 20 points despite consuming five times more tokens. We propose {\dag}DAGGER, which reformulates mathematical problem solving as executable computational graph generation with explicit modeling of distractor nodes. Fine-tuning Gemma-3 models using supervised fine-tuning followed by Group Relative Policy Optimization achieves comparable weighted accuracy on augmented benchmarks while using 89 percent fewer tokens than reasoning models. Importantly, this robustness emerges without explicit training on distractor-augmented examples. Our results suggest that enforcing structured intermediate representations improves robustness and inference efficiency in mathematical reasoning compared to free-form approaches, particularly in noisy, low-resource settings.
Authors: Qikai Xiao, Kehui Li, Binggui Zhou, Shaodan Ma
Abstract: Future wireless communication systems will increasingly rely on the integration of millimeter wave (mmWave) and sub-6 GHz bands to meet heterogeneous demands on high-speed data transmission and extensive coverage. To fully exploit the benefits of mmWave bands in massive multiple-input multiple-output (MIMO) systems, highly accurate channel state information (CSI) is required. However, directly estimating the mmWave channel demands substantial pilot overhead due to the large CSI dimension and low signal-to-noise ratio (SNR) led by severe path loss and blockage attenuation. In this paper, we propose an efficient \textbf{M}ulti-\textbf{D}omain \textbf{F}usion \textbf{C}hannel \textbf{E}xtrapolator (MDFCE) to extrapolate sub-6 GHz band CSI to mmWave band CSI, so as to reduce the pilot overhead for mmWave CSI acquisition in dual band massive MIMO systems. Unlike traditional channel extrapolation methods based on mathematical modeling, the proposed MDFCE combines the mixture-of-experts framework and the multi-head self-attention mechanism to fuse multi-domain features of sub-6 GHz CSI, aiming to characterize the mapping from sub-6 GHz CSI to mmWave CSI effectively and efficiently. The simulation results demonstrate that MDFCE can achieve superior performance with less training pilots compared with existing methods across various antenna array scales and signal-to-noise ratio levels while showing a much higher computational efficiency.
Authors: Michael Sidorov, Ofer Hadar
Abstract: The rapid growth of multimedia consumption, driven by major advances in mobile devices since the mid-2000s, has led to widespread use of video conferencing applications (VCAs) such as Zoom and Google Meet, as well as instant messaging applications (IMAs) like WhatsApp and Telegram, which increasingly support video conferencing as a core feature. Many of these systems rely on the Web Real-Time Communication (WebRTC) protocol, enabling direct peer-to-peer media streaming without requiring a third-party server to relay data, reducing the latency and facilitating a real-time communication. Despite WebRTC's potential, adverse network conditions can degrade streaming quality and consequently reduce users' Quality of Experience (QoE). Maintaining high QoE therefore requires continuous monitoring and timely intervention when QoE begins to deteriorate. While content providers can often estimate QoE by directly comparing transmitted and received media, this task is significantly more challenging for internet service providers (ISPs). End-to-end encryption, commonly used by modern VCAs and IMAs, prevent ISPs from accessing the original media stream, leaving only Quality of Service (QoS) and routing information available. To address this limitation, we propose the QoE Attention Convolutional Neural Network (qAttCNN), a model that leverages packet size parameter of the traffic to infer two no-reference QoE metrics viz. BRISQUE and frames per second (FPS). We evaluate qAttCNN on a custom dataset collected from WhatsApp video calls and compare it against existing QoE models. Using mean absolute error percentage (MAEP), our approach achieves 2.14% error for BRISQUE and 7.39% for FPS prediction.
Authors: Mustafa Abdool, Soumyadip Banerjee, Moutupsi Paul, Do-kyum Kim, Xioawei Liu, Bin Xu, Tracy Yu, Hui Gao, Karen Ouyang, Huiji Gao, Liwei He, Stephanie Moyerman, Sanjeev Katariya
Abstract: The goal of Airbnb search is to match guests with the ideal accommodation that fits their travel needs. This is a challenging problem, as popular search locations can have around a hundred thousand available homes, and guests themselves have a wide variety of preferences. Furthermore, the launch of new product features, such as \textit{flexible date search,} significantly increased the number of eligible homes per search query. As such, there is a need for a sophisticated retrieval system which can provide high-quality candidates with low latency in a way that integrates with the overall ranking stack. This paper details our journey to build an efficient and high-quality retrieval system for Airbnb search. We describe the key unique challenges we encountered when implementing an Embedding-Based Retrieval (EBR) system for a two sided marketplace like Airbnb -- such as the dynamic nature of the inventory, a lengthy user funnel with multiple stages, and a variety of product surfaces. We cover unique insights when modeling the retrieval problem, how to build robust evaluation systems, and design choices for online serving. The EBR system was launched to production and powers several use-cases such as regular search, flexible date and promotional emails for marketing campaigns. The system demonstrated statistically-significant improvements in key metrics, such as booking conversion, via A/B testing.
Authors: Masahiro Kaneko
Abstract: The use of large language models (LLMs) in peer review systems has attracted growing attention, making it essential to examine their potential vulnerabilities. Prior attacks rely on prompt injection, which alters manuscript content and conflates injection susceptibility with evaluation robustness. We propose the Paraphrasing Adversarial Attack (PAA), a black-box optimization method that searches for paraphrased sequences yielding higher review scores while preserving semantic equivalence and linguistic naturalness. PAA leverages in-context learning, using previous paraphrases and their scores to guide candidate generation. Experiments across five ML and NLP conferences with three LLM reviewers and five attacking models show that PAA consistently increases review scores without changing the paper's claims. Human evaluation confirms that generated paraphrases maintain meaning and naturalness. We also find that attacked papers exhibit increased perplexity in reviews, offering a potential detection signal, and that paraphrasing submissions can partially mitigate attacks.
Authors: Shaoning Sun, Mingzhu Cai, Huang He, Bingjin Chen, Siqi Bao, Yujiu Yang, Hua Wu, Haifeng Wang
Abstract: Language model families exhibit striking disparity in their capacity to benefit from reinforcement learning: under identical training, models like Qwen achieve substantial gains, while others like Llama yield limited improvements. Complementing data-centric approaches, we reveal that this disparity reflects a hidden structural property: \textbf{distributional clarity} in probability space. Through a three-stage analysis-from phenomenon to mechanism to interpretation-we uncover that RL-friendly models exhibit intra-class compactness and inter-class separation in their probability assignments to correct vs. incorrect responses. We quantify this clarity using the \textbf{Silhouette Coefficient} ($S$) and demonstrate that (1) high $S$ correlates strongly with RL performance; (2) low $S$ is associated with severe logic errors and reasoning instability. To confirm this property, we introduce a Silhouette-Aware Reweighting strategy that prioritizes low-$S$ samples during training. Experiments across six mathematical benchmarks show consistent improvements across all model families, with gains up to 5.9 points on AIME24. Our work establishes distributional clarity as a fundamental, trainable property underlying RL-Friendliness.
Authors: Fozle Rabbi Shafi, M. Anwar Hossain, Salimur Choudhury
Abstract: Large Language Model (LLM)-based systems increasingly rely on function calling to enable structured and controllable interaction with external data sources, yet existing datasets do not address mental health-oriented access to wearable sensor data. This paper presents a synthetic function-calling dataset designed for mental health assistance grounded in wearable health signals such as sleep, physical activity, cardiovascular measures, stress indicators, and metabolic data. The dataset maps diverse natural language queries to standardized API calls derived from a widely adopted health data schema. Each sample includes a user query, a query category, an explicit reasoning step, a normalized temporal parameter, and a target function. The dataset covers explicit, implicit, behavioral, symptom-based, and metaphorical expressions, which reflect realistic mental health-related user interactions. This resource supports research on intent grounding, temporal reasoning, and reliable function invocation in LLM-based mental health agents and is publicly released to promote reproducibility and future work.
Authors: Jie Wu, Haoling Li, Xin Zhang, Jiani Guo, Jane Luo, Steven Liu, Yangyu Huang, Ruihang Chu, Scarlett Li, Yujiu Yang
Abstract: Competitive programming presents great challenges for Code LLMs due to its intensive reasoning demands and high logical complexity. However, current Code LLMs still rely heavily on real-world data, which limits their scalability. In this paper, we explore a fully synthetic approach: training Code LLMs with entirely generated tasks, solutions, and test cases, to empower code reasoning models without relying on real-world data. To support this, we leverage feature-based synthesis to propose a novel data synthesis pipeline called SynthSmith. SynthSmith shows strong potential in producing diverse and challenging tasks, along with verified solutions and tests, supporting both supervised fine-tuning and reinforcement learning. Based on the proposed synthetic SFT and RL datasets, we introduce the X-Coder model series, which achieves a notable pass rate of 62.9 avg@8 on LiveCodeBench v5 and 55.8 on v6, outperforming DeepCoder-14B-Preview and AReal-boba2-14B despite having only 7B parameters. In-depth analysis reveals that scaling laws hold on our synthetic dataset, and we explore which dimensions are more effective to scale. We further provide insights into code-centric reinforcement learning and highlight the key factors that shape performance through detailed ablations and analysis. Our findings demonstrate that scaling high-quality synthetic data and adopting staged training can greatly advance code reasoning, while mitigating reliance on real-world coding data.
Authors: Taishi Watanabe, Ryo Karakida, Jun-nosuke Teramae
Abstract: The success of deep neural networks largely depends on the statistical structure of the training data. While learning dynamics and generalization on isotropic data are well-established, the impact of pronounced anisotropy on these crucial aspects is not yet fully understood. We examine the impact of data anisotropy, represented by a spiked covariance structure, a canonical yet tractable model, on the learning dynamics and generalization error of a two-layer linear network in a linear regression setting. Our analysis reveals that the learning dynamics proceed in two distinct phases, governed initially by the input-output correlation and subsequently by other principal directions of the data structure. Furthermore, we derive an analytical expression for the generalization error, quantifying how the alignment of the spike structure of the data with the learning task improves performance. Our findings offer deep theoretical insights into how data anisotropy shapes the learning trajectory and final performance, providing a foundation for understanding complex interactions in more advanced network architectures.
Authors: Qinshan Zhang, Bin Chen, Yong Jiang, Shu-Tao Xia
Abstract: Transformer channel decoders, such as the Error Correction Code Transformer (ECCT), have shown strong empirical performance in channel decoding, yet their generalization behavior remains theoretically unclear. This paper studies the generalization performance of ECCT from a learning-theoretic perspective. By establishing a connection between multiplicative noise estimation errors and bit-error-rate (BER), we derive an upper bound on the generalization gap via bit-wise Rademacher complexity. The resulting bound characterizes the dependence on code length, model parameters, and training set size, and applies to both single-layer and multi-layer ECCTs. We further show that parity-check-based masked attention induces sparsity that reduces the covering number, leading to a tighter generalization bound. To the best of our knowledge, this work provides the first theoretical generalization guarantees for this class of decoders.
Authors: Zhiyuan Tang, Wanning Chen, Kan Xu
Abstract: Matching markets face increasing needs to learn the matching qualities between demand and supply for effective design of matching policies. In practice, the matching rewards are high-dimensional due to the growing diversity of participants. We leverage a natural low-rank matrix structure of the matching rewards in these two-sided markets, and propose to utilize matrix completion to accelerate reward learning with limited offline data. A unique property for matrix completion in this setting is that the entries of the reward matrix are observed with matching interference -- i.e., the entries are not observed independently but dependently due to matching or budget constraints. Such matching dependence renders unique technical challenges, such as sub-optimality or inapplicability of the existing analytical tools in the matrix completion literature, since they typically rely on sample independence. In this paper, we first show that standard nuclear norm regularization remains theoretically effective under matching interference. We provide a near-optimal Frobenius norm guarantee in this setting, coupled with a new analytical technique. Next, to guide certain matching decisions, we develop a novel ``double-enhanced'' estimator, based off the nuclear norm estimator, with a near-optimal entry-wise guarantee. Our double-enhancement procedure can apply to broader sampling schemes even with dependence, which may be of independent interest. Additionally, we extend our approach to online learning settings with matching constraints such as optimal matching and stable matching, and present improved regret bounds in matrix dimensions. Finally, we demonstrate the practical value of our methods using both synthetic data and real data of labor markets.
Authors: Roman Hornung (Institute for Medical Information Processing, Biometry and Epidemiology, Faculty of Medicine, Ludwig Maximilian University of Munich, Munich Center for Machine Learning), Alexander Hapfelmeier (Institute of General Practice and Health Services Research, Department Clinical Medicine, TUM School of Medicine and Health, Technical University of Munich, Institute of AI and Informatics in Medicine, TUM School of Medicine and Health, Technical University of Munich)
Abstract: Random forests (RFs) are widely used for prediction and variable importance analysis and are often believed to capture any types of interactions via recursive splitting. However, since the splits are chosen locally, interactions are only reliably captured when at least one involved covariate has a marginal effect. We introduce unity forests (UFOs), an RF variant designed to better exploit interactions involving covariates without marginal effects. In UFOs, the first few splits of each tree are optimized jointly across a random covariate subset to form a "tree root" capturing such interactions; the remainder is grown conventionally. We further propose the unity variable importance measure (VIM), which is based on out-of-bag split criterion values from the tree roots. Here, only a small fraction of tree root splits with the highest in-bag criterion values are considered per covariate, reflecting that covariates with purely interaction-based effects are discriminative only if a split in an interacting covariate occurred earlier in the tree. Finally, we introduce covariate-representative tree roots (CRTRs), which select representative tree roots per covariate and provide interpretable insight into the conditions - marginal or interactive - under which each covariate has its strongest effects. In a simulation study, the unity VIM reliably identified interacting covariates without marginal effects, unlike conventional RF-based VIMs. In a large-scale real-data comparison, UFOs achieved higher discrimination and predictive accuracy than standard RFs, with comparable calibration. The CRTRs reproduced the covariates' true effect types reliably in simulated data and provided interesting insights in a real data analysis.
Authors: Luke S. Lagunowich, Guoxiang Grayson Tong, Daniele E. Schiavazzi
Abstract: Traditional filtering algorithms for state estimation -- such as classical Kalman filtering, unscented Kalman filtering, and particle filters - show performance degradation when applied to nonlinear systems whose uncertainty follows arbitrary non-Gaussian, and potentially multi-modal distributions. This study reviews recent approaches to state estimation via nonlinear filtering based on conditional normalizing flows, where the conditional embedding is generated by standard MLP architectures, transformers or selective state-space models (like Mamba-SSM). In addition, we test the effectiveness of an optimal-transport-inspired kinetic loss term in mitigating overparameterization in flows consisting of a large collection of transformations. We investigate the performance of these approaches on applications relevant to autonomous driving and patient population dynamics, paying special attention to how they handle time inversion and chained predictions. Finally, we assess the performance of various conditioning strategies for an application to real-world COVID-19 joint SIR system forecasting and parameter estimation.
Authors: Wang Yang, Debargha Ganguly, Xinpeng Li, Chaoda Song, Shouren Wang, Vikash Singh, Vipin Chaudhary, Xiaotian Han
Abstract: Hybrid reasoning language models are commonly controlled through high-level Think/No-think instructions to regulate reasoning behavior, yet we found that such mode switching is largely driven by a small set of trigger tokens rather than the instructions themselves. Through attention analysis and controlled prompting experiments, we show that a leading ``Okay'' token induces reasoning behavior, while the newline pattern following ``'' suppresses it. Based on this observation, we propose Mid-Think, a simple training-free prompting format that combines these triggers to achieve intermediate-budget reasoning, consistently outperforming fixed-token and prompt-based baselines in terms of the accuracy-length trade-off. Furthermore, applying Mid-Think to RL training after SFT reduces training time by approximately 15% while improving final performance of Qwen3-8B on AIME from 69.8% to 72.4% and on GPQA from 58.5% to 61.1%, demonstrating its effectiveness for both inference-time control and RL-based reasoning training.
Authors: Zhuoyi Yang, Yurun Song, Iftekhar Ahmed, Ian Harris
Abstract: Multi-hop question answering is widely used to evaluate the reasoning capabilities of large language models (LLMs), as it requires integrating multiple pieces of supporting knowledge to arrive at a correct answer. While prior work has explored different mechanisms for providing knowledge to LLMs, such as finetuning and retrieval-augmented generation (RAG), their relative effectiveness for multi-hop question answering remains insufficiently understood, particularly when the required knowledge is temporally novel. In this paper, we systematically compare parametric and non-parametric knowledge injection methods for open-domain multi-hop question answering. We evaluate unsupervised fine-tuning (continual pretraining), supervised fine-tuning, and retrieval-augmented generation across three 7B-parameter open-source LLMs. Experiments are conducted on two benchmarks: QASC, a standard multi-hop science question answering dataset, and a newly constructed dataset of over 10,000 multi-hop questions derived from Wikipedia events in 2024, designed to test knowledge beyond the models' pretraining cutoff. Our results show that unsupervised fine-tuning provides only limited gains over base models, suggesting that continual pretraining alone is insufficient for improving multi-hop reasoning accuracy. In contrast, retrieval-augmented generation yields substantial and consistent improvements, particularly when answering questions that rely on temporally novel information. Supervised fine-tuning achieves the highest overall accuracy across models and datasets. These findings highlight fundamental differences in how knowledge injection mechanisms support multi-hop question answering and underscore the importance of retrieval-based methods when external or compositional knowledge is required.
Authors: Alex Kokot, Anand Hemmady, Vydhourie Thiyageswaran, Marina Meila
Abstract: We introduce the setting of continuous index learning, in which a function of many variables varies only along a small number of directions at each point. For efficient estimation, it is beneficial for a learning algorithm to adapt, near each point $x$, to the subspace that captures the local variability of the function $f$. We pose this task as kernel adaptation along a manifold with noise, and introduce Local EGOP learning, a recursive algorithm that utilizes the Expected Gradient Outer Product (EGOP) quadratic form as both a metric and inverse-covariance of our target distribution. We prove that Local EGOP learning adapts to the regularity of the function of interest, showing that under a supervised noisy manifold hypothesis, intrinsic dimensional learning rates are achieved for arbitrarily high-dimensional noise. Empirically, we compare our algorithm to the feature learning capabilities of deep learning. Additionally, we demonstrate improved regression quality compared to two-layer neural networks in the continuous single-index setting.
Authors: Pedro Abdalla, Junren Chen
Abstract: We consider the problem of mean estimation under quantization and adversarial corruption. We construct multivariate robust estimators that are optimal up to logarithmic factors in two different settings. The first is a one-bit setting, where each bit depends only on a single sample, and the second is a partial quantization setting, in which the estimator may use a small fraction of unquantized data.
Authors: Osama Yousuf, Andreu L. Glasmann, Martin Lueker-Boden, Sina Najmaei, Gina C. Adam
Abstract: Emerging memory technologies have gained significant attention as a promising pathway to overcome the limitations of conventional computing architectures in deep learning applications. By enabling computation directly within memory, these technologies - built on nanoscale devices with tunable and nonvolatile conductance - offer the potential to drastically reduce energy consumption and latency compared to traditional von Neumann systems. This paper introduces XBTorch (short for CrossBarTorch), a novel simulation framework that integrates seamlessly with PyTorch and provides specialized tools for accurately and efficiently modeling crossbar-based systems based on emerging memory technologies. Through detailed comparisons and case studies involving hardware-aware training and inference, we demonstrate how XBTorch offers a unified interface for key research areas such as device-level modeling, cross-layer co-design, and inference-time fault tolerance. While exemplar studies utilize ferroelectric field-effect transistor (FeFET) models, the framework remains technology-agnostic - supporting other emerging memories such as resistive RAM (ReRAM), as well as enabling user-defined custom device models. The code is publicly available at: https://github.com/ADAM-Lab-GW/xbtorch
Authors: Jiguang Li, Hengrui Luo
Abstract: Bayesian optimization (BO) iteratively fits a Gaussian process (GP) surrogate to accumulated evaluations and selects new queries via an acquisition function such as expected improvement (EI). In practice, BO often concentrates evaluations near the current incumbent, causing the surrogate to become overconfident and to understate predictive uncertainty in the region guiding subsequent decisions. We develop a robust GP-based BO via tempered posterior updates, which downweight the likelihood by a power $\alpha \in (0,1]$ to mitigate overconfidence under local misspecification. We establish cumulative regret bounds for tempered BO under a family of generalized improvement rules, including EI, and show that tempering yields strictly sharper worst-case regret guarantees than the standard posterior $(\alpha=1)$, with the most favorable guarantees occurring near the classical EI choice. Motivated by our theoretic findings, we propose a prequential procedure for selecting $\alpha$ online: it decreases $\alpha$ when realized prediction errors exceed model-implied uncertainty and returns $\alpha$ toward one as calibration improves. Empirical results demonstrate that tempering provides a practical yet theoretically grounded tool for stabilizing BO surrogates under localized sampling.
Authors: Yixiao Peng, Hao Hu, Feiyang Li, Xinye Cao, Yingchang Jiang, Jipeng Tang, Guoshun Nan, Yuling Liu
Abstract: While virtualization and resource pooling empower cloud networks with structural flexibility and elastic scalability, they inevitably expand the attack surface and challenge cyber resilience. Reinforcement Learning (RL)-based defense strategies have been developed to optimize resource deployment and isolation policies under adversarial conditions, aiming to enhance system resilience by maintaining and restoring network availability. However, existing approaches lack robustness as they require retraining to adapt to dynamic changes in network structure, node scale, attack strategies, and attack intensity. Furthermore, the lack of Human-in-the-Loop (HITL) support limits interpretability and flexibility. To address these limitations, we propose CyberOps-Bots, a hierarchical multi-agent reinforcement learning framework empowered by Large Language Models (LLMs). Inspired by MITRE ATT&CK's Tactics-Techniques model, CyberOps-Bots features a two-layer architecture: (1) An upper-level LLM agent with four modules--ReAct planning, IPDRR-based perception, long-short term memory, and action/tool integration--performs global awareness, human intent recognition, and tactical planning; (2) Lower-level RL agents, developed via heterogeneous separated pre-training, execute atomic defense actions within localized network regions. This synergy preserves LLM adaptability and interpretability while ensuring reliable RL execution. Experiments on real cloud datasets show that, compared to state-of-the-art algorithms, CyberOps-Bots maintains network availability 68.5% higher and achieves a 34.7% jumpstart performance gain when shifting the scenarios without retraining. To our knowledge, this is the first study to establish a robust LLM-RL framework with HITL support for cloud defense. We will release our framework to the community, facilitating the advancement of robust and autonomous defense in cloud networks.
Authors: James Calo, Benny Lo
Abstract: Consensus mechanisms are the core of any blockchain system. However, the majority of these mechanisms do not target federated learning directly nor do they aid in the aggregation step. This paper introduces Proof of Reasoning (PoR), a novel consensus mechanism specifically designed for federated learning using blockchain, aimed at preserving data privacy, defending against malicious attacks, and enhancing the validation of participating networks. Unlike generic blockchain consensus mechanisms commonly found in the literature, PoR integrates three distinct processes tailored for federated learning. Firstly, a masked autoencoder (MAE) is trained to generate an encoder that functions as a feature map and obfuscates input data, rendering it resistant to human reconstruction and model inversion attacks. Secondly, a downstream classifier is trained at the edge, receiving input from the trained encoder. The downstream network's weights, a single encoded datapoint, the network's output and the ground truth are then added to a block for federated aggregation. Lastly, this data facilitates the aggregation of all participating networks, enabling more complex and verifiable aggregation methods than previously possible. This three-stage process results in more robust networks with significantly reduced computational complexity, maintaining high accuracy by training only the downstream classifier at the edge. PoR scales to large IoT networks with low latency and storage growth, and adapts to evolving data, regulations, and network conditions.
Authors: Linus Bleistein, Mathieu Dagr\'eou, Francisco Andrade, Thomas Boudou, Aur\'elien Bellet
Abstract: Ensuring fairness in matching algorithms is a key challenge in allocating scarce resources and positions. Focusing on Optimal Transport (OT), we introduce a novel notion of group fairness requiring that the probability of matching two individuals from any two given groups in the OT plan satisfies a predefined target. We first propose \texttt{FairSinkhorn}, a modified Sinkhorn algorithm to compute perfectly fair transport plans efficiently. Since exact fairness can significantly degrade matching quality in practice, we then develop two relaxation strategies. The first one involves solving a penalised OT problem, for which we derive novel finite-sample complexity guarantees. This result is of independent interest as it can be generalized to arbitrary convex penalties. Our second strategy leverages bilevel optimization to learn a ground cost that induces a fair OT solution, and we establish a bound guaranteeing that the learned cost yields fair matchings on unseen data. Finally, we present empirical results that illustrate the trade-offs between fairness and performance.
Authors: Xinzi Cao, Jianyang Zhai, Pengfei Li, Zhiheng Hu, Cen Yan, Bingxu Mu, Guanghuan Fang, Bin She, Jiayu Li, Yihan Su, Dongyang Tao, Xiansong Huang, Fan Xu, Feidiao Yang, Yao Lu, Chang-Dong Wang, Yutong Lu, Weicheng Xue, Bin Zhou, Yonghong Tian
Abstract: To meet the ever-increasing demand for computational efficiency, Neural Processing Units (NPUs) have become critical in modern AI infrastructure. However, unlocking their full potential requires developing high-performance compute kernels using vendor-specific Domain-Specific Languages (DSLs), a task that demands deep hardware expertise and is labor-intensive. While Large Language Models (LLMs) have shown promise in general code generation, they struggle with the strict constraints and scarcity of training data in the NPU domain. Our preliminary study reveals that state-of-the-art general-purpose LLMs fail to generate functional complex kernels for Ascend NPUs, yielding a near-zero success rate. To address these challenges, we propose AscendKernelGen, a generation-evaluation integrated framework for NPU kernel development. We introduce Ascend-CoT, a high-quality dataset incorporating chain-of-thought reasoning derived from real-world kernel implementations, and KernelGen-LM, a domain-adaptive model trained via supervised fine-tuning and reinforcement learning with execution feedback. Furthermore, we design NPUKernelBench, a comprehensive benchmark for assessing compilation, correctness, and performance across varying complexity levels. Experimental results demonstrate that our approach significantly bridges the gap between general LLMs and hardware-specific coding. Specifically, the compilation success rate on complex Level-2 kernels improves from 0% to 95.5% (Pass@10), while functional correctness achieves 64.3% compared to the baseline's complete failure. These results highlight the critical role of domain-specific reasoning and rigorous evaluation in automating accelerator-aware code generation.
Authors: E. R. Filimoshina, D. S. Shirokov
Abstract: This paper introduces Lie groups in degenerate geometric (Clifford) algebras that preserve four fundamental subspaces determined by the grade involution and reversion under the adjoint and twisted adjoint representations. We prove that these Lie groups can be equivalently defined using norm functions of multivectors applied in the theory of spin groups. We also study the corresponding Lie algebras. Some of these Lie groups and algebras are closely related to Heisenberg Lie groups and algebras. The introduced groups are interesting for various applications in physics and computer science, in particular, for constructing equivariant neural networks.
Authors: Yang Zhao, Yangou Ouyang, Xiao Ding, Hepeng Wang, Bibo Cai, Kai Xiong, Jinglong Gao, Zhouhao Sun, Li Du, Bing Qin, Ting Liu
Abstract: While Hybrid Supervised Fine-Tuning (SFT) followed by Reinforcement Learning (RL) has become the standard paradigm for training LLM agents, effective mechanisms for data allocation between these stages remain largely underexplored. Current data arbitration strategies often rely on surface-level heuristics that fail to diagnose intrinsic learning needs. Since SFT targets pattern consolidation through imitation while RL drives structural adaptation via exploration, misaligning data with these functional roles causes severe optimization interference. We propose PRISM, a dynamics-aware framework grounded in Schema Theory that arbitrates data based on its degree of cognitive conflict with the model's existing knowledge. By analyzing the spatial geometric structure of gradients, PRISM identifies data triggering high spatial concentration as high-conflict signals that require RL for structural restructuring. In contrast, data yielding diffuse updates is routed to SFT for efficient consolidation. Extensive experiments on WebShop and ALFWorld demonstrate that PRISM achieves a Pareto improvement, outperforming state-of-the-art hybrid methods while reducing computational costs by up to 3.22$\times$. Our findings suggest that disentangling data based on internal optimization regimes is crucial for scalable and robust agent alignment.
Authors: Pranav Kallem
Abstract: Large language models (LLMs) achieve strong aver- age performance yet remain unreliable at the instance level, with frequent hallucinations, brittle failures, and poorly calibrated confidence. We study reliability through the lens of multi-model consensus: given responses from several heterogeneous LLMs, can we learn which answer is most likely correct for a given query? We introduce a Multi-Model Consensus Reasoning Engine that treats the set of LLM outputs as input to a supervised meta-learner. The system maps natural language responses into structured features using semantic embeddings, pairwise similarity and clustering statistics, lexical and structural cues, reasoning-quality scores, confidence estimates, and model-specific priors, and then applies gradient-boosted trees, listwise ranking, and graph neural networks over similarity graphs of answers. Using three open-weight LLMs evaluated on compact, resource- constrained subsets of GSM8K, ARC-Challenge, HellaSwag, and TruthfulQA, our best graph-attention-based consensus model improves macro-average accuracy by 4.6 percentage points over the strongest single LLM and by 8.1 points over majority vote, while also yielding lower Brier scores and fewer TruthfulQA hal- lucinations. Ablation and feature-importance analyses show that semantic agreement and clustering features are most influential, with reasoning-quality and model-prior features providing com- plementary gains, suggesting supervised multi-model consensus is a practical route toward more reliable LLM behavior, even in a modest single-machine setup.
Authors: Yiran Jia
Abstract: Learning models that can handle distribution shifts is a key challenge in domain generalization. Invariance learning, an approach that focuses on identifying features invariant across environments, improves model generalization by capturing stable relationships, which may represent causal effects when the data distribution is encoded within a structural equation model (SEM) and satisfies modularity conditions. This has led to a growing body of work that builds on invariance learning, leveraging the inherent heterogeneity across environments to develop methods that provide causal explanations while enhancing robust prediction. However, in many practical scenarios, obtaining complete outcome data from each environment is challenging due to the high cost or complexity of data collection. This limitation in available data hinders the development of models that fully leverage environmental heterogeneity, making it crucial to address missing outcomes to improve both causal insights and robust prediction. In this work, we derive an estimator from the invariance objective under missing outcomes. We establish non-asymptotic guarantees on variable selection property and $\ell_2$ error convergence rates, which are influenced by the proportion of missing data and the quality of imputation models across environments. We evaluate the performance of the new estimator through extensive simulations and demonstrate its application using the UCI Bike Sharing dataset to predict the count of bike rentals. The results show that despite relying on a biased imputation model, the estimator is efficient and achieves lower prediction error, provided the bias is within a reasonable range.
Authors: Karthikeyan V. R., Premnath S., Kavinraaj S., J. Sangeetha
Abstract: Fraudulent activities on digital banking services are becoming more intricate by the day, challenging existing defenses. While older rule driven methods struggle to keep pace, even precision focused algorithms fall short when new scams are introduced. These tools typically overlook subtle shifts in criminal behavior, missing crucial signals. Because silent breaches cost institutions far more than flagged but legitimate actions, catching every possible case is crucial. High sensitivity to actual threats becomes essential when oversight leads to heavy losses. One key aim here involves reducing missed fraud cases without spiking incorrect alerts too much. This study builds a system using group learning methods adjusted through smart threshold choices. Using real world transaction records shared openly, where cheating acts rarely appear among normal activities, tests are run under practical skewed distributions. The outcomes reveal that approximately 91 percent of actual fraud is detected, outperforming standard setups that rely on unchanging rules when dealing with uneven examples across classes. When tested in live settings, the fraud detection system connects directly to an online banking transaction flow, stopping questionable activities before they are completed. Alongside this setup, a browser add on built for Chrome is designed to flag deceptive web links and reduce threats from harmful sites. These results show that adjusting decisions by cost impact and validating across entire systems makes deployment more stable and realistic for today's digital banking platforms.
Authors: Likun Zhang, Wei Ma
Abstract: Decision trees are powerful machine learning algorithms, widely used in fields such as economics and medicine for their simplicity and interpretability. However, decision trees such as CART are prone to overfitting, especially when grown deep or the sample size is small. Conventional methods to reduce overfitting include pre-pruning and post-pruning, which constrain the growth of uninformative branches. In this paper, we propose a complementary approach by introducing a covariance-driven splitting criterion for regression trees (CovRT). This method is more robust to overfitting than the empirical risk minimization criterion used in CART, as it produces more balanced and stable splits and more effectively identifies covariates with true signals. We establish an oracle inequality of CovRT and prove that its predictive accuracy is comparable to that of CART in high-dimensional settings. We find that CovRT achieves superior prediction accuracy compared to CART in both simulations and real-world tasks.
Authors: Zhuoka Feng, Kang Chen, Sihan Zhao, Kai Xiong, Yaoning Wang, Minshen Yu, Junjie Nian, Changyi Xiao, Yixin Cao, Yugang Jiang
Abstract: Interactive large language model agents have advanced rapidly, but most remain specialized to a single environment and fail to adapt robustly to other environments. Model merging offers a training-free alternative by integrating multiple experts into a single model. In this paper, we propose Agent-Role Merging (ARM), an activation-guided, role-conditioned neuron transplantation method for model merging in LLM agents. ARM improves existing merging methods from static natural language tasks to multi-turn agent scenarios, and over the generalization ability across various interactive environments. This is achieved with a well designed 3-step framework: 1) constructing merged backbones, 2) selection based on its role-conditioned activation analysis, and 3) neuron transplantation for fine-grained refinements. Without gradient-based optimization, ARM improves cross-benchmark generalization while enjoying efficiency. Across diverse domains, the model obtained via ARM merging outperforms prior model merging methods and domain-specific expert models, while demonstrating strong out-of-domain generalization.
Authors: EL Mahdi Khribch, Pierre Alquier
Abstract: The $\rho$-posterior framework provides universal Bayesian estimation with explicit contamination rates and optimal convergence guarantees, but has remained computationally difficult due to an optimization over reference distributions that precludes intractable posterior computation. We develop a PAC-Bayesian framework that recovers these theoretical guarantees through temperature-dependent Gibbs posteriors, deriving finite-sample oracle inequalities with explicit rates and introducing tractable variational approximations that inherit the robustness properties of exact $\rho$-posteriors. Numerical experiments demonstrate that this approach achieves theoretical contamination rates while remaining computationally feasible, providing the first practical implementation of $\rho$-posterior inference with rigorous finite-sample guarantees.
Authors: Huan Li, Yiming Dong, Zhouchen Lin
Abstract: This paper studies the AdamW-style Shampoo optimizer, an effective implementation of classical Shampoo that notably won the external tuning track of the AlgoPerf neural network training algorithm competition. Our analysis unifies one-sided and two-sided preconditioning and establishes the convergence rate $\frac{1}{K}\sum_{k=1}^K E\left[\|\nabla f(X_k)\|_*\right]\leq O(\frac{\sqrt{m+n}C}{K^{1/4}})$ measured by nuclear norm, where $K$ represents the iteration number, $(m,n)$ denotes the size of matrix parameters, and $C$ matches the constant in the optimal convergence rate of SGD. Theoretically, we have $\|\nabla f(X)\|_F\leq \|\nabla f(X)\|_*\leq \sqrt{m+n}\|\nabla f(X)\|_F$, supporting that our convergence rate can be considered to be analogous to the optimal $\frac{1}{K}\sum_{k=1}^KE\left[\|\nabla f(X_k)\|_F\right]\leq O(\frac{C}{K^{1/4}})$ convergence rate of SGD in the ideal case of $\|\nabla f(X)\|_*= \Theta(\sqrt{m+n})\|\nabla f(X)\|_F$.
Authors: Yuanhe Zhang, Jiayu Tian, Yibo Zhang, Shilinlu Yan, Liang Lin, Zhenhong Zhou, Li Sun, Sen Su
Abstract: Large Audio Language Models (LALMs) have been widely applied in real-time scenarios, such as in-car assistants and online meeting comprehension. In practice, audio inputs are often corrupted by device and environmental noise, leading to performance degradation. However, existing LALM studies on noise lack quantitative analysis and rely mainly on intuition and empirical observation, thus failing to understand practical robustness. To address this issue, we introduce Signal Embedding Energy (SEE), a method for quantifying the impact of noise intensity on LALM inputs, enabling the differentiation of LALM robustness in real-world deployments. SEE introduces a perspective based on structured activation subspaces derived from the model's internal representations, which more accurately captures its perception of noise than raw audio features. Across experiments, SEE exhibits a strong correlation with LALM performance, achieving a correlation of 0.98. Surprisingly, traditional audio denoising methods are only marginally effective for LALMs, and, in some cases, even increase SEE and impair performance. This suggests a mismatch between speech-centric denoising objectives and the noise sensitivity of modern LALMs. Therefore, we propose a mitigation strategy derived from SEE to denoise LALM inputs, outperforming existing denoising methods. This paper introduces a novel metric for noise quantification in LALMs, providing guidance for robustness improvements in real-world deployments.
Authors: Ziheng Li, Liu Kang, Feng Xiao, Luxi Xing, Qingyi Si, Zhuoran Li, Weikang Gong, Deqing Yang, Yanghua Xiao, Hongcheng Guo
Abstract: Group Relative Policy Optimization (GRPO) has emerged as a promising critic-free reinforcement learning paradigm for reasoning tasks. However, standard GRPO employs a coarse-grained credit assignment mechanism that propagates group-level rewards uniformly to to every token in a sequence, neglecting the varying contribution of individual reasoning steps. We address this limitation by introducing Outcome-grounded Advantage Reshaping (OAR), a fine-grained credit assignment mechanism that redistributes advantages based on how much each token influences the model's final answer. We instantiate OAR via two complementary strategies: (1) OAR-P, which estimates outcome sensitivity through counterfactual token perturbations, serving as a high-fidelity attribution signal; (2) OAR-G, which uses an input-gradient sensitivity proxy to approximate the influence signal with a single backward pass. These importance signals are integrated with a conservative Bi-Level advantage reshaping scheme that suppresses low-impact tokens and boosts pivotal ones while preserving the overall advantage mass. Empirical results on extensive mathematical reasoning benchmarks demonstrate that while OAR-P sets the performance upper bound, OAR-G achieves comparable gains with negligible computational overhead, both significantly outperforming a strong GRPO baseline, pushing the boundaries of critic-free LLM reasoning.
Authors: Niklas Kormann, Benjamin Doerr, Johannes F. Lutzeyer
Abstract: Over-smoothing and over-squashing have been extensively studied in the literature on Graph Neural Networks (GNNs) over the past years. We challenge this prevailing focus in GNN research, arguing that these phenomena are less critical for practical applications than assumed. We suggest that performance decreases often stem from uninformative receptive fields rather than over-smoothing. We support this position with extensive experiments on several standard benchmark datasets, demonstrating that accuracy and over-smoothing are mostly uncorrelated and that optimal model depths remain small even with mitigation techniques, thus highlighting the negligible role of over-smoothing. Similarly, we challenge that over-squashing is always detrimental in practical applications. Instead, we posit that the distribution of relevant information over the graph frequently factorises and is often localised within a small k-hop neighbourhood, questioning the necessity of jointly observing entire receptive fields or engaging in an extensive search for long-range interactions. The results of our experiments show that architectural interventions designed to mitigate over-squashing fail to yield significant performance gains. This position paper advocates for a paradigm shift in theoretical research, urging a diligent analysis of learning tasks and datasets using statistics that measure the underlying distribution of label-relevant information to better understand their localisation and factorisation.
Authors: Zicong Jiang, Magnus Karlsson, Erik Agrell, Christian H\"ager
Abstract: We propose physics-informed digital twin (PIDT): a fiber parameter estimation approach that combines a parameterized split-step method with a physics-informed loss. PIDT improves accuracy and convergence speed with lower complexity compared to previous neural operators.
Authors: Himanshu Patil, Geo Jolly, Ramana Raja Buddala, Ganesh Ramakrishnan, Rohit Saluja
Abstract: Video Question Answering (VideoQA) models enhance understanding and interaction with audiovisual content, making it more accessible, searchable, and useful for a wide range of fields such as education, surveillance, entertainment, and content creation. Due to heavy compute requirements, most large visual language models (VLMs) for VideoQA rely on a fixed number of frames by uniformly sampling the video. However, this process does not pick important frames or capture the context of the video. We present a novel query-based selection of frames relevant to the questions based on the submodular mutual Information (SMI) functions. By replacing uniform frame sampling with query-based selection, our method ensures that the chosen frames provide complementary and essential visual information for accurate VideoQA. We evaluate our approach on the MVBench dataset, which spans a diverse set of multi-action video tasks. VideoQA accuracy on this dataset was assessed using two VLMs, namely Video-LLaVA and LLaVA-NeXT, both of which originally employed uniform frame sampling. Experiments were conducted using both uniform and query-based sampling strategies. An accuracy improvement of up to \textbf{4\%} was observed when using query-based frame selection over uniform sampling. Qualitative analysis further highlights that query-based selection, using SMI functions, consistently picks frames better aligned with the question. We opine that such query-based frame selection can enhance accuracy in a wide range of tasks that rely on only a subset of video frames.
Authors: Sijia li, Xinran Li, Shibo Chen, Jun Zhang
Abstract: Offline multi-agent reinforcement learning (MARL) aims to solve cooperative decision-making problems in multi-agent systems using pre-collected datasets. Existing offline MARL methods primarily constrain training within the dataset distribution, resulting in overly conservative policies that struggle to generalize beyond the support of the data. While model-based approaches offer a promising solution by expanding the original dataset with synthetic data generated from a learned world model, the high dimensionality, non-stationarity, and complexity of multi-agent systems make it challenging to accurately estimate the transitions and reward functions in offline MARL. Given the difficulty of directly modeling joint dynamics, we propose a local-to-global (LOGO) world model, a novel framework that leverages local predictions-which are easier to estimate-to infer global state dynamics, thus improving prediction accuracy while implicitly capturing agent-wise dependencies. Using the trained world model, we generate synthetic data to augment the original dataset, expanding the effective state-action space. To ensure reliable policy learning, we further introduce an uncertainty-aware sampling mechanism that adaptively weights synthetic data by prediction uncertainty, reducing approximation error propagation to policies. In contrast to conventional ensemble-based methods, our approach requires only an additional encoder for uncertainty estimation, significantly reducing computational overhead while maintaining accuracy. Extensive experiments across 8 scenarios against 8 baselines demonstrate that our method surpasses state-of-the-art baselines on standard offline MARL benchmarks, establishing a new model-based baseline for generalizable offline multi-agent learning.
Authors: Helia Karisani, Mohammadreza Daneshvaramoli, Hedyeh Beyhaghi, Mohammad Hajiesmaili, Cameron Musco
Abstract: We study a learning-augmented variant of the secretary problem, recently introduced by Fujii and Yoshida (2023), in which the decision-maker has access to machine-learned predictions of candidate values. The central challenge is to balance consistency and robustness: when predictions are accurate, the algorithm should select a near-optimal secretary, while under inaccurate predictions it should still guarantee a bounded competitive ratio. We consider both the classical Random Order Secretary Problem (ROSP), where candidates arrive in a uniformly random order, and a more natural learning-augmented model in which the decision-maker may choose the arrival order based on predicted values. We call this model the Chosen Order Secretary Problem (COSP), capturing scenarios such as interview schedules set in advance. We propose a new randomized algorithm applicable to both ROSP and COSP. Our method switches from fully trusting predictions to a threshold-based rule once a large prediction deviation is detected. Let $\epsilon \in [0,1]$ denote the maximum multiplicative prediction error. For ROSP, our algorithm achieves a competitive ratio of $\max\{0.221, (1-\epsilon)/(1+\epsilon)\}$, improving upon the prior bound of $\max\{0.215, (1-\epsilon)/(1+\epsilon)\}$. For COSP, we achieve $\max\{0.262, (1-\epsilon)/(1+\epsilon)\}$, surpassing the $0.25$ worst-case bound for prior approaches and moving closer to the classical secretary benchmark of $1/e \approx 0.368$. These results highlight the benefit of combining predictions with arrival-order control in online decision-making.
Authors: Yongqi Li, Hao Lang, Tieyun Qian, Yongbin Li
Abstract: Vision-language models are increasingly employed as multimodal conversational agents (MCAs) for diverse conversational tasks. Recently, reinforcement learning (RL) has been widely explored for adapting MCAs to various human-AI interaction scenarios. Despite showing great enhancement in generalization performance, fine-tuning MCAs via RL still faces challenges in handling the extremely large text token space. To address this, we learn a compact latent action space for RL fine-tuning instead. Specifically, we adopt the learning from observation mechanism to construct the codebook for the latent action space, where future observations are leveraged to estimate current latent actions that could further be used to reconstruct future observations. However, the scarcity of paired image-text data hinders learning a codebook with sufficient coverage. Thus, we leverage both paired image-text data and text-only data to construct the latent action space, using a cross-modal projector for transforming text embeddings into image-text embeddings. We initialize the cross-modal projector on paired image-text data, and further train it on massive text-only data with a novel cycle consistency loss to enhance its robustness. We show that our latent action based method outperforms competitive baselines on two conversation tasks across various RL algorithms.
Authors: Victor Thuot (MISTEA), Sebastian Vogt (LRR-TUM), Debarghya Ghoshdastidar (LRR-TUM), Nicolas Verzelen (MISTEA)
Abstract: Clustering with bandit feedback refers to the problem of partitioning a set of items, where the clustering algorithm can sequentially query the items to receive noisy observations. The problem is formally posed as the task of partitioning the arms of an N-armed stochastic bandit according to their underlying distributions, grouping two arms together if and only if they share the same distribution, using samples collected sequentially and adaptively. This setting has gained attention in recent years due to its applicability in recommendation systems and crowdsourcing. Existing works on clustering with bandit feedback rely on a strong assumption that the underlying distributions are sub-Gaussian. As a consequence, the existing methods mainly cover settings with linearly-separable clusters, which has little practical relevance. We introduce a framework of ``nonparametric clustering with bandit feedback'', where the underlying arm distributions are not constrained to any parametric, and hence, it is applicable for active clustering of real-world datasets. We adopt a kernel-based approach, which allows us to reformulate the nonparametric problem as the task of clustering the arms according to their kernel mean embeddings in a reproducing kernel Hilbert space (RKHS). Building on this formulation, we introduce the KABC algorithm with theoretical correctness guarantees and analyze its sampling budget. We introduce a notion of signal-to-noise ratio for this problem that depends on the maximum mean discrepancy (MMD) between the arm distributions and on their variance in the RKHS. Our algorithm is adaptive to this unknown quantity: it does not require it as an input yet achieves instance-dependent guarantees.
Authors: Donglin Liu, Francisco Garc\'ia Atienza, Mengwu Guo
Abstract: Reduced-order modeling lies at the interface of numerical analysis and data-driven scientific computing, providing principled ways to compress high-fidelity simulations in science and engineering. We propose a training framework that couples a continuous-time form of operator inference with the adjoint-state method to obtain robust data-driven reduced-order models. This method minimizes a trajectory-based loss between reduced-order solutions and projected snapshot data, which removes the need to estimate time derivatives from noisy measurements and provides intrinsic temporal regularization through time integration. We derive the corresponding continuous adjoint equations to compute gradients efficiently and implement a gradient based optimizer to update the reduced model parameters. Each iteration only requires one forward reduced order solve and one adjoint solve, followed by inexpensive gradient assembly, making the method attractive for large-scale simulations. We validate the proposed method on three partial differential equations: viscous Burgers' equation, the two-dimensional Fisher-KPP equation, and an advection-diffusion equation. We perform systematic comparisons against standard operator inference under two perturbation regimes, namely reduced temporal snapshot density and additive Gaussian noise. For clean data, both approaches deliver similar accuracy, but in situations with sparse sampling and noise, the proposed adjoint-based training provides better accuracy and enhanced roll-out stability.
Authors: Sara Zoccheddu, Shah Rukh Qasim, Patrick Owen, Nicola Serra
Abstract: We study the use of large language models (LLMs) for physics instrument design and compare their performance to reinforcement learning (RL). Using only prompting, LLMs are given task constraints and summaries of prior high-scoring designs and propose complete detector configurations, which we evaluate with the same simulators and reward functions used in RL-based optimization. Although RL yields stronger final designs, we find that modern LLMs consistently generate valid, resource-aware, and physically meaningful configurations that draw on broad pretrained knowledge of detector design principles and particle--matter interactions, despite having no task-specific training. Based on this result, as a first step toward hybrid design workflows, we explore pairing the LLMs with a dedicated trust region optimizer, serving as a precursor to future pipelines in which LLMs propose and structure design hypotheses while RL performs reward-driven optimization. Based on these experiments, we argue that LLMs are well suited as meta-planners: they can design and orchestrate RL-based optimization studies, define search strategies, and coordinate multiple interacting components within a unified workflow. In doing so, they point toward automated, closed-loop instrument design in which much of the human effort required to structure and supervise optimization can be reduced.
Authors: Yunhao Fan, Sheng Zhang, Gia-Wei Chern
Abstract: Nonequilibrium electronic forces play a central role in voltage-driven phase transitions but are notoriously expensive to evaluate in dynamical simulations. Here we develop a machine learning framework for adiabatic lattice dynamics coupled to nonequilibrium electrons, and demonstrate it for a gating induced insulator to metal transition out of a charge density wave state in the Holstein model. Although exact electronic forces can be obtained from nonequilibrium Green's function (NEGF) calculations, their high computational cost renders long time dynamical simulations prohibitively expensive. By exploiting the locality of the electronic response, we train a neural network to directly predict instantaneous local electronic forces from the lattice configuration, thereby bypassing repeated NEGF calculations during time evolution. When combined with Brownian dynamics, the resulting machine learning force field quantitatively reproduces domain wall motion and nonequilibrium phase transition dynamics obtained from full NEGF simulations, while achieving orders of magnitude gains in computational efficiency. Our results establish direct force learning as an efficient and accurate approach for simulating nonequilibrium lattice dynamics in driven quantum materials.
Authors: O. Didkovskyi, A. Vidali, N. Jean, G. Le Pera
Abstract: This paper presents a meta-learning framework for credit risk assessment of Italian Small and Medium Enterprises (SMEs) that explicitly addresses the temporal misalignment of credit scoring models. The approach aligns financial statement reference dates with evaluation dates, mitigating bias arising from publication delays and asynchronous data sources. It is based on a two-step temporal decomposition that at first estimates annual probabilities of default (PDs) anchored to balance-sheet reference dates (December 31st) through a static model. Then it models the monthly evolution of PDs using higher-frequency behavioral data. Finally, we employ stacking-based architecture to aggregate multiple scoring systems, each capturing complementary aspects of default risk, into a unified predictive model. In this way, first level model outputs are treated as learned representations that encode non-linear relationships in financial and behavioral indicators, allowing integration of new expert-based features without retraining base models. This design provides a coherent and interpretable solution to challenges typical of low-default environments, including heterogeneous default definitions and reporting delays. Empirical validation shows that the framework effectively captures credit risk evolution over time, improving temporal consistency and predictive stability relative to standard ensemble methods.
Authors: Dimple Vijay Kochar, Nathaniel Pinckney, Guan-Ting Liu, Chia-Tung Ho, Chenhui Deng, Haoxing Ren, Brucek Khailany
Abstract: RTL design often relies heavily on ad-hoc testbench creation early in the design cycle. While large language models (LLMs) show promise for RTL code generation, their ability to reason about hardware specifications and generate targeted test plans remains largely unexplored. We present the first systematic study of LLM reasoning capabilities for RTL verification stimuli generation, establishing a two-stage framework that decomposes test plan generation from testbench execution. Our benchmark reveals that state-of-the-art models, including DeepSeek-R1 and Claude-4.0-Sonnet, achieve only 15.7-21.7% success rates on generating stimuli that pass golden RTL designs. To improve LLM generated stimuli, we develop a comprehensive training methodology combining supervised fine-tuning with a novel reinforcement learning approach, GRPO with State Mutation (GRPO-SMu), which enhances exploration by varying input mutations. Our approach leverages a tree-based branching mutation strategy to construct training data comprising equivalent and mutated trees, moving beyond linear mutation approaches to provide rich learning signals. Training on this curated dataset, our 7B parameter model achieves a 33.3% golden test pass rate and a 13.9% mutation detection rate, representing a 17.6% absolute improvement over baseline and outperforming much larger general-purpose models. These results demonstrate that specialized training methodologies can significantly enhance LLM reasoning capabilities for hardware verification tasks, establishing a foundation for automated sub-unit testing in semiconductor design workflows.
Authors: Denis D. Caprioti, Matheus Haas, Constantino F. Vasconcelos, Mauricio Girardi-Schappo
Abstract: The Hopfield model, originally inspired by spin-glass physics, occupies a central place at the intersection of statistical mechanics, neural networks, and modern artificial intelligence. Despite its conceptual simplicity and broad applicability -- from associative memory to near-optimal solutions of combinatorial optimization problems -- it is rarely integrated into standard undergraduate physics curricula. In this paper, we present the Hopfield model as a pedagogically rich framework that naturally unifies core topics from undergraduate statistical physics, dynamical systems, linear algebra, and computational methods. We provide a concise and illustrated theoretical introduction grounded in familiar physics concepts, analyze the model's energy function, dynamics, and pattern stability, and discuss practical aspects of simulation, including a freely available simulation code. To support instruction, we conclude with classroom-ready example problems designed to mirror research practice. By explicitly connecting fundamental physics to contemporary AI applications, this work aims to help prepare physics students to understand, apply, and critically engage with the computational tools increasingly central to research, industry, and society.
Authors: Benjamin Avanzi, Ronald Richman, Bernard Wong, Mario W\"uthrich, Yagebu Xie
Abstract: Outstanding claim liabilities are revised repeatedly as claims develop, yet most modern reserving models are trained as one-shot predictors and typically learn only from settled claims. We formulate individual claims reserving as a claim-level Markov decision process in which an agent sequentially updates outstanding claim liability (OCL) estimates over development, using continuous actions and a reward design that balances accuracy with stable reserve revisions. A key advantage of this reinforcement learning (RL) approach is that it can learn from all observed claim trajectories, including claims that remain open at valuation, thereby avoiding the reduced sample size and selection effects inherent in supervised methods trained on ultimate outcomes only. We also introduce practical components needed for actuarial use -- initialisation of new claims, temporally consistent tuning via a rolling-settlement scheme, and an importance-weighting mechanism to mitigate portfolio-level underestimation driven by the rarity of large claims. On CAS and SPLICE synthetic general insurance datasets, the proposed Soft Actor-Critic implementation delivers competitive claim-level accuracy and strong aggregate OCL performance, particularly for the immature claim segments that drive most of the liability.
Authors: Mahdi Nasiri, Johanna Kortelainen, Simo S\"arkk\"a
Abstract: This paper develops an approach for multi-step forecasting of dynamical systems by integrating probabilistic input forecasting with physics-informed output prediction. Accurate multi-step forecasting of time series systems is important for the automatic control and optimization of physical processes, enabling more precise decision-making. While mechanistic-based and data-driven machine learning (ML) approaches have been employed for time series forecasting, they face significant limitations. Incomplete knowledge of process mathematical models limits mechanistic-based direct employment, while purely data-driven ML models struggle with dynamic environments, leading to poor generalization. To address these limitations, this paper proposes a dual-level strategy for physics-informed forecasting of dynamical systems. On the first level, input variables are forecast using a hybrid method that integrates a long short-term memory (LSTM) network into probabilistic state transition models (STMs). On the second level, these stochastically predicted inputs are sequentially fed into a physics-informed neural network (PINN) to generate multi-step output predictions. The experimental results of the paper demonstrate that the hybrid input forecasting models achieve a higher log-likelihood and lower mean squared errors (MSE) compared to conventional STMs. Furthermore, the PINNs driven by the input forecasting models outperform their purely data-driven counterparts in terms of MSE and log-likelihood, exhibiting stronger generalization and forecasting performance across multiple test cases.
Authors: Jos\'e Pulido, Francesc Wilhelmi, Sergio Fortes, Alfonso Fern\'andez-Dur\'an, Lorenzo Galati Giordano, Raquel Barco
Abstract: Synthetic data generation is an appealing tool for augmenting and enriching datasets, playing a crucial role in advancing artificial intelligence (AI) and machine learning (ML). Not only does synthetic data help build robust AI/ML datasets cost-effectively, but it also offers privacy-friendly solutions and bypasses the complexities of storing large data volumes. This paper proposes a novel method to generate synthetic data, based on first-order auto-regressive noise statistics, for large-scale Wi-Fi deployments. The approach operates with minimal real data requirements while producing statistically rich traffic patterns that effectively mimic real Access Point (AP) behavior. Experimental results show that ML models trained on synthetic data achieve Mean Absolute Error (MAE) values within 10 to 15 of those obtained using real data when trained on the same APs, while requiring significantly less training data. Moreover, when generalization is required, synthetic-data-trained models improve prediction accuracy by up to 50 percent compared to real-data-trained baselines, thanks to the enhanced variability and diversity of the generated traces. Overall, the proposed method bridges the gap between synthetic data generation and practical Wi-Fi traffic forecasting, providing a scalable, efficient, and real-time solution for modern wireless networks.
Authors: Marc Lanctot, Kate Larson, Ian Gemp, Michael Kaisers
Abstract: As intelligent agents become more generally-capable, i.e. able to master a wide variety of tasks, the complexity and cost of properly evaluating them rises significantly. Tasks that assess specific capabilities of the agents can be correlated and stochastic, requiring many samples for accurate comparisons, leading to added costs. In this paper, we propose a formal definition and a conceptual framework for active evaluation of agents across multiple tasks, which assesses the performance of ranking algorithms as a function of number of evaluation data samples. Rather than curating, filtering, or compressing existing data sets as a preprocessing step, we propose an online framing: on every iteration, the ranking algorithm chooses the task and agents to sample scores from. Then, evaluation algorithms report a ranking of agents on each iteration and their performance is assessed with respect to the ground truth ranking over time. Several baselines are compared under different experimental contexts, with synthetic generated data and simulated online access to real evaluation data from Atari game-playing agents. We find that the classical Elo rating system -- while it suffers from well-known failure modes, in theory -- is a consistently reliable choice for efficient reduction of ranking error in practice. A recently-proposed method, Soft Condorcet Optimization, shows comparable performance to Elo on synthetic data and significantly outperforms Elo on real Atari agent evaluation. When task variation from the ground truth is high, selecting tasks based on proportional representation leads to higher rate of ranking error reduction.
Authors: Andrea Martin, Giuseppe Belgioioso
Abstract: We introduce a principled learning to optimize (L2O) framework for solving fixed-point problems involving general nonexpansive mappings. Our idea is to deliberately inject summable perturbations into a standard Krasnosel'skii-Mann iteration to improve its average-case performance over a specific distribution of problems while retaining its convergence guarantees. Under a metric sub-regularity assumption, we prove that the proposed parametrization includes only iterations that locally achieve linear convergence-up to a vanishing bias term-and that it encompasses all iterations that do so at a sufficiently fast rate. We then demonstrate how our framework can be used to augment several widely-used operator splitting methods to accelerate the solution of structured monotone inclusion problems, and validate our approach on a best approximation problem using an L2O-augmented Douglas-Rachford splitting algorithm.
Authors: Rei Taniguchi, Yuyang Dong, Makoto Onizuka, Chuan Xiao
Abstract: Due to the prevalence of large language models (LLMs), key-value (KV) cache reduction for LLM inference has received remarkable attention. Among numerous works that have been proposed in recent years, layer-wise token pruning approaches, which select a subset of tokens at particular layers to retain in KV cache and prune others, are one of the most popular schemes. They primarily adopt a set of pre-defined layers, at which tokens are selected. Such design is inflexible in the sense that the accuracy significantly varies across tasks and deteriorates in harder tasks such as KV retrieval. In this paper, we propose ASL, a training-free method that adaptively chooses the selection layer for KV cache reduction, exploiting the variance of token ranks ordered by attention score. The proposed method balances the performance across different tasks while meeting the user-specified KV budget requirement. ASL operates during the prefilling stage and can be jointly used with existing KV cache reduction methods such as SnapKV to optimize the decoding stage. By evaluations on the InfiniteBench, RULER, and NIAH benchmarks, we show that equipped with one-shot token selection, where tokens are selected at a layer and propagated to deeper layers, ASL outperforms state-of-the-art layer-wise token selection methods in accuracy while maintaining decoding speed and KV cache reduction.
Authors: Xingran Chen, Parimal Parag, Rohit Bhagat, Salim El Rouayheb
Abstract: Random walk (RW)-based algorithms have long been popular in distributed systems due to low overheads and scalability, with recent growing applications in decentralized learning. However, their reliance on local interactions makes them inherently vulnerable to malicious behavior. In this work, we investigate an adversarial threat that we term the ``Pac-Man'' attack, in which a malicious node probabilistically terminates any RW that visits it. This stealthy behavior gradually eliminates active RWs from the network, effectively halting the learning process without triggering failure alarms. To counter this threat, we propose the CREATE-IF-LATE (CIL) algorithm, which is a fully decentralized, resilient mechanism that enables self-creating RWs and prevents RW extinction in the presence of Pac-Man. Our theoretical analysis shows that the CIL algorithm guarantees several desirable properties, such as (i) non-extinction of the RW population, (ii) almost sure boundedness of the RW population, and (iii) convergence of RW-based stochastic gradient descent even in the presence of Pac-Man with a quantifiable deviation from the true optimum. Moreover, the learning process experiences at most a linear time delay due to Pac-Man interruptions and RW regeneration. Our extensive empirical results on both synthetic and public benchmark datasets validate our theoretical findings.
Authors: Efstratios Manolakis, Christian Bongiorno, Rosario Nunzio Mantegna
Abstract: A new wave of work on covariance cleaning and nonlinear shrinkage has delivered asymptotically optimal analytical solutions for large covariance matrices. Building on this progress, these ideas have been generalized to empirical cross-covariance matrices, whose singular-value shrinkage characterizes comovements between one set of assets and another. Existing analytical cross-covariance cleaners are derived under strong stationarity and large-sample assumptions, and they typically rely on mesoscopic regularity conditions such as bounded spectra; macroscopic common modes (e.g., a global market factor) violate these conditions. When applied to real equity returns, where dependence structures drift over time and global modes are prominent, we find that these theoretically optimal formulas do not translate into robust out-of-sample performance. We address this gap by designing a random-matrix-inspired neural architecture that operates in the empirical singular-vector basis and learns a nonlinear mapping from empirical singular values to their corresponding cleaned values. By construction, the network can recover the analytical solution as a special case, yet it remains flexible enough to adapt to non-stationary dynamics and mode-driven distortions. Trained on a long history of equity returns, the proposed method achieves a more favorable bias-variance trade-off than purely analytical cleaners and delivers systematically lower out-of-sample cross-covariance prediction errors. Our results demonstrate that combining random-matrix theory with machine learning makes asymptotic theories practically effective in realistic time-varying markets.
Authors: Jonathan D. Socha, Seyed F. Maroufi, Dipankar Biswas, Richard Um, Aruna S. Rao, Mark G. Luciano
Abstract: We present a novel framework for analyzing intracranial pressure monitoring data by applying interpretability principles. Intracranial pressure monitoring data was collected from 60 patients at Johns Hopkins. The data was segmented into individual cardiac cycles. A convolutional neural network was trained to classify each cardiac cycle into one of seven body positions. Neural network attention was extracted and was used to identify regions of interest in the waveform. Further directions for exploration are identified. This framework provides an extensible method to further understand the physiological and clinical underpinnings of the intracranial pressure waveform, which could lead to better diagnostic capabilities for intracranial pressure monitoring.
Authors: Jakob Paul Zimmermann, Georg Loho
Abstract: It has been demonstrated in various contexts that monotonicity leads to better explainability in neural networks. However, not every function can be well approximated by a monotone neural network. We demonstrate that monotonicity can still be used in two ways to boost explainability. First, we use an adaptation of the decomposition of a trained ReLU network into two monotone and convex parts, thereby overcoming numerical obstacles from an inherent blowup of the weights in this procedure. Our proposed saliency methods -- SplitCAM and SplitLRP -- improve on state of the art results on both VGG16 and Resnet18 networks on ImageNet-S across all Quantus saliency metric categories. Second, we exhibit that training a model as the difference between two monotone neural networks results in a system with strong self-explainability properties.
Authors: Joseph L. Shomberg
Abstract: We present a physics-informed Wasserstein GAN with gradient penalty (WGAN-GP) for solving the inverse Chafee--Infante problem on two-dimensional domains with Dirichlet boundary conditions. The objective is to reconstruct an unknown initial condition from a near-equilibrium state obtained after 100 explicit forward Euler iterations of the reaction-diffusion equation \[ u_t - \gamma\Delta u + \kappa\left(u^3 - u\right)=0. \] Because this mapping strongly damps high-frequency content, the inverse problem is severely ill-posed and sensitive to noise. Our approach integrates a U-Net generator, a PatchGAN critic with spectral normalization, Wasserstein loss with gradient penalty, and several physics-informed auxiliary terms, including Lyapunov energy matching, distributional statistics, and a crucial forward-simulation penalty. This penalty enforces consistency between the predicted initial condition and its forward evolution under the \emph{same} forward Euler discretization used for dataset generation. Earlier experiments employing an Eyre-type semi-implicit solver were not compatible with this residual mechanism due to the cost and instability of Newton iterations within batched GPU training. On a dataset of 50k training and 10k testing pairs on $128\times128$ grids (with natural $[-1,1]$ amplitude scaling), the best trained model attains a mean absolute error (MAE) of approximately \textbf{0.23988159} on the full test set, with a sample-wise standard deviation of about \textbf{0.00266345}. The results demonstrate stable inversion, accurate recovery of interfacial structure, and robustness to high-frequency noise in the initial data.
Authors: Teddy Koker, Abhijeet Gangan, Mit Kotak, Jaime Marian, Tess Smidt
Abstract: Many materials properties depend on higher-order derivatives of the potential energy surface, yet machine learned interatomic potentials (MLIPs) trained with standard a standard loss on energy, force, and stress errors can exhibit error in curvature, degrading the prediction of vibrational properties. We introduce phonon fine-tuning (PFT), which directly supervises second-order force constants of materials by matching MLIP energy Hessians to DFT-computed force constants from finite displacement phonon calculations. To scale to large supercells, PFT stochastically samples Hessian columns and computes the loss with a single Hessian-vector product. We also use a simple co-training scheme to incorporate upstream data to mitigate catastrophic forgetting. On the MDR Phonon benchmark, PFT improves Nequix MP (trained on Materials Project) by 55% on average across phonon thermodynamic properties and achieves state-of-the-art performance among models trained on Materials Project trajectories. PFT also generalizes to improve properties beyond second-derivatives, improving thermal conductivity predictions that rely on third-order derivatives of the potential energy.
Authors: Masahiro Kato
Abstract: Estimating the Riesz representer is a central problem in debiased machine learning for causal and structural parameter estimation. Various methods for Riesz representer estimation have been proposed, including Riesz regression and covariate balancing. This study unifies these methods within a single framework. Our framework fits a Riesz representer model to the true Riesz representer under a Bregman divergence, which includes the squared loss and the Kullback--Leibler (KL) divergence as special cases. We show that the squared loss corresponds to Riesz regression, and the KL divergence corresponds to tailored loss minimization, where the dual solutions correspond to stable balancing weights and entropy balancing weights, respectively, under specific model specifications. We refer to our method as generalized Riesz regression, and we refer to the associated duality as automatic covariate balancing. Our framework also generalizes density ratio fitting under a Bregman divergence to Riesz representer estimation, and it includes various applications beyond density ratio estimation. We also provide a convergence analysis for both cases where the model class is a reproducing kernel Hilbert space (RKHS) and where it is a neural network.
Authors: Johannes Erdmann, Nitish Kumar Kasaraguppe, Florian Mausolf
Abstract: Categorizing events using discriminant observables is central to many high-energy physics analyses. Yet, bin boundaries are often chosen by hand. A simple, popular choice is to apply argmax projections of multi-class scores and equidistant binning of one-dimensional discriminants. We propose a binning optimization for signal significance directly in multi-dimensional discriminants. We use a Gaussian Mixture Model (GMM) to define flexible bin boundary shapes for multi-class scores, while in one dimension (binary classification) we move bin boundaries directly. On this binning model, we study two optimization strategies: a differentiable and a Bayesian optimization approach. We study two toy setups: a binary classification and a three-class problem with two signals and backgrounds. In the one-dimensional case, both approaches achieve similar gains in signal sensitivity compared to equidistant binnings for a given number of bins. In the multi-dimensional case, the GMM-based binning defines sensitive categories as well, with the differentiable approach performing best. We show that, in particular for limited separability of the signal processes, our approach outperforms argmax classification even with optimized binning in the one-dimensional projections. Both methods are released as lightweight Python plugins intended for straightforward integration into existing analyses.
Authors: Ahmed Sabir, Markus K\"angsepp, Rajesh Sharma
Abstract: The increased use of Large Language Models (LLMs) in sensitive domains leads to growing interest in how their confidence scores correspond to fairness and bias. This study examines the alignment between LLM-predicted confidence and human-annotated bias judgments. Focusing on gender bias, the research investigates probability confidence calibration in contexts involving gendered pronoun resolution. The goal is to evaluate if calibration metrics based on predicted confidence scores effectively capture fairness-related disparities in LLMs. The results show that, among the six state-of-the-art models, Gemma-2 demonstrates the worst calibration according to the gender bias benchmark. The primary contribution of this work is a fairness-aware evaluation of LLMs' confidence calibration, offering guidance for ethical deployment. In addition, we introduce a new calibration metric, Gender-ECE, designed to measure gender disparities in resolution tasks.
Authors: Huanyu Li, Kun Lei, Sheng Zang, Kaizhe Hu, Yongyuan Liang, Bo An, Xiaoli Li, Huazhe Xu
Abstract: Post-training algorithms based on deep reinforcement learning can push the limits of robotic models for specific objectives, such as generalizability, accuracy, and robustness. However, Intervention-requiring Failures (IR Failures) (e.g., a robot spilling water or breaking fragile glass) during real-world exploration happen inevitably, hindering the practical deployment of such a paradigm. To tackle this, we introduce Failure-Aware Offline-to-Online Reinforcement Learning (FARL), a new paradigm minimizing failures during real-world reinforcement learning. We create FailureBench, a benchmark that incorporates common failure scenarios requiring human intervention, and propose an algorithm that integrates a world-model-based safety critic and a recovery policy trained offline to prevent failures during online exploration. Extensive simulation and real-world experiments demonstrate the effectiveness of FARL in significantly reducing IR Failures while improving performance and generalization during online reinforcement learning post-training. FARL reduces IR Failures by 73.1% while elevating performance by 11.3% on average during real-world RL post-training. Videos and code are available at https://failure-aware-rl.github.io.
Authors: Samuel Duffield
Abstract: We show that any stochastic differential equation with prescribed time-dependent marginal distributions admits a decomposition into three components: a unique scalar field governing marginal evolution, a symmetric positive-semidefinite diffusion matrix field and a skew-symmetric matrix field.
Authors: Siddharth Chandak, Vivek S. Borkar
Abstract: We derive uniform all-time concentration bound of the type 'for all $n \geq n_0$ for some $n_0$' for TD(0) with linear function approximation. We work with online TD learning with samples from a single sample path of the underlying Markov chain. This makes our analysis significantly different from offline TD learning or TD learning with access to independent samples from the stationary distribution of the Markov chain. We treat TD(0) as a contractive stochastic approximation algorithm, with both martingale and Markov noises. Markov noise is handled using the Poisson equation and the lack of almost sure guarantees on boundedness of iterates is handled using the concept of relaxed concentration inequalities.
Authors: Yilei Chen, Aldo Pacchiano, Ioannis Ch. Paschalidis
Abstract: We study the multiple-policy evaluation problem where we are given a set of $K$ policies and the goal is to evaluate their performance (expected total reward over a fixed horizon) to an accuracy $\epsilon$ with probability at least $1-\delta$. We propose an algorithm named $\mathrm{CAESAR}$ for this problem. Our approach is based on computing an approximate optimal offline sampling distribution and using the data sampled from it to perform the simultaneous estimation of the policy values. $\mathrm{CAESAR}$ has two phases. In the first we produce coarse estimates of the visitation distributions of the target policies at a low order sample complexity rate that scales with $\tilde{O}(\frac{1}{\epsilon})$. In the second phase, we approximate the optimal offline sampling distribution and compute the importance weighting ratios for all target policies by minimizing a step-wise quadratic loss function inspired by the DualDICE \cite{nachum2019dualdice} objective. Up to low order and logarithmic terms $\mathrm{CAESAR}$ achieves a sample complexity $\tilde{O}\left(\frac{H^4}{\epsilon^2}\sum_{h=1}^H\max_{k\in[K]}\sum_{s,a}\frac{(d_h^{\pi^k}(s,a))^2}{\mu^*_h(s,a)}\right)$, where $d^{\pi}$ is the visitation distribution of policy $\pi$, $\mu^*$ is the optimal sampling distribution, and $H$ is the horizon.
Authors: Elynn Chen, Xi Chen, Wenbo Jing
Abstract: In data-driven decision-making in marketing, healthcare, and education, it is desirable to utilize a large amount of data from existing ventures to navigate high-dimensional feature spaces and address data scarcity in new ventures. We explore knowledge transfer in dynamic decision-making by concentrating on batch stationary environments and formally defining task discrepancies through the lens of Markov decision processes (MDPs). We propose a framework of Transferred Fitted $Q$-Iteration algorithm with general function approximation, enabling the direct estimation of the optimal action-state function $Q^*$ using both target and source data. We establish the relationship between statistical performance and MDP task discrepancy under sieve approximation, shedding light on the impact of source and target sample sizes and task discrepancy on the effectiveness of knowledge transfer. We show that the final learning error of the $Q^*$ function is significantly improved from the single task rate both theoretically and empirically.
Authors: Hyunjun Na, Donghwan Lee
Abstract: $Q$-learning is one of the most fundamental reinforcement learning (RL) algorithms. Despite its widespread success in various applications, it is prone to overestimation bias in the $Q$-learning update. To address this issue, double $Q$-learning employs two independent $Q$-estimators which are randomly selected and updated during the learning process. This paper proposes a modified double $Q$-learning, called simultaneous double $Q$-learning (SDQ), with its finite-time analysis. SDQ eliminates the need for random selection between the two $Q$-estimators, and this modification allows us to analyze double $Q$-learning through the lens of a novel switching system framework facilitating efficient finite-time analysis. Empirical studies demonstrate that SDQ converges faster than double $Q$-learning while retaining the ability to mitigate the maximization bias. Finally, we derive a finite-time expected error bound for SDQ.
Authors: Seongho Son, William Bankes, Sayak Ray Chowdhury, Brooks Paige, Ilija Bogunovic
Abstract: Current Large Language Model (LLM) preference optimization algorithms do not account for temporal preference drift, which can lead to severe misalignment. To address this limitation, we propose Non-Stationary Direct Preference Optimisation (NS-DPO) that models time-dependent reward functions with a Dynamic Bradley-Terry model. NS-DPO proposes a computationally efficient solution by introducing only a single discount parameter in the loss function, which is used for exponential weighting that proportionally focuses learning on more time-relevant datapoints. We theoretically analyze the convergence of NS-DPO in a general setting where the exact nature of the preference drift is not known, providing upper bounds on the estimation error and regret caused by non-stationary preferences. Finally, we demonstrate the effectiveness of NS-DPO for fine-tuning LLMs under drifting preferences. Using scenarios where various levels of preference drift is introduced, with popular LLM reward models and datasets, we show that NS-DPO fine-tuned LLMs remain robust under non-stationarity, significantly outperforming baseline algorithms that ignore temporal preference changes, without sacrificing performance in stationary cases.
Authors: Jinying Xiao, Ping Li, Jie Nie, Bin Ji, Shasha Li, Xiaodong Liu, Jun Ma, Qingbo Wu, Jie Yu
Abstract: Recently, large language and vision models have shown strong performance, but due to high pre-training and fine-tuning costs, research has shifted towards faster training via dataset pruning. Previous methods used sample loss as an evaluation criterion, aiming to select the most "difficult" samples for training. However, when the pruning rate increases, the number of times each sample is trained becomes more evenly distributed, which causes many critical or general samples to not be effectively fitted. We refer to this as Low-Frequency Learning (LFL). In other words, LFL prevents the model from remembering most samples. In our work, we decompose the scoring function of LFL, provide a theoretical explanation for the inefficiency of LFL, and propose adding a memory term to the scoring function to enhance the model's memory capability, along with an approximation of this memory term. Similarly, we explore memory in Self-Supervised Learning (SSL), marking the first discussion on SSL memory. Using contrastive learning, we derive the memory term both theoretically and experimentally. Finally, we propose Enhance Memory Pruning (EMP), which addresses the issue of insufficient memory under high pruning rates by enhancing the model's memory of data, thereby improving its performance. We evaluated the performance of EMP in tasks such as image classification, natural language understanding, and model pre-training. The results show that EMP can improve model performance under extreme pruning rates. For example, in the CIFAR100-ResNet50 pre-training task, with 70\% pruning, EMP outperforms current methods by 2.2\%.
Authors: Mohammad Pivezhandi, Abusayeed Saifullah
Abstract: Reinforcement learning (RL) in few-shot scenarios with limited sensor data is challenging due to insufficient training samples, particularly in applications like Dynamic Voltage and Frequency Scaling (DVFS) where sensor readings are semi-structured with inherent correlations. We propose Flow-Augmented Reinforcement Learning (FlowRL), a novel method that leverages continuous normalizing flows to generate high-quality synthetic data for few-shot RL. By integrating latent space bootstrapping for diversity and feature-weighted flow matching to preserve critical data correlations, FlowRL enhances sample efficiency and policy robustness. Evaluated on a DVFS case study using the NVIDIA Jetson TX2, our approach achieves up to 35\% higher frame rates and faster Q-value convergence compared to baselines, demonstrating its effectiveness in resource-constrained environments. FlowRL generalizes to other semi-structured domains, such as robotics and smart grids, offering a scalable solution for data-scarce RL settings.
Authors: Florian Vincent, Wa\"iss Azizian, Franck Iutzeler, J\'er\^ome Malick
Abstract: We present skwdro, a Python library for training robust machine learning models. The library is based on distributionally robust optimization using Wasserstein distances, popular in optimal transport and machine learnings. The goal of the library is to make the training of robust models easier for a wide audience by proposing a wrapper for PyTorch modules, enabling model loss' robustification with minimal code changes. It comes along with scikit-learn compatible estimators for some popular objectives. The core of the implementation relies on an entropic smoothing of the original robust objective, in order to ensure maximal model flexibility. The library is available at https://github.com/iutzeler/skwdro and the documentation at https://skwdro.readthedocs.io.
URLs: https://github.com/iutzeler/skwdro, https://skwdro.readthedocs.io.
Authors: Yonatan Kurniawan (Brigham Young University, Provo, UT, USA), Tracianne B. Neilsen (Brigham Young University, Provo, UT, USA), Benjamin L. Francis (Achilles Heel Technologies, Orem, UT, USA), Alex M. Stankovic (SLAC National Accelerator Laboratory, Menlo Park, CA, USA), Mingjian Wen (University of Electronic Science and Technology of China, Chengdu, China), Ilia Nikiforov (University of Minnesota, Minneapolis, MN, USA), Ellad B. Tadmor (University of Minnesota, Minneapolis, MN, USA), Vasily V. Bulatov (Lawrence Livermore National Laboratory), Vincenzo Lordi (Lawrence Livermore National Laboratory), Mark K. Transtrum (Brigham Young University, Provo, UT, USA, SLAC National Accelerator Laboratory, Menlo Park, CA, USA)
Abstract: The efficacy of mathematical models heavily depends on the quality of the training data, yet collecting sufficient data is often expensive and challenging. Many modeling applications require inferring parameters only as a means to predict other quantities of interest (QoI). Because models often contain many unidentifiable (sloppy) parameters, QoIs often depend on a relatively small number of parameter combinations. Therefore, we introduce an information-matching criterion based on the Fisher Information Matrix to select the most informative training data from a candidate pool. This method ensures that the selected data contain sufficient information to learn only those parameters that are needed to constrain downstream QoIs. It is formulated as a convex optimization problem, making it scalable to large models and datasets. We demonstrate the effectiveness of this approach across various modeling problems in diverse scientific fields, including power systems and underwater acoustics. Finally, we use information-matching as a query function within an Active Learning loop for material science applications. In all these applications, we find that a relatively small set of optimal training data can provide the necessary information for achieving precise predictions. These results are encouraging for diverse future applications, particularly active learning in large machine learning models.
Authors: Chenxi Yang, Divyanshu Saxena, Rohit Dwivedula, Kshiteej Mahajan, Swarat Chaudhuri, Aditya Akella
Abstract: Learning-based congestion controllers offer better adaptability compared to traditional heuristics. However, the unreliability of learning techniques can cause learning-based controllers to behave poorly, creating a need for formal guarantees. While methods for formally verifying learned congestion controllers exist, these methods offer binary feedback that cannot optimize the controller toward better behavior. We improve this state-of-the-art via Canopy, a new property-driven framework that integrates learning with formal reasoning in the learning loop. Canopy uses novel quantitative certification with an abstract interpreter to guide the training process, rewarding models, and evaluating robust and safe model performance on worst-case inputs. Our evaluation demonstrates that unlike state-of-the-art learned controllers, Canopy-trained controllers provide both adaptability and worst-case reliability across a range of network conditions.
Authors: Zihang Li, Yangdong Ruan, Wenjun Liu, Zhengyang Wang, Tong Yang
Abstract: Although retrieval-augmented generation(RAG) significantly improves generation quality by retrieving external knowledge bases and integrating generated content, it faces computational efficiency bottlenecks, particularly in knowledge retrieval tasks involving hierarchical structures for Tree-RAG. This paper proposes a Tree-RAG acceleration method based on the improved Cuckoo Filter, which optimizes entity localization during the retrieval process to achieve significant performance improvements. Tree-RAG effectively organizes entities through the introduction of a hierarchical tree structure, while the Cuckoo Filter serves as an efficient data structure that supports rapid membership queries and dynamic updates. The experiment results demonstrate that our method is much faster than naive Tree-RAG while maintaining high levels of generative quality. When the number of trees is large, our method is hundreds of times faster than naive Tree-RAG. Our work is available at https://github.com/TUPYP7180/CFT-RAG-2025.
Authors: Shawn Im, Sharon Li
Abstract: Latent space steering methods provide a practical approach to controlling large language models by applying steering vectors to intermediate activations, guiding outputs toward desired behaviors while avoiding retraining. Despite their growing importance, the field lacks a unified understanding and consistent evaluation across tasks and datasets, hindering progress. This paper introduces a unified framework for analyzing and evaluating steering methods, formalizing their core principles and offering theoretical insights into their effectiveness. Through comprehensive empirical evaluations on multiple-choice and open-ended text generation tasks, we validate these insights, identifying key factors that influence performance and demonstrating the superiority of certain methods. Our work bridges theoretical and practical perspectives, offering actionable guidance for advancing the design, optimization, and deployment of latent space steering methods in LLMs.
Authors: Davide Domini, Gianluca Aguzzi, Lukas Esterle, Mirko Viroli
Abstract: In the last years, Federated learning (FL) has become a popular solution to train machine learning models in domains with high privacy concerns. However, FL scalability and performance face significant challenges in real-world deployments where data across devices are non-independently and identically distributed (non-IID). The heterogeneity in data distribution frequently arises from spatial distribution of devices, leading to degraded model performance in the absence of proper handling. Additionally, FL typical reliance on centralized architectures introduces bottlenecks and single-point-of-failure risks, particularly problematic at scale or in dynamic environments. To close this gap, we propose Field-Based Federated Learning (FBFL), a novel approach leveraging macroprogramming and field coordination to address these limitations through: (i) distributed spatial-based leader election for personalization to mitigate non-IID data challenges; and (ii) construction of a self-organizing, hierarchical architecture using advanced macroprogramming patterns. Moreover, FBFL not only overcomes the aforementioned limitations, but also enables the development of more specialized models tailored to the specific data distribution in each subregion. This paper formalizes FBFL and evaluates it extensively using MNIST, FashionMNIST, and Extended MNIST datasets. We demonstrate that, when operating under IID data conditions, FBFL performs comparably to the widely-used FedAvg algorithm. Furthermore, in challenging non-IID scenarios, FBFL not only outperforms FedAvg but also surpasses other state-of-the-art methods, namely FedProx and Scaffold, which have been specifically designed to address non-IID data distributions. Additionally, we showcase the resilience of FBFL's self-organizing hierarchical architecture against server failures.
Authors: Ganghua Wang, Yuhong Yang, Jie Ding
Abstract: The use of machine learning (ML) has become increasingly prevalent in various domains, highlighting the importance of understanding and ensuring its safety. One pressing concern is the vulnerability of ML applications to model stealing attacks. These attacks involve adversaries attempting to recover a learned model through limited query-response interactions, such as those found in cloud-based services or on-chip artificial intelligence interfaces. While existing literature proposes various attack and defense strategies, these often lack a theoretical foundation and standardized evaluation criteria. In response, this work presents a framework called ``Model Privacy'', providing a foundation for comprehensively analyzing model stealing attacks and defenses. We establish a rigorous formulation for the threat model and objectives, propose methods to quantify the goodness of attack and defense strategies, and analyze the fundamental tradeoffs between utility and privacy in ML models. Our developed theory offers valuable insights into enhancing the security of ML models, especially highlighting the importance of the attack-specific structure of perturbations for effective defenses. We demonstrate the application of model privacy from the defender's perspective through various learning scenarios. Extensive experiments corroborate the insights and the effectiveness of defense mechanisms developed under the proposed framework.
Authors: Abhilash Neog, Arka Daw, Sepideh Fatemi Khorasgani, Medha Sawhney, Aanish Pradhan, Mary E. Lofton, Bennett J. McAfee, Adrienne Breef-Pilz, Heather L. Wander, Dexter W Howard, Cayelan C. Carey, Paul Hanson, Anuj Karpatne
Abstract: Modeling Irregularly-sampled and Multivariate Time Series (IMTS) is crucial across a variety of applications where different sets of variates may be missing at different time-steps due to sensor malfunctions or high data acquisition costs. Existing approaches for IMTS either consider a two-stage impute-then-model framework or involve specialized architectures specific to a particular model and task. We perform a series of experiments to derive novel insights about the performance of IMTS methods on a variety of semi-synthetic and real-world datasets for both classification and forecasting. We also introduce Missing Feature-aware Time Series Modeling (MissTSM) or MissTSM, a novel model-agnostic and imputation-free approach for IMTS modeling. We show that MissTSM shows competitive performance compared to other IMTS approaches, especially when the amount of missing values is large and the data lacks simplistic periodic structures - conditions common to real-world IMTS applications.
Authors: Yi Shen, Jian Zhang, Jieyun Huang, Shuming Shi, Wenjing Zhang, Jiangze Yan, Ning Wang, Kai Wang, Zhaoxiang Liu, Shiguo Lian
Abstract: Recent advancements in slow thinking reasoning models have shown exceptional performance in complex reasoning tasks. However, these models often exhibit overthinking (generating redundant reasoning steps for simple problems), leading to excessive computational resource usage. While current mitigation strategies uniformly reduce reasoning tokens, they risk degrading performance on challenging tasks that require extended reasoning. This paper introduces Difficulty-Adaptive Slow Thinking (DAST), a novel framework that enables models to autonomously adjust the length of Chain-of-Thought (CoT) based on problem difficulty. We first propose a Token Length Budget (TLB) metric to quantify difficulty, then leverage budget-aware reward shaping and budget preference optimization to implement DAST. DAST penalizes overlong responses for simple tasks while incentivizing sufficient reasoning for complex problems. Experiments on diverse datasets and model scales demonstrate that DAST effectively mitigates overthinking (reducing token usage by over 30\% on average) while preserving reasoning accuracy on complex problems. Our codes and models are available at https://github.com/AnonymousUser0520/AnonymousRepo01.
Authors: Hanyang Zhao, Haoxian Chen, Yucheng Guo, Genta Indra Winata, Tingting Ou, Ziyu Huang, David D. Yao, Wenpin Tang
Abstract: Traditional preference tuning methods for LLMs/Visual Generative Models often rely solely on reward model labeling, which can be opaque, offer limited insights into the rationale behind preferences, and are prone to issues such as reward hacking or overfitting. We introduce Rich Preference Optimization (RPO), a novel pipeline that leverages rich feedback signals from Vision Language Models (VLMs) to improve the curation of preference pairs for fine-tuning visual generative models like text-to-image diffusion models. Our approach begins with prompting VLMs to generate detailed critiques of synthesized images, from which we further prompt VLMs to extract reliable and actionable image editing instructions. By implementing these instructions, we create refined images, resulting in synthetic, informative preference pairs that serve as enhanced tuning datasets. We demonstrate the effectiveness of our pipeline and the resulting datasets in fine-tuning state-of-the-art diffusion models.
Authors: Marimuthu Kalimuthu, David Holzm\"uller, Mathias Niepert
Abstract: Modeling high-frequency information is a critical challenge in scientific machine learning. For instance, fully turbulent flow simulations of the Navier-Stokes equations at Reynolds numbers 3500 and above can generate high-frequency signals due to swirling fluid motions caused by eddies and vortices. Faithfully modeling such signals using neural nets depends on the accurate reconstruction of moderate to high frequencies. However, it has been well known that neural nets exhibit spectral or frequency bias towards learning low-frequency components. Meanwhile, Fourier Neural Operators (FNOs) have emerged as a popular class of data-driven models for surrogate modeling and solving PDEs. Although impressive results were achieved on several PDE benchmark problems, FNOs perform poorly in learning non-dominant frequencies characterized by local features. This limitation stems from spectral bias inherent in neural nets and the explicit exclusion of high-frequency modes in FNOs and their variants. Therefore, to mitigate these issues and improve FNO's spectral learning capabilities to represent a broad range of frequency components, we propose two key architectural enhancements: (i) a parallel branch performing local spectral convolution (ii) a high-frequency propagation module. Moreover, we propose a novel frequency-sensitive loss based on radially binned spectral errors. This introduction of a parallel branch for local convolution reduces the trainable parameters by up to 50% while achieving the accuracy of FNO that relies solely on global convolution. Moreover, our findings demonstrate that the proposed model improves stability over longer rollouts. Experiments on six challenging PDEs in fluid mechanics, wave propagation, and biological pattern formation, and the qualitative and spectral analysis of predictions, show the effectiveness of our method over SOTA neural operator families of baselines.
Authors: Philippe Goulet Coulombe
Abstract: I show that ordinary least squares (OLS) predictions can be rewritten as the output of a restricted attention module, akin to those forming the backbone of large language models. This connection offers an alternative perspective on attention beyond the conventional information retrieval framework, making it more accessible to researchers and analysts with a background in traditional statistics. It falls into place when OLS is framed as a similarity-based method in a transformed regressor space, distinct from the standard view based on partial correlations. In fact, the OLS solution can be recast as the outcome of an alternative problem: minimizing squared prediction errors by optimizing the embedding space in which training and test vectors are compared via inner products. Rather than estimating coefficients directly, we equivalently learn optimal encoding and decoding operations for predictors. From this vantage point, OLS maps naturally onto the query-key-value structure of attention mechanisms. Building on this foundation, I discuss key elements of Transformer-style attention and draw connections to classic ideas from time series econometrics.
Authors: Neil Mehta, Ines Goncalves, Alberto Montagna, Mathis Fleury, Gustavo Caetano, Ines Esteves, Athanasios Vourvopoulos, Pulkit Grover, Patricia Figueiredo
Abstract: A growing interest has developed in the problem of training models of EEG features to predict brain activity measured using fMRI, i.e. the problem of EEG-to-fMRI synthesis. Despite some reported success, the statistical significance and generalizability of EEG-to-fMRI predictions remains to be fully demonstrated. Here, we investigate the predictive power of EEG for both task-evoked and spontaneous activity of the somatomotor network measured by fMRI, based on data collected from healthy subjects in two different sessions. We trained subject-specific distributed-lag linear models of time-varying, multi-channel EEG spectral power using Sparse Group LASSO regularization, and we showed that learned models outperformed conventional EEG somatomotor rhythm predictors as well as massive univariate correlation models. Furthermore, we showed that learned models were statistically significantly better than appropriate null models in most subjects and conditions, although less frequently for spontaneous compared to task-evoked activity. Critically, predictions improved significantly when training and testing on data acquired in the same session relative to across sessions, highlighting the importance of temporally separating the collection of train and test data to avoid data leakage and optimistic bias in model generalization. In sum, while we demonstrate that EEG models can provide fMRI predictions with statistical significance, we also show that predictive power is impaired for spontaneous fluctuations in brain activity and for models trained on data acquired in a different session. Our findings highlight the need to explicitly consider these often overlooked issues in the growing literature of EEG-to-fMRI synthesis.
Authors: K M Sajjadul Islam, Ravi Teja Karri, Srujan Vegesna, Jiawei Wu, Praveen Madiraju
Abstract: Understanding patient feedback is crucial for improving healthcare services, yet analyzing unlabeled short-text feedback presents significant challenges due to limited data and domain-specific nuances. Traditional supervised learning approaches require extensive labeled datasets, making unsupervised methods more viable for uncovering meaningful insights from patient feedback. This study explores unsupervised methods to extract meaningful topics from 439 survey responses collected from a healthcare system in Wisconsin, USA. A keyword-based filtering approach was applied to isolate complaint-related feedback using a domain-specific lexicon. To delve deeper and analyze dominant topics in feedback, we explored traditional topic modeling methods, including Latent Dirichlet Allocation (LDA) and Gibbs Sampling Dirichlet Multinomial Mixture (GSDMM), alongside BERTopic, an advanced neural embedding-based clustering approach. To improve coherence and interpretability where data are scarce and consist of short-texts, we propose kBERT, an integration of BERT embeddings with k-means clustering. Model performance was assessed using coherence scores (Cv ) for topic interpretability and average Inverted Rank-Biased Overlap (IRBOavg) for topic diversity. Results indicate that kBERT achieves the highest coherence (Cv = 0.53) and distinct topic separation (IRBOavg = 1.00), outperforming all other models in short-text healthcare feedback analysis. Our findings emphasize the importance of embedding-based techniques for topic identification and highlight the need for context-aware models in healthcare analytics.
Authors: Julian Kranz, Davide Gallon, Steffen Dereich, Arnulf Jentzen
Abstract: We study gradient flows for loss landscapes of fully connected feedforward neural networks with commonly used continuously differentiable activation functions such as the logistic, hyperbolic tangent, softplus or GELU function. We prove that the gradient flow either converges to a critical point or diverges to infinity while the loss converges to an asymptotic critical value. Moreover, we prove the existence of a threshold $\varepsilon>0$ such that the loss value of any gradient flow initialized at most $\varepsilon$ above the optimal level converges to it. For polynomial target functions and sufficiently big architecture and data set, we prove that the optimal loss value is zero and can only be realized asymptotically. From this setting, we deduce our main result that any gradient flow with sufficiently good initialization diverges to infinity. Our proof heavily relies on the geometry of o-minimal structures. We confirm these theoretical findings with numerical experiments and extend our investigation to more realistic scenarios, where we observe an analogous behavior.
Authors: Kehan Long, Jorge Cort\'es, Nikolay Atanasov
Abstract: Establishing stability certificates for closed-loop systems under reinforcement learning (RL) policies is essential to move beyond empirical performance and offer guarantees of system behavior. Classical Lyapunov methods require a strict stepwise decrease in the Lyapunov function but such certificates are difficult to construct for learned policies. The RL value function is a natural candidate but it is not well understood how it can be adapted for this purpose. To gain intuition, we first study the linear quadratic regulator (LQR) problem and make two key observations. First, a Lyapunov function can be obtained from the value function of an LQR policy by augmenting it with a residual term related to the system dynamics and stage cost. Second, the classical Lyapunov decrease requirement can be relaxed to a generalized Lyapunov condition requiring only decrease on average over multiple time steps. Using this intuition, we consider the nonlinear setting and formulate an approach to learn generalized Lyapunov functions by augmenting RL value functions with neural network residual terms. Our approach successfully certifies the stability of RL policies trained on Gymnasium and DeepMind Control benchmarks. We also extend our method to jointly train neural controllers and stability certificates using a multi-step Lyapunov loss, resulting in larger certified inner approximations of the region of attraction compared to the classical Lyapunov approach. Overall, our formulation enables stability certification for a broad class of systems with learned policies by making certificates easier to construct, thereby bridging classical control theory and modern learning-based methods.
Authors: Ziyao Cui, Minxing Zhang, Jian Pei
Abstract: Large language models (LLMs) are trained on massive corpora that may contain sensitive information, creating privacy risks under membership inference attacks (MIAs). Knowledge distillation is widely used to compress LLMs into smaller student models, but its privacy implications are poorly understood. We systematically evaluate how distillation affects MIA vulnerability across six teacher-student model pairs and six attack methods. We find that distilled student models do not consistently exhibit lower MIA success than their teacher models, and in some cases demonstrate substantially higher member-specific attack success, challenging the assumption that knowledge distillation inherently improves privacy. We attribute this to mixed supervision in distillation: for vulnerable training data points, teacher predictions often align with ground-truth labels, causing student models to learn overly confident predictions that amplify the separability between members and non-members; conversely, for non-vulnerable points, teacher predictions and ground truth frequently diverge, providing inconsistent learning signals. To mitigate this, we propose three practical interventions -- restricting distillation to non-vulnerable points, adding a low-dimensional Bottleneck Projection, and a normalization variant (NoNorm). Experiments show these methods reduce both aggregate and member-specific MIA success while preserving model utility, improving privacy-utility trade-offs for distilled LLMs.
Authors: Adam R. Klivans, Konstantinos Stavropoulos, Kevin Tian, Arsen Vasilyan
Abstract: Inspired by recent work on learning with distribution shift, we give a general outlier removal algorithm called iterative polynomial filtering and show a number of striking applications for supervised learning with contamination: (1) We show that any function class that can be approximated by low-degree polynomials with respect to a hypercontractive distribution can be efficiently learned under bounded contamination (also known as nasty noise). This is a surprising resolution to a longstanding gap between the complexity of agnostic learning and learning with contamination, as it was widely believed that low-degree approximators only implied tolerance to label noise. In particular, it implies the first efficient algorithm for learning halfspaces with $\eta$-bounded contamination up to error $2\eta+\epsilon$ with respect to the Gaussian distribution. (2) For any function class that admits the (stronger) notion of sandwiching approximators, we obtain near-optimal learning guarantees even with respect to heavy additive contamination, where far more than $1/2$ of the training set may be added adversarially. Prior related work held only for regression and in a list-decodable setting. (3) We obtain the first efficient algorithms for tolerant testable learning of functions of halfspaces with respect to any fixed log-concave distribution. Even the non-tolerant case for a single halfspace in this setting had remained open. These results significantly advance our understanding of efficient supervised learning under contamination, a setting that has been much less studied than its unsupervised counterpart.
Authors: Joseph Rowan, Buu Phan, Ashish Khisti
Abstract: We study a relaxation of the problem of coupling probability distributions -- a list of samples is generated from one distribution and an accept is declared if any one of these samples is identical to the sample generated from the other distribution. We propose a novel method for generating samples, which extends the Gumbel-max sampling suggested in Daliri et al. (arXiv:2408.07978) for coupling probability distributions. We also establish a corresponding lower bound on the acceptance probability, which we call the list matching lemma. We next discuss two applications of our setup. First, we develop a new mechanism for multi-draft speculative sampling that is simple to implement and achieves performance competitive with baselines such as SpecTr and SpecInfer across a range of language tasks. Our method also guarantees a certain degree of drafter invariance with respect to the output tokens which is not supported by existing schemes. We also provide a theoretical lower bound on the token level acceptance probability. As our second application, we consider distributed lossy compression with side information in a setting where a source sample is compressed and available to multiple decoders, each with independent side information. We propose a compression technique that is based on our generalization of Gumbel-max sampling and show that it provides significant gains in experiments involving synthetic Gaussian sources and the MNIST image dataset.
Authors: Tailin Zhou, Zhilin Chen, Wenlong Lyu, Zhitang Chen, Danny H. K. Tsang, Jun Zhang
Abstract: Optimizing complex systems, from discovering therapeutic drugs to designing high-performance materials, remains a fundamental challenge across science and engineering, as the underlying rules are often unknown and costly to evaluate. Offline optimization aims to optimize designs for target scores using pre-collected datasets without system interaction. However, conventional approaches may fail beyond training data, predicting inaccurate scores and generating inferior designs. This paper introduces ManGO, a diffusion-based framework that learns the design-score manifold, capturing the design-score interdependencies holistically. Unlike existing methods that treat design and score spaces in isolation, ManGO unifies forward prediction and backward generation, attaining generalization beyond training data. Key to this is its derivative-free guidance for conditional generation, coupled with adaptive inference-time scaling that dynamically optimizes denoising paths. Extensive evaluations demonstrate that ManGO outperforms 24 single- and 10 multi-objective optimization methods across diverse domains, including synthetic tasks, robot control, material design, DNA sequence, and real-world engineering optimization.
Authors: Yunpeng Qing, Yixiao Chi, Shuo Chen, Shunyu Liu, Kelu Yao, Sixu Lin, Litao Liu, Changqing Zou
Abstract: Recent advances in offline Reinforcement Learning (RL) have proven that effective policy learning can benefit from imposing conservative constraints on pre-collected datasets. However, such static datasets often exhibit distribution bias, resulting in limited generalizability. To address this limitation, a straightforward solution is data augmentation (DA), which leverages generative models to enrich data distribution. Despite the promising results, current DA techniques focus solely on reconstructing future trajectories from given states, while ignoring the exploration of history transitions that reach them. This single-direction paradigm inevitably hinders the discovery of diverse behavior patterns, especially those leading to critical states that may have yielded high-reward outcomes. In this work, we introduce Bidirectional Trajectory Diffusion (BiTrajDiff), a novel DA framework for offline RL that models both future and history trajectories from any intermediate states. Specifically, we decompose the trajectory generation task into two independent yet complementary diffusion processes: one generating forward trajectories to predict future dynamics, and the other generating backward trajectories to trace essential history transitions.BiTrajDiff can efficiently leverage critical states as anchors to expand into potentially valuable yet underexplored regions of the state space, thereby facilitating dataset diversity. Extensive experiments on the D4RL benchmark suite demonstrate that BiTrajDiff achieves superior performance compared to other advanced DA methods across various offline RL backbones.
Authors: Haoyu Wang, Peihao Wang, Mufei Li, Shikun Liu, Siqi Miao, Zhangyang Wang, Pan Li
Abstract: Modern large language models (LLMs) are inherently auto-regressive, requiring input to be serialized into flat sequences regardless of their structural dependencies. This serialization hinders the model's ability to leverage structural inductive biases, especially in tasks such as retrieval-augmented generation (RAG) and reasoning on data with native graph structures, where inter-segment dependencies are crucial. We introduce Graph-KV with the potential to overcome this limitation. Graph-KV leverages the KV-cache of text segments as condensed representations and governs their interaction through structural inductive biases. In this framework, 'target' segments selectively attend only to the KV-caches of their designated 'source' segments, rather than all preceding segments in a serialized sequence. This approach induces a graph-structured block mask, sparsifying attention and enabling a message-passing-like step within the LLM. Furthermore, strategically allocated positional encodings for source and target segments reduce positional bias and context window consumption. We evaluate Graph-KV across three scenarios: (1) seven RAG benchmarks spanning direct inference, multi-hop reasoning, and long-document understanding; (2) Arxiv-QA, a novel academic paper QA task with full-text scientific papers structured as citation ego-graphs; and (3) paper topic classification within a citation network. By effectively reducing positional bias and harnessing structural inductive biases, Graph-KV substantially outperforms baselines, including standard costly sequential encoding, across various settings. Code and the Graph-KV data are publicly available.
Authors: Manuel Brenner, Georgia Koppe
Abstract: Sequence modeling tasks across domains such as natural language processing, time series forecasting, and control require learning complex input-output mappings. Nonlinear recurrence is theoretically required for universal approximation of sequence-to-sequence functions, yet linear recurrent models often prove surprisingly effective. This raises the question of when nonlinearity is truly required. We present a framework to systematically dissect the functional role of nonlinearity in recurrent networks, identifying when it is computationally necessary and what mechanisms it enables. We address this using Almost Linear Recurrent Neural Networks (AL-RNNs), which allow recurrence nonlinearity to be gradually attenuated and decompose network dynamics into analyzable linear regimes, making computational mechanisms explicit. We illustrate the framework across diverse synthetic and real-world tasks, including classic sequence modeling benchmarks, a neuroscientific stimulus-selection task, and a multi-task suite. We demonstrate how the AL-RNN's piecewise linear structure enables identification of computational primitives such as gating, rule-based integration, and memory-dependent transients, revealing that these operations emerge within predominantly linear backbones. Across tasks, sparse nonlinearity improves interpretability by reducing and localizing nonlinear computations, promotes shared representations in multi-task settings, and reduces computational cost. Moreover, sparse nonlinearity acts as a useful inductive bias: in low-data regimes or when tasks require discrete switching between linear regimes, sparsely nonlinear models often match or exceed fully nonlinear architectures. Our findings provide a principled approach for identifying where nonlinearity is functionally necessary, guiding the design of recurrent architectures that balance performance, efficiency, and interpretability.
Authors: Katharina Limbeck, Lydia Mezrag, Guy Wolf, Bastian Rieck
Abstract: Graph Neural Networks (GNNs) have shown significant success for graph-based tasks. Motivated by the prevalence of large datasets in real-world applications, pooling layers are crucial components of GNNs. By reducing the size of input graphs, pooling enables faster training and potentially better generalisation. However, existing pooling operations often optimise for the learning task at the expense of discarding fundamental graph structures, thus reducing interpretability. This leads to unreliable performance across dataset types, downstream tasks and pooling ratios. Addressing these concerns, we propose novel graph pooling layers for structure-aware pooling via edge collapses. Our methods leverage diffusion geometry and iteratively reduce a graph's size while preserving both its metric structure and its structural diversity. We guide pooling using magnitude, an isometry-invariant diversity measure, which permits us to control the fidelity of the pooling process. Further, we use the spread of a metric space as a faster and more stable alternative ensuring computational efficiency. Empirical results demonstrate that our methods (i) achieve top performance compared to alternative pooling layers across a range of diverse graph classification tasks, (ii) preserve key spectral properties of the input graphs, and (iii) retain high accuracy across varying pooling ratios.
Authors: Elif Y{\i}lmaz, Christos Dimitrakakis
Abstract: We study a two-player zero-sum game in which the row player aims to maximize their payoff against an adversarial column player, under an unknown payoff matrix estimated through bandit feedback. We propose three algorithms based on the Explore-Then-Commit (ETC) framework. The first adapts it to zero-sum games, the second incorporates adaptive elimination that leverages the $\varepsilon$-Nash Equilibrium property to efficiently select the optimal action pair, and the third extends the elimination algorithm by employing non-uniform exploration. Our objective is to demonstrate the applicability of ETC in a zero-sum game setting by focusing on learning pure strategy Nash Equilibria. A key contribution of our work is a derivation of instance-dependent upper bounds on the expected regret of our proposed algorithms, which has received limited attention in the literature on zero-sum games. Particularly, after $T$ rounds, we achieve an instance-dependent regret upper bounds of $O(\Delta + \sqrt{T})$ for ETC in zero-sum game setting and $O(\log (T \Delta^2)/\Delta)$ for the adaptive elimination algorithm and its variant with non-uniform exploration, where $\Delta$ denotes the suboptimality gap. Therefore, our results indicate that ETC-based algorithms perform effectively in zero-sum game settings, achieving regret bounds comparable to existing methods while providing insight through instance-dependent analysis.
Authors: Vanessa Kosoy
Abstract: We propose a new variant of online learning that we call "ambiguous online learning". In this setting, the learner is allowed to produce multiple predicted labels. Such an "ambiguous prediction" is considered correct when at least one of the labels is correct, and none of the labels are "predictably wrong". The definition of "predictably wrong" comes from a hypothesis class in which hypotheses are also multi-valued. Thus, a prediction is "predictably wrong" if it's not allowed by the (unknown) true hypothesis. In particular, this setting is natural in the context of multivalued dynamical systems, recommendation algorithms and lossless compression. It is also strongly related to so-called "apple tasting". We show that in this setting, there is a trichotomy of mistake bounds: up to logarithmic factors, any hypothesis class has an optimal mistake bound of either Theta(1), Theta(sqrt(N)) or N.
Authors: Defeng Liu, Ying Liu, Carson Eisenach
Abstract: In this work, we study how to efficiently apply reinforcement learning (RL) for solving large-scale stochastic optimization problems by leveraging intervention models. The key of the proposed methodology is to better explore the solution space by simulating and composing the stochastic processes using pre-trained deep learning (DL) models. We demonstrate our approach on a challenging real-world application, the multi-sourcing multi-period inventory management problem in supply chain optimization. In particular, we employ deep RL models for learning and forecasting the stochastic supply chain processes under a range of assumptions. Moreover, we also introduce a constraint coordination mechanism, designed to forecast dual costs given the cross-products constraints in the inventory network. We highlight that instead of directly modeling the complex physical constraints into the RL optimization problem and solving the stochastic problem as a whole, our approach breaks down those supply chain processes into scalable and composable DL modules, leading to improved performance on large real-world datasets. We also outline open problems for future research to further investigate the efficacy of such models.
Authors: Yiyuan Yang, Zichuan Liu, Lei Song, Kai Ying, Zhiguang Wang, Tom Bamford, Svitlana Vyetrenko, Jiang Bian, Qingsong Wen
Abstract: Time series anomaly detection (TSAD) has traditionally focused on binary classification and often lacks the fine-grained categorization and explanatory reasoning required for transparent decision-making. To address these limitations, we propose Time-series Reasoning for Anomaly (Time-RA), a novel task that reformulates TSAD from a discriminative into a generative, reasoning-intensive paradigm. To facilitate this, we introduce RATs40K, the first real-world large-scale multimodal benchmark with ~40,000 samples across 10 domains, integrating raw time series, textual context, and visual plots with structured reasoning annotations. Extensive benchmarking shows that while supervised fine-tuning and visual representations boost diagnostic accuracy and reasoning consistency, performance varies across complex scenarios. Notably, fine-tuned models demonstrate strong "plug-and-play" transferability, outperforming traditional baselines on unseen real-world datasets. Our work establishes a foundation for interpretable, multimodal time series analysis. All code (https://github.com/yyysjz1997/Time-RA) and the RATs40K dataset (https://huggingface.co/datasets/Time-RA/RATs40K) are fully open-sourced to facilitate future research.
URLs: https://github.com/yyysjz1997/Time-RA), https://huggingface.co/datasets/Time-RA/RATs40K)
Authors: Zihang Ma, Qitian Yin
Abstract: Graph node classification is a fundamental task in graph neural networks (GNNs), aiming to assign predefined class labels to nodes. On the PubMed citation network dataset, we observe significant classification difficulty disparities, with Category 2 achieving only 74.4% accuracy in traditional GCN, 7.5% lower than Category 1. To address this, we propose a Wasserstein-Rubinstein (WR) distance enhanced Expert Fusion Model (WR-EFM), training specialized GNN models for Categories 0/1 (with layer normalization and residual connections) and Multi-hop Graph Attention Networks (GAT) for Category 2. The WR distance metric optimizes representation similarity between models, particularly focusing on improving Category 2 performance. Our adaptive fusion strategy dynamically weights models based on category-specific performance, with Category 2 assigned a GAT weight of 0.8. WR distance further guides the fusion process by measuring distributional differences between model representations, enabling more principled integration of complementary features. Experimental results show WR-EFM achieves balanced accuracy across categories: 77.8% (Category 0), 78.0% (Category 1), and 79.9% (Category 2), outperforming both single models and standard fusion approaches. The coefficient of variation (CV) of WR-EFM's category accuracies is 0.013, 77.6% lower than GCN's 0.058, demonstrating superior stability. Notably, WR-EFM improves Category 2 accuracy by 5.5% compared to GCN, verifying the effectiveness of WR-guided fusion in capturing complex structural patterns. This work provides a novel paradigm for handling class-imbalanced graph classification tasks. To promote the research community, we release our project at https://github.com/s010m00n/GASEM4NC.
Authors: Nhut Truong, Uri Hasson
Abstract: Topographic convolutional neural networks (TCNNs) are computational models that can simulate aspects of the brain's spatial and functional organization. However, it is unclear whether and how different types of topographic regularization shape robustness, representational structure, and functional organization during end-to-end training. We address this question by comparing TCNNs trained with two local spatial losses applied to a penultimate-layer topographic grid: i) Weight Similarity (WS), whose objective penalizes differences between neighboring units' incoming weight vectors, and ii) Activation Similarity (AS), whose objective penalizes differences between neighboring units' activation patterns over stimuli. We evaluate the trained models on classification accuracy, robustness to weight perturbations and input degradation, the spatial organization of learned representations, and development of category-selective "expert units" in the penultimate layer. Both losses changed inter-unit correlation structure, but in qualitatively different ways. WS produced smooth topographies, with correlated neighborhoods. In contrast, AS produced a bimodal inter-unit correlation structure that lacked spatial smoothness. AS and WS training increased robustness relative to control (non-topographic) models: AS improved robustness to image degradation on CIFAR-10, WS did so on MNIST, and both improved robustness to weight perturbations. WS was also associated with greater input sensitivity at the unit level and stronger functional localization. In addition, as compared to control models, both AS and WS produced differences in orientation tuning, symmetry sensitivity, and eccentricity profiles of units. Together, these results show that local topographic regularization can improve robustness during end-to-end training while systematically reshaping representational structure.
Authors: Chengsong Huang, Wenhao Yu, Xiaoyang Wang, Hongming Zhang, Zongxia Li, Ruosen Li, Jiaxin Huang, Haitao Mi, Dong Yu
Abstract: Self-evolving Large Language Models (LLMs) offer a scalable path toward super-intelligence by autonomously generating, refining, and learning from their own experiences. However, existing methods for training such models still rely heavily on vast human-curated tasks and labels, typically via fine-tuning or reinforcement learning, which poses a fundamental bottleneck to advancing AI systems toward capabilities beyond human intelligence. To overcome this limitation, we introduce R-Zero, a fully autonomous framework that generates its own training data from scratch. Starting from a single base LLM, R-Zero initializes two independent models with distinct roles, a Challenger and a Solver. These models are optimized separately and co-evolve through interaction: the Challenger is rewarded for proposing tasks near the edge of the Solver capability, and the Solver is rewarded for solving increasingly challenging tasks posed by the Challenger. This process yields a targeted, self-improving curriculum without any pre-existing tasks and labels. Empirically, R-Zero substantially improves reasoning capability across different backbone LLMs, e.g., boosting the Qwen3-4B-Base by +6.49 on math-reasoning benchmarks and +7.54 on general-domain reasoning benchmarks.
Authors: Haitong Luo, Suhang Wang, Weiyao Zhang, Ruiqi Meng, Xuying Meng, Yujun Zhang
Abstract: Graph ``pre-training and prompt-tuning'' aligns downstream tasks with pre-trained objectives to enable efficient knowledge transfer under limited supervision. However, current methods typically rely on single-filter backbones (e.g., low-pass), whereas real-world graphs exhibit inherent spectral diversity. Our theoretical \textit{Spectral Specificity} principle reveals that effective knowledge transfer requires alignment between pre-trained spectral filters and the intrinsic spectrum of downstream graphs. This identifies two fundamental limitations: (1) Knowledge Bottleneck: single-filter models suffer from irreversible information loss by suppressing signals from other frequency bands (e.g., high-frequency); (2) Utilization Bottleneck: spectral mismatches between pre-trained filters and downstream spectra lead to significant underutilization of pre-trained knowledge. To bridge this gap, we propose HS-GPPT. We utilize a hybrid spectral backbone to construct an abundant knowledge basis. Crucially, we introduce Spectral-Aligned Prompt Tuning to actively align the downstream graph's spectrum with diverse pre-trained filters, facilitating comprehensive knowledge utilization across both homophily and heterophily. Extensive experiments validate the effectiveness under both transductive and inductive learning settings.
Authors: Pengyu Lai, Yixiao Chen, Hui Xu
Abstract: A primary challenge in using neural networks to approximate nonlinear dynamical systems governed by partial differential equations (PDEs) is transforming these systems into a suitable format, especially when dealing with non-linearizable dynamics or the need for infinite-dimensional spaces for linearization. This paper introduces DyMixOp, a novel neural operator framework for PDEs that integrates insights from complex dynamical systems to address this challenge. Grounded in inertial manifold theory, DyMixOp transforms infinite-dimensional nonlinear PDE dynamics into a finite-dimensional latent space, establishing a structured foundation that maintains essential nonlinear interactions and enhances physical interpretability. A key innovation is the Local-Global-Mixing (LGM) transformation, inspired by convection dynamics in turbulence. This transformation effectively captures both fine-scale details and nonlinear interactions, while mitigating spectral bias commonly found in existing neural operators. The framework is further strengthened by a dynamics-informed architecture that connects multiple LGM layers to approximate linear and nonlinear dynamics, reflecting the temporal evolution of dynamical systems. Experimental results across diverse PDE benchmarks demonstrate that DyMixOp achieves state-of-the-art performance, significantly reducing prediction errors, particularly in convection-dominated scenarios reaching up to 86.7\%, while maintaining computational efficiency and scalability.
Authors: Zhihao Liu, Jian Lou, Yuke Hu, Xiaochen Li, Yitian Chen, Tailun Chen, Zhizhen Qin, Kui Ren, Zhan Qin
Abstract: Large language models (LLMs) are trained on massive datasets that may include private or copyrighted content. Due to growing privacy and ownership concerns, data owners may request the removal of their data from trained models. Machine unlearning provides a practical solution by removing the influence of specific data without full retraining. However, most existing methods still suffer from over-unlearning due to the lack of a principled mechanism to regulate the forgetting boundary, leading to unnecessary utility degradation and heightened privacy and robustness risks. In this work, we propose EGUP (Entanglement-Guided Unlearning with Proxy Constraint), a novel framework that leverages entanglement and proxy constraint to guide the unlearning process while mitigating over-unlearning. Within each iteration, EGUP employs inter-sample entanglement to adaptively reweight the unlearning strength, assigning greater unlearning efforts to forget samples that are semantically closer to retained knowledge. Across iterations, EGUP leverages intra-sample entanglement to track the representation shift of each forget sample and dynamically adjust its unlearning effort. In addition, we incorporate a proxy constraint that approximates the model's expected outputs after unlearning, forming a reference boundary that softly regularizes the unlearning process. EGUP is compatible with existing gradient-based objectives and serves as a plug-and-play enhancement. We evaluate EGUP on the TOFU and MUSE benchmarks, demonstrating consistent improvements in the unlearning-utility trade-off across multiple LLMs. Moreover, EGUP achieves performance close to the retrained model while remaining scalable and robust.
Authors: Lu Han, Xiuying Wang
Abstract: Traditional decision tree models, which rely exclusively on numerical variables, often face challenges in handling high-dimensional data and are limited in their ability to incorporate textual information effectively. To address these limitations, we propose the integrated multivariate segmentation tree (IMST), a comprehensive framework designed to improve credit evaluation for small- and medium-sized enterprises (SMEs) by integrating financial data with textual sources. This method comprises three core stages: (1) transforming textual data into numerical matrices through matrix factorization, (2) selecting salient financial features using Lasso regression, and (3) constructing a multivariate segmentation tree based on either the Gini index or entropy, with weakest-link pruning applied to control model complexity. Experimental results based on a dataset of 1,428 Chinese SMEs demonstrated that IMST achieved an accuracy rate of 88.9%, surpassing both baseline decision trees (87.4%) and conventional models such as support vector machines and neural networks. Furthermore, the proposed model demonstrated superior interpretability and computational efficiency, featuring a more streamlined architecture and improved risk detection capabilities.
Authors: Dongseok Kim, Hyoungsun Choi, Mohamed Jismy Aashik Rasool, Gisung Oh
Abstract: We introduce ORACLE, a framework for explaining neural networks on tabular data and scientific factorial designs. ORACLE summarizes a trained network's prediction surface with main effects and pairwise interactions by treating the network as a black-box response, discretizing the inputs onto a grid, and fitting an orthogonal factorial (ANOVA-style) surrogate -- the $L^2$ orthogonal projection of the model response onto a finite-dimensional factorial subspace. A simple centering and $\mu$-rebalancing step then expresses this surrogate as main- and interaction-effect tables that remain faithful to the original model in the $L^2$ sense. The resulting grid-based interaction maps are easy to visualize, comparable across backbones, and directly aligned with classical design-of-experiments practice. On synthetic factorial benchmarks and low- to medium-dimensional tabular regression tasks, ORACLE more accurately recovers ground-truth interaction structure and hotspots than Monte Carlo SHAP-family interaction methods, as measured by ranking, localization, and cross-backbone stability. We also discuss its scope in latent image and text settings: grid-based factorial surrogates are most effective when features admit an interpretable factorial structure, making ORACLE particularly well-suited to scientific and engineering workflows that require stable DoE-style interaction summaries.
Authors: Seul Lee, Karsten Kreis, Srimukh Prasad Veccham, Meng Liu, Danny Reidenbach, Saee Paliwal, Weili Nie, Arash Vahdat
Abstract: A well-known pitfall of molecular generative models is that they are not guaranteed to generate synthesizable molecules. Existing solutions for this problem often struggle to effectively navigate exponentially large combinatorial space of synthesizable molecules and suffer from poor coverage. To address this problem, we introduce ReaSyn, an iterative generative pathway refinement framework that obtains synthesizable analogs to input molecules by projecting them onto synthesizable space. Specifically, we propose a simple synthetic pathway representation that allows for generating pathways in both bottom-up and top-down traversal of synthetic trees. We design ReaSyn so that both bottom-up and top-down pathways can be sampled with a single unified autoregressive model. ReaSyn can thus iteratively refine subtrees of generated synthetic trees in a bidirectional manner. Further, we introduce a discrete flow model that refines the generated pathway at the entire pathway level with edit operations: insertion, deletion, and substitution. The iterative refinement cycle of (1) bottom-up decoding, (2) top-down decoding, and (3) holistic editing constitutes a powerful pathway reasoning strategy, allowing the model to explore the vast space of synthesizable molecules. Experimentally, ReaSyn achieves the highest reconstruction rate and pathway diversity in synthesizable molecule reconstruction and the highest optimization performance in synthesizable goal-directed molecular optimization, and significantly outperforms previous synthesizable projection methods in synthesizable hit expansion. These results highlight ReaSyn's superior ability to navigate combinatorially-large synthesizable chemical space.
Authors: \'Alvaro Parafita, Tomas Garriga, Axel Brando, Francisco J. Cazorla
Abstract: Among explainability techniques, SHAP stands out as one of the most popular, but often overlooks the causal structure of the problem. In response, do-SHAP employs interventional queries, but its reliance on estimands hinders its practical application. To address this problem, we propose the use of estimand-agnostic approaches, which allow for the estimation of any identifiable query from a single model, making do-SHAP feasible on complex graphs. We also develop a novel algorithm to significantly accelerate its computation at a negligible cost, as well as a method to explain inaccessible Data Generating Processes. We demonstrate the estimation and computational performance of our approach, and validate it on two real-world datasets, highlighting its potential in obtaining reliable explanations.
Authors: Luca Zhou, Pratham Yashwante, Marshall Fisher, Alessio Sampieri, Zihao Zhou, Fabio Galasso, Rose Yu
Abstract: Time series captioning, the task of describing time series in natural language, requires numeric and temporal reasoning, trend interpretation, and contextual understanding. Existing benchmarks, however, often rely on fully synthetic or generic captions, and typically neglect metadata and visual representations. We introduce CaTS-Bench, a comprehensive benchmark for Context-aware Time Series reasoning across $11$ diverse domains, centered on a gold-standard evaluation set of $1746$ human-rewritten captions that measure how effectively models translate numeric trends into immediately interpretable narratives. To address the scarcity of human-annotated data, we also propose a scalable pipeline for generating high-fidelity synthetic captions, the quality of which we validate. We evaluate leading Vision-Language Models on our benchmark, revealing that even proprietary models struggle to capture numeric nuances in temporal descriptions, while finetuning open-source models on synthetic data yields substantial performance gains. Finally, we release a diagnostic suite of $910$ multiple-choice questions and tailored numeric metrics to gauge time-series-specific reasoning capabilities, establishing CaTS-Bench as a reliable foundation for grounded, multimodal language generation in numeric domains.
Authors: Bingshuai Liu, Ante Wang, Zijun Min, Liang Yao, Haibo Zhang, Yang Liu, Xu Han, Peng Li, Anxiang Zeng, Jinsong Su
Abstract: Large Language Models (LLMs) increasingly rely on reinforcement learning with verifiable rewards (RLVR) to elicit reliable chain-of-thought reasoning. However, the training process remains bottlenecked by the computationally expensive rollout stage. Existing acceleration methods-such as parallelization, objective- and data-driven modifications, and replay buffers-either incur diminishing returns, introduce bias, or overlook redundancy across iterations. We identify that rollouts from consecutive training epochs frequently share a large portion of overlapping segments, wasting computation. To address this, we propose SPEC-RL, a novel framework that integrates SPECulative decoding with the RL rollout process. SPEC-RL reuses prior trajectory segments as speculative prefixes and extends them via a draft-and-verify mechanism, avoiding redundant generation while ensuring policy consistency. Experiments on diverse math reasoning and generalization benchmarks, including AIME24, MATH-500, OlympiadBench, MMLU-STEM, and others, demonstrate that SPEC-RL reduces rollout time by 2-3x without compromising policy quality. As a purely rollout-stage enhancement, SPEC-RL integrates seamlessly with mainstream algorithms (e.g., PPO, GRPO, DAPO), offering a general and practical path to scale RLVR for large reasoning models. Our code is available at https://github.com/ShopeeLLM/Spec-RL
Authors: Sophia V. Kuhn, Rafael Bischof, Marius Weber, Antoine Binggeli, Michael A. Kraus, Walter Kaufmann, Fernando P\'erez-Cruz
Abstract: Aging infrastructure portfolios pose a critical resource allocation challenge: deciding which structures require intervention and which can safely remain in service. Structural assessments must balance the trade-off between cheaper, conservative analysis methods and accurate but costly simulations that do not scale portfolio-wide. We propose Bayesian neural network (BNN) surrogates for rapid structural pre-assessment of worldwide common bridge types, such as reinforced concrete frame bridges. Trained on a large-scale database of non-linear finite element analyses generated via a parametric pipeline and developed based on the Swiss Federal Railway's bridge portfolio, the models accurately and efficiently estimate high-fidelity structural analysis results by predicting code compliance factors with calibrated epistemic uncertainty. Our BNN surrogate enables fast, uncertainty-aware triage: flagging likely critical structures and providing guidance where refined analysis is pertinent. We demonstrate the framework's effectiveness in a real-world case study of a railway underpass, showing its potential to significantly reduce costs and emissions by avoiding unnecessary analyses and physical interventions across entire infrastructure portfolios.
Authors: Hanyang Zhao, Dawen Liang, Wenpin Tang, David Yao, Nathan Kallus
Abstract: We propose DiFFPO, Diffusion Fast and Furious Policy Optimization, a unified framework for training masked diffusion large language models (dLLMs) to reason not only better (furious), but also faster via reinforcement learning (RL). We first unify the existing baseline approach such as d1 by proposing to train surrogate policies via off-policy RL, whose likelihood is much more tractable as an approximation to the true dLLM policy. This naturally motivates a more accurate and informative two-stage likelihood approximation combined with importance sampling correction, which leads to generalized RL algorithms with better sample efficiency and superior task performance. Second, we propose a new direction of joint training efficient samplers/controllers of dLLMs policy. Via RL, we incentivize dLLMs' natural multi-token prediction capabilities by letting the model learn to adaptively allocate an inference threshold for each prompt. By jointly training the sampler, we yield better accuracies with lower number of function evaluations (NFEs) compared to training the model only, obtaining the best performance in improving the Pareto frontier of the inference-time compute of dLLMs. We showcase the effectiveness of our pipeline by training open source large diffusion language models over benchmark math and planning tasks.
Authors: Jacob Feitelberg, Dwaipayan Saha, Kyuseong Choi, Zaid Ahmad, Anish Agarwal, Raaz Dwivedi
Abstract: Missing data is a pervasive problem in tabular settings. Existing solutions range from simple averaging to complex generative adversarial networks, but due to each method's large variance in performance across real-world domains and time-consuming hyperparameter tuning, no default imputation method exists. Building on TabPFN, a recent tabular foundation model for supervised learning, we propose TabImpute, a pre-trained transformer that delivers accurate and fast zero-shot imputations requiring no fitting or hyperparameter tuning at inference time. To train and evaluate TabImpute, we introduce (i) an entry-wise featurization for tabular settings, which enables a 100x speedup over the previous TabPFN imputation method, (ii) a synthetic training data generation pipeline incorporating realistic missingness patterns, and (iii) MissBench, a comprehensive benchmark with 42 OpenML datasets and 13 new missingness patterns. MissBench spans domains such as medicine, finance, and engineering, showcasing TabImpute's robust performance compared to 12 established imputation methods.
Authors: Jinhui Bai, Andreas Christmann, Lei Shi
Abstract: In this paper, we propose a novel kernel stochastic gradient descent (SGD) algorithm for large-scale supervised learning with general losses. Compared to traditional kernel SGD, our algorithm improves efficiency and scalability through an innovative regularization strategy. By leveraging the infinite series expansion of spherical radial basis functions, this strategy projects the stochastic gradient onto a finite-dimensional hypothesis space, which is adaptively scaled according to the bias-variance trade-off, thereby enhancing generalization performance. Based on a new estimation of the spectral structure of the kernel-induced covariance operator, we develop an analytical framework that unifies optimization and generalization analyses. We prove that both the last iterate and the suffix average converge at minimax-optimal rates, and we further establish optimal strong convergence in the reproducing kernel Hilbert space. Our framework accommodates a broad class of classical loss functions, including least-squares, Huber, and logistic losses. Moreover, the proposed algorithm significantly reduces computational complexity and achieves optimal storage complexity by incorporating coordinate-wise updates from linear SGD, thereby avoiding the costly pairwise operations typical of kernel SGD and enabling efficient processing of streaming data. Finally, extensive numerical experiments demonstrate the efficiency of our approach.
Authors: John Dunbar, Scott Aaronson
Abstract: We establish that randomly initialized neural networks, with large width and a natural choice of hyperparameters, have nearly independent outputs exactly when their activation function is nonlinear with zero mean under the Gaussian measure: $\mathbb{E}_{z \sim \mathcal{N}(0,1)}[\sigma(z)]=0$. For example, this includes ReLU and GeLU with an additive shift, as well as tanh, but not ReLU or GeLU by themselves. Because of their nearly independent outputs, we propose neural networks with zero-mean activation functions as a promising candidate for the Alignment Research Center's computational no-coincidence conjecture -- a conjecture that aims to measure the limits of AI interpretability.
Authors: Han Wu, Jie Yin
Abstract: Few-shot knowledge graph relational learning seeks to perform reasoning over relations given only a limited number of training examples. While existing approaches largely adopt a meta-learning framework for enabling fast adaptation to new relations, they suffer from two key pitfalls. First, they learn relation meta-knowledge in isolation, failing to capture common relational patterns shared across tasks. Second, they struggle to effectively incorporate local, task-specific contexts crucial for rapid adaptation. To address these limitations, we propose MoEMeta, a novel meta-learning framework that disentangles globally shared knowledge from task-specific contexts to enable both effective model generalization and rapid adaptation. MoEMeta introduces two key innovations: (i) a mixture-of-experts (MoE) model that learns globally shared relational prototypes to enhance generalization, and (ii) a task-tailored adaptation mechanism that captures local contexts for fast task-specific adaptation. By balancing global generalization with local adaptability, MoEMeta significantly advances few-shot relational learning. Extensive experiments and analyses on three KG benchmarks show that MoEMeta consistently outperforms existing baselines, achieving state-of-the-art performance.
Authors: Di Zhang, Xun Wu, Shaohan Huang, Lingjie Jiang, Yaru Hao, Li Dong, Zewen Chi, Zhifang Sui, Furu Wei
Abstract: Recent advances in reinforcement learning (RL) have substantially improved the training of large-scale language models, leading to significant gains in generation quality and reasoning ability. However, most existing research focuses on dense models, while RL training for Mixture-of-Experts (MoE) architectures remains underexplored. To address the instability commonly observed in MoE training, we propose a novel router-aware approach to optimize importance sampling (IS) weights in off-policy RL. Specifically, we design a rescaling strategy guided by router logits, which effectively reduces gradient variance and mitigates training divergence. Experimental results demonstrate that our method significantly improves both the convergence stability and the final performance of MoE models, highlighting the potential of RL algorithmic innovations tailored to MoE architectures and providing a promising direction for efficient training of large-scale expert models.
Authors: Ruifeng Ren, Sheng Ouyang, Huayi Tang, Yong Liu
Abstract: Attention-based Transformers have demonstrated strong adaptability across a wide range of tasks and have become the backbone of modern Large Language Models (LLMs). However, their underlying mechanisms remain open for further exploration. The energy-based perspective has long provided a valuable principle for understanding neural computation. In this paper, we revisit the principle of energy as a lens to understand attention-based Transformer models. We present a unified energy-based framework which is composed of three key components: the local energy $E_i$, the global energy $F$, and the employed optimization algorithms. We show that different attention forms including unnormalized linear attention, gated linear attention and standard softmax attention can be induced by choosing their corresponding recipes within this framework. Building on this framework, we propose energy-based modifications of attention structures. Inspired by classical gradient descent (GD) algorithms, we extend the original attention formulation based on standard GD to the momentum-based GD, Nesterov Accelerated Gradient (NAG), and Newton's method, each inducing a corresponding new attention structure. Our experiments provide preliminary support for the potential of the energy-based framework for designing attention mechanisms.
Authors: Ziyi Wang, Lijian Jiang
Abstract: Data assimilation (DA) has increasingly emerged as a critical tool for state estimation across a wide range of applications. It is significantly challenging when the governing equations of the underlying dynamics are unknown. To this end, various machine learning approaches have been employed to construct a surrogate state transition model in a supervised learning framework, which relies on pre-computed training datasets. However, it is often infeasible to obtain noise-free ground-truth state sequences in practice. To address this challenge, we propose a novel method that integrates reinforcement learning with ensemble-based Bayesian filtering methods, enabling the learning of surrogate state transition model for unknown dynamics directly from noisy observations, without using true state trajectories. Specifically, we treat the process for computing maximum likelihood estimation of surrogate model parameters as a sequential decision-making problem, which can be formulated as a discrete-time Markov decision process (MDP). Under this formulation, learning the surrogate transition model is equivalent to finding an optimal policy of the MDP, which can be effectively addressed using reinforcement learning techniques. Once the model is trained offline, state estimation can be performed in the online stage using filtering methods based on the learned dynamics. The proposed framework accommodates a wide range of observation scenarios, including nonlinear and partially observed measurement models. A few numerical examples demonstrate that the proposed method achieves superior accuracy and robustness in high-dimensional settings.
Authors: Yicheng Zhan, Fahim Ahmed, Amy Burrell, Matthew J. Tonkin, Sarah Galambos, Jessica Woodhams, Dalal Alrajeh
Abstract: Effective crime linkage analysis is crucial for identifying serial offenders and enhancing public safety. To address limitations of traditional crime linkage methods in handling high-dimensional, sparse, and heterogeneous data, we propose a Siamese Autoencoder framework that learns meaningful latent representations and uncovers correlations in complex crime data. Using data from the Violent Crime Linkage Analysis System (ViCLAS), maintained by the Serious Crime Analysis Section of the UK's National Crime Agency, our approach mitigates signal dilution in sparse feature spaces by integrating geographic-temporal features at the decoder stage. This design amplifies behavioral representations rather than allowing them to be overshadowed at the input level, yielding consistent improvements across multiple evaluation metrics. We further analyze how different domain-informed data reduction strategies influence model performance, providing practical guidance for preprocessing in crime linkage contexts. Our results show that advanced machine learning approaches can substantially enhance linkage accuracy, improving AUC by up to 9% over traditional methods while offering interpretable insights to support investigative decision-making.
Authors: Tianlei Chen, Yuxiao Chen, Yang Li, Feifei Wang
Abstract: Automated international classification of diseases (ICD) coding aims to assign multiple disease codes to clinical documents and plays a critical role in healthcare informatics. However, its performance is hindered by the extreme long-tail distribution of the ICD ontology, where a few common codes dominate while thousands of rare codes have very few examples. To address this issue, we propose a Probability-Biased Directed Graph Attention model (ProBias) that partitions codes into common and rare sets and allows information to flow only from common to rare codes. Edge weights are determined by conditional co-occurrence probabilities, which guide the attention mechanism to enrich rare-code representations with clinically related signals. To provide higher-quality semantic representations as model inputs, we further employ large language models to generate enriched textual descriptions for ICD codes, offering external clinical context that complements statistical co-occurrence signals. Applied to automated ICD coding, our approach significantly improves the representation and prediction of rare codes, achieving state-of-the-art performance on three benchmark datasets. In particular, we observe substantial gains in macro-averaged F1 score, a key metric for long-tail classification.
Authors: Youming Chen, Zhaoqiang Liu
Abstract: Diffusion models (DMs) have demonstrated to be powerful priors for signal recovery, but their application to 1-bit quantization tasks, such as 1-bit compressed sensing and logistic regression, remains a challenge. This difficulty stems from the inherent non-linear link function in these tasks, which is either non-differentiable or lacks an explicit characterization. To tackle this issue, we introduce Diff-OneBit, which is a fast and effective DM-based approach for signal recovery under 1-bit quantization. Diff-OneBit addresses the challenge posed by non-differentiable or implicit links functions via leveraging a differentiable surrogate likelihood function to model 1-bit quantization, thereby enabling gradient based iterations. This function is integrated into a flexible plug-and-play framework that decouples the data-fidelity term from the diffusion prior, allowing any pretrained DM to act as a denoiser within the iterative reconstruction process. Extensive experiments on the FFHQ, CelebA and ImageNet datasets demonstrate that Diff-OneBit gives high-fidelity reconstructed images, outperforming state-of-the-art methods in both reconstruction quality and computational efficiency across 1-bit compressed sensing and logistic regression tasks. Our code is available at https://github.com/Chenyouming123/DiffOneBit.
Authors: Sanjit Dandapanthula, Aaditya Ramdas
Abstract: Deep equilibrium models (DEQs) have recently emerged as a powerful paradigm for training infinitely deep weight-tied neural networks that achieve state of the art performance across many modern machine learning tasks. Despite their practical success, theoretically understanding the gradient descent dynamics for training DEQs remains an area of active research. In this work, we rigorously study the gradient descent dynamics for DEQs in the simple setting of linear models and single-index models, filling several gaps in the literature. We prove a conservation law for linear DEQs which implies that the parameters remain trapped on spheres during training and use this property to show that gradient flow remains well-conditioned for all time. We then prove linear convergence of gradient descent to a global minimizer for linear DEQs and deep equilibrium single-index models under appropriate initialization and with a sufficiently small step size. Finally, we validate our theoretical findings through experiments.
Authors: Artem Frenk, Roee Shraga
Abstract: Our work introduces SAVeD (Semantically Aware Version Detection), a contrastive learning-based framework for identifying versions of structured datasets without relying on metadata, labels, or integration-based assumptions. SAVeD addresses a common challenge in data science of repeated labor due to a difficulty of similar work or transformations on datasets. SAVeD employs a modified SimCLR pipeline, generating augmented table views through random transformations (e.g., row deletion, encoding perturbations). These views are embedded via a custom transformer encoder and contrasted in latent space to optimize semantic similarity. Our model learns to minimize distances between augmented views of the same dataset and maximize those between unrelated tables. We evaluate performance using validation accuracy and separation, defined respectively as the proportion of correctly classified version/non-version pairs on a hold-out set, and the difference between average similarities of versioned and non-versioned tables (defined by a benchmark, and not provided to the model). Our experiments span five canonical datasets from the Semantic Versioning in Databases Benchmark, and demonstrate substantial gains post-training. SAVeD achieves significantly higher accuracy on completely unseen tables in, and a significant boost in separation scores, confirming its capability to distinguish semantically altered versions. Compared to untrained baselines and prior state-of-the-art dataset-discovery methods like Starmie, our custom encoder achieves competitive or superior results.
Authors: PengYu Chen, Xiaohou Shi, Yuan Chang, Yan Sun, Sajal K. Das
Abstract: Anomaly prediction (AP) in multivariate time series (MTS) is crucial to ensure system dependability. Existing methods either focus solely on whether an anomaly is imminent without providing precise predictions for the future anomaly, or performing predictions directly on historical data, which is easily drowned out by the normal patterns. To address the challenges in AP task, we propose RED-F, a novel framework comprised of the Reconstruction-Elimination Model (REM) and the Dual-stream Contrastive Forecasting Model (DFM). We utilize REM to construct a baseline of normal patterns from historical data, providing a foundation for subsequent predictions of anomalies. Then DFM simultaneously predicts both the constructed normal pattern and the current window, employing a contrastive forecast that transforms the difficult AP task into a simpler, more robust task of relative trajectory comparison by computing the divergence between these two predictions. To enable the forecasting model to generate a prediction not easily obscured by normal patterns, we propose a Multi-Series Prediction (MSP) training objective to enhance its sensitivity to the current window. Extensive experiments on multiple real-world datasets demonstrate the superior capability of RED-F in anomaly prediction tasks. Our code is available at http://github.com/PenyChen/RED-F.
Authors: Sabrina Islam, Md. Atiqur Rahman, Md. Bakhtiar Hasan, Md. Hasanul Kabir
Abstract: Accurate prediction of compound potency accelerates early-stage drug discovery by prioritizing candidates for experimental testing. However, many Quantitative Structure-Activity Relationship (QSAR) approaches for this prediction are constrained by their choice of molecular representation: handcrafted descriptors capture global properties but miss local topology, graph neural networks encode structure but often lack broader chemical context, and SMILES-based language models provide contextual patterns learned from large corpora but are seldom combined with structural features. To exploit these complementary signals, we introduce Rep3Net, a unified multimodal architecture that fuses RDKit molecular descriptors, graph-derived features from a residual graph-convolutional backbone, and ChemBERTa SMILES embeddings. We evaluate Rep3Net on a curated ChEMBL subset for Human PARP1 using fivefold cross validation. Rep3Net attains an MSE of $0.83\pm0.06$, RMSE of $0.91\pm0.03$, $R^{2}=0.43\pm0.01$, and yields Pearson and Spearman correlations of $0.66\pm0.01$ and $0.67\pm0.01$, respectively, substantially improving over several strong GNN baselines. In addition, Rep3Net achieves a favorable latency-to-parameter trade-off thanks to a single-layer GCN backbone and parallel frozen encoders. Ablations show that graph topology, ChemBERTa semantics, and handcrafted descriptors each contribute complementary information, with full fusion providing the largest error reduction. These results demonstrate that multimodal representation fusion can improve potency prediction for PARP1 and provide a scalable framework for virtual screening in early-stage drug discovery.
Authors: Dongseok Kim, Hyoungsun Choi, Mohamed Jismy Aashik Rasool, Gisung Oh
Abstract: We present CLAPS, a posterior-aware conformal regression method that pairs a Last-Layer Laplace Approximation with split-conformal calibration. From the resulting Gaussian posterior, CLAPS defines a simple two-sided posterior CDF score that aligns the conformity metric with the full predictive shape, not just a point estimate. This alignment can yield substantially narrower prediction intervals at a fixed target coverage, particularly on small to medium tabular datasets where data are scarce and uncertainty modeling is informative. We also provide a lightweight diagnostic suite that separates aleatoric and epistemic components and visualizes posterior behavior, helping practitioners assess when and why intervals shrink. Across multiple benchmarks using the same MLP backbone, CLAPS achieves nominal coverage and offers the most efficient intervals on small to medium datasets with mild heterogeneity, while remaining competitive and diagnostically transparent on large-scale heterogeneous data where Normalized-CP and CQR attain the tightest intervals.
Authors: Minglai Yang, Xinyu Guo, Zhengliang Shi, Jinhe Bi, Mihai Surdeanu, Liangming Pan
Abstract: Large Language Models (LLMs) encode factual knowledge within hidden parametric spaces that are difficult to inspect or control. While Sparse Autoencoders (SAEs) can decompose hidden activations into more fine-grained, interpretable features, they often struggle to reliably align these features with human-defined concepts, resulting in entangled and distributed feature representations. To address this, we introduce AlignSAE, a method that aligns SAE features with a predefined ontology through a "pre-train, then post-train" curriculum. After an initial unsupervised training phase, we apply supervised post-training to bind specific concepts to dedicated latent slots while preserving the remaining capacity for general reconstruction. This separation creates an interpretable interface where specific concepts can be inspected and controlled without interference from unrelated features. Empirical results demonstrate that AlignSAE enables precise causal interventions, such as reliable "concept swaps", by targeting single, semantically aligned slots, and further supports multi-hop reasoning and a mechanistic probe of grokking-like generalization dynamics.
Authors: Eduardo Di Santi
Abstract: Real-world phenomena do not generate arbitrary variability: their signals concentrate on compact, low-variability subsets of functional space, enabling rapid generalization from few examples. A small child can recognize a dog after extremely limited exposure because the perceptual manifold of "dog" is compact, structured, and low-dimensional. We formalize this principle through a deterministic functional-topological framework in which the set of valid realizations produced by a physical process forms a compact subset of a Banach space, endowed with stable invariants, a finite Hausdorff radius, and an induced continuous perceptual functional. This geometry provides explicit limits on knowledge, conditions for identifiability, and guarantees for generalization from sparse evidence -- properties fundamental to both natural and artificial intelligence. Across electromechanical, electrochemical, and physiological domains, we show that real-world processes consistently generate compact perceptual manifolds with the same geometric characteristics. Their boundaries can be discovered in a fully self-supervised manner as the empirical radius saturates with increasing sampling, even when the governing equations are unknown. These results demonstrate that deterministic functional topology offers a unified mathematical foundation for perception, representation, and world-model construction. It provides a geometric explanation for why biological learners and self-supervised AI systems can generalize from few observations, and establishes compact perceptual manifolds as a fundamental building block for future AI architectures. Finally, this work unifies biological perception and modern self-supervised models under a single geometric principle: both derive their generalization ability from the compactness and invariants of real-world perceptual manifolds.
Authors: Xiaocan Li, Shiliang Wu, Zheng Shen
Abstract: Decoupled PPO has been a successful reinforcement learning (RL) algorithm to deal with the high data staleness under the asynchronous RL setting. Decoupled loss used in decoupled PPO improves coupled-loss style of algorithms' (e.g., standard PPO, GRPO) learning stability by introducing a proximal policy to decouple the off-policy correction (importance weight) from the policy update constraint (trust region). However, the proximal policy requires an extra forward pass through the model at each training step, creating a computational overhead for large language models training. We observe that since the proximal policy only serves as a trust region anchor between the behavior and target policies, we can approximate it through simple interpolation without explicit computation. We call this approach A-3PO (APproximated Proximal Policy Optimization). A-3PO eliminates this overhead, accelerating training by 1.8x speedup while maintaining comparable performance. Code \& off-the-shelf example are available at: https://github.com/inclusionAI/AReaL/blob/main/docs/algorithms/prox_approx.md
URLs: https://github.com/inclusionAI/AReaL/blob/main/docs/algorithms/prox_approx.md
Authors: Rongmei Liang, Zizheng Liu, Xiaofei Wu, Jingwen Tu
Abstract: Mathematical modelling, particularly through approaches such as structured sparse support vector machines (SS-SVM), plays a crucial role in processing data with complex feature structures, yet efficient algorithms for distributed large-scale data remain lacking. To address this gap, this paper proposes a unified optimization framework based on a consensus structure. This framework is not only applicable to various loss functions and combined regularization terms but can also be effectively extended to non-convex regularizers, demonstrating strong scalability. Building upon this framework, we develop a distributed parallel alternating direction method of multipliers (ADMM) algorithm to efficiently solve SS-SVMs under distributed data storage. To ensure convergence, we incorporate a Gaussian back-substitution technique. Additionally, for completeness, we introduce a family of sparse group Lasso support vector machine (SGL-SVM) and apply it to music information retrieval. Theoretical analysis confirms that the computational complexity of the proposed algorithm is independent of the choice of regularization terms and loss functions, underscoring the universality of the parallel approach. Experiments on both synthetic and real-world music archive datasets validate the reliability, stability, and efficiency of our algorithm.
Authors: Ci Zhang, Huayu Li, Changdi Yang, Jiangnan Xia, Yanzhi Wang, Xiaolong Ma, Jin Lu, Ao Li, Geng Yuan
Abstract: Recent studies show that using diffusion models for time series signal reconstruction holds great promise. However, such approaches remain largely unexplored in the domain of medical time series. The unique characteristics of the physiological time series signals, such as multivariate, high temporal variability, highly noisy, and artifact-prone, make deep learning-based approaches still challenging for tasks such as imputation. Hence, we propose a novel Mixture of Experts (MoE)-based noise estimator within a score-based diffusion framework. Specifically, the Receptive Field Adaptive MoE (RFAMoE) module is designed to enable each channel to adaptively select desired receptive fields throughout the diffusion process. Moreover, recent literature has found that when generating a physiological signal, performing multiple inferences and averaging the reconstructed signals can effectively reduce reconstruction errors, but at the cost of significant computational and latency overhead. We design a Fusion MoE module and innovatively leverage the nature of MoE module to generate K noise signals in parallel, fuse them using a routing mechanism, and complete signal reconstruction in a single inference step. This design not only improves performance over previous methods but also eliminates the substantial computational cost and latency associated with multiple inference processes. Extensive results demonstrate that our proposed framework consistently outperforms diffusion-based SOTA works on different tasks and datasets.
Authors: Xiang Chen, Yuling Shi, Qizhen Lan, Yuchao Qiu, Min Wang, Xiaodong Gu, Yanfu Yan
Abstract: LLM agents are widely deployed in complex interactive tasks, yet privacy constraints often preclude centralized optimization and co-evolution across dynamic environments. Despite the demonstrated success of Federated Learning (FL) on static datasets, its effectiveness in open-ended, self-evolving agent systems remains largely unexplored. In such settings, the direct application of standard FL is particularly challenging, as heterogeneous tasks and sparse, trajectory-level reward signals give rise to severe gradient instability, which undermines the global optimization process. To bridge this gap, we propose Fed-SE, a Federated Self-Evolution framework for LLM agents that establishes a local evolution-global aggregation paradigm. Locally, agents employ parameter-efficient fine-tuning on filtered, high-return trajectories to achieve stable gradient updates. Globally, Fed-SE aggregates updates within a low-rank subspace, reducing communication cost across clients. Experiments across five heterogeneous environments demonstrate that Fed-SE improves average task success rates by 10\% over the state-of-the-art FedIT, validating its effectiveness in cross-environment knowledge transfer under privacy constraints.
Authors: Haitao Lin, Peiyan Hu, Minsi Ren, Zhifeng Gao, Zhi-Ming Ma, Guolin ke, Tailin Wu, Stan Z. Li
Abstract: Recent advances in few-step diffusion models have demonstrated their efficiency and effectiveness by shortcutting the probabilistic paths of diffusion models, especially in training one-step diffusion models from scratch (\emph{a.k.a.} shortcut models). However, their theoretical derivation and practical implementation are often closely coupled, which obscures the design space. To address this, we propose a common design framework for representative shortcut models. This framework provides theoretical justification for their validity and disentangles concrete component-level choices, thereby enabling systematic identification of improvements. With our proposed improvements, the resulting one-step model achieves a new state-of-the-art FID50k of 2.85 on ImageNet-256x256 under the classifier-free guidance setting with one step generation, and further reaches FID50k of 2.53 with 2x training steps. Remarkably, the model requires no pre-training, distillation, or curriculum learning. We believe our work lowers the barrier to component-level innovation in shortcut models and facilitates principled exploration of their design space.
Authors: Mohammad Pivezhandi, Mahdi Banisharif, Saeed Bakhshan, Abusayeed Saifullah, Ali Jannesari
Abstract: Performance prediction for OpenMP workloads on heterogeneous embedded SoCs is challenging due to complex interactions between task DAG structure, control-flow irregularity, cache and branch behavior, and thermal dynamics; classical heuristics struggle under workload irregularity, tabular regressors discard structural information, and model-free RL risks overheating resource-constrained devices. We introduce GraphPerf-RT, the first surrogate that unifies task DAG topology, CFG-derived code semantics, and runtime context (per-core DVFS, thermal state, utilization) in a heterogeneous graph representation with typed edges encoding precedence, placement, and contention. Multi-task evidential heads predict makespan, energy, cache and branch misses, and utilization with calibrated uncertainty (Normal-Inverse-Gamma), enabling risk-aware scheduling that filters low-confidence rollouts. We validate GraphPerf-RT on three embedded ARM platforms (Jetson TX2, Jetson Orin NX, RUBIK Pi), achieving R^2 > 0.95 with well-calibrated uncertainty (ECE < 0.05). To demonstrate end-to-end scheduling utility, we integrate the surrogate with four RL methods on Jetson TX2: single-agent model-free (SAMFRL), single-agent model-based (SAMBRL), multi-agent model-free (MAMFRL-D3QN), and multi-agent model-based (MAMBRL-D3QN). Experiments across 5 seeds (200 episodes each) show that MAMBRL-D3QN with GraphPerf-RT as the world model achieves 66% makespan reduction (0.97 +/- 0.35s) and 82% energy reduction (0.006 +/- 0.005J) compared to model-free baselines, demonstrating that accurate, uncertainty-aware surrogates enable effective model-based planning on thermally constrained embedded systems.
Authors: Honey Singh Chauhan, Zahraa S. Abdallah
Abstract: Kernel-based methods such as Rocket are among the most effective default approaches for univariate time series classification (TSC), yet they do not perform equally well across all datasets. We revisit the long-standing intuition that different representations capture complementary structure and show that selectively fusing them can yield consistent improvements over Rocket on specific, systematically identifiable kinds of datasets. We introduce Fusion-3 (F3), a lightweight framework that adaptively fuses Rocket, SAX, and SFA representations. To understand when fusion helps, we cluster UCR datasets into six groups using meta-features capturing series length, spectral structure, roughness, and class imbalance, and treat these clusters as interpretable data-structure regimes. Our analysis shows that fusion typically outperforms strong baselines in regimes with structured variability or rich frequency content, while offering diminishing returns in highly irregular or outlier-heavy settings. To support these findings, we combine three complementary analyses: non-parametric paired statistics across datasets, ablation studies isolating the roles of individual representations, and attribution via SHAP to identify which dataset properties predict fusion gains. Sample-level case studies further reveal the underlying mechanism: fusion primarily improves performance by rescuing specific errors, with adaptive increases in frequency-domain weighting precisely where corrections occur. Using 5-fold cross-validation on the 113 UCR datasets, F3 yields small but consistent average improvements over Rocket, supported by frequentist and Bayesian evidence and accompanied by clearly identifiable failure cases. Our results show that selectively applied fusion provides dependable and interpretable extension to strong kernel-based methods, correcting their weaknesses precisely where the data support it.
Authors: Dimitrios Danopoulos, Enrico Lupi, Chang Sun, Sebastian Dittmeier, Michael Kagan, Vladimir Loncar, Maurizio Pierini
Abstract: Efficient AI inference on AMD's Versal AI Engine (AIE) is challenging due to tightly coupled VLIW execution, explicit datapaths, and local memory management. Prior work focused on first-generation AIE kernel optimizations, without tackling full neural network execution across the 2D array. In this work, we present AIE4ML, the first comprehensive framework for converting AI models automatically into optimized firmware targeting the AIE-ML generation devices, also with forward compatibility for the newer AIE-MLv2 architecture. At the single-kernel level, we attain performance close to the architectural peak. At the graph and system levels, we provide a structured parallelization method that can scale across the 2D AIE-ML fabric and exploit its dedicated memory tiles to stay entirely on-chip throughout the model execution. As a demonstration, we designed a generalized and highly efficient linear-layer implementation with intrinsic support for fused bias addition and ReLU activation. Also, as our framework necessitates the generation of multi-layer implementations, our approach systematically derives deterministic, compact, and topology-optimized placements tailored to the physical 2D grid of the device through a novel graph placement and search algorithm. Finally, the framework seamlessly accepts quantized models imported from high-level tools such as hls4ml or PyTorch while preserving bit-exactness. In layer scaling benchmarks, we achieve up to 98.6% efficiency relative to the single-kernel baseline, utilizing 296 of 304 AIE tiles (97.4%) of the device with entirely on-chip data movement. With evaluations across real-world model topologies, we demonstrate that AIE4ML delivers GPU-class throughput under microsecond latency constraints, making it a practical companion for ultra-low-latency environments such as trigger systems in particle physics experiments.
Authors: Yizhou Zhang
Abstract: Empirical power--law scaling has been widely observed across modern deep learning systems, yet its theoretical origins and scope of validity remain incompletely understood. The Generalized Resolution--Shell Dynamics (GRSD) framework models learning as spectral energy transport across logarithmic resolution shells, providing a coarse--grained dynamical description of training. Within GRSD, power--law scaling corresponds to a particularly simple renormalized shell dynamics; however, such behavior is not automatic and requires additional structural properties of the learning process. In this work, we identify a set of sufficient conditions under which the GRSD shell dynamics admits a renormalizable coarse--grained description. These conditions constrain the learning configuration at multiple levels, including boundedness of gradient propagation in the computation graph, weak functional incoherence at initialization, controlled Jacobian evolution along training, and log--shift invariance of renormalized shell couplings. We further show that power--law scaling does not follow from renormalizability alone, but instead arises as a rigidity consequence: once log--shift invariance is combined with the intrinsic time--rescaling covariance of gradient flow, the renormalized GRSD velocity field is forced into a power--law form.
Authors: Rui Pan, Zhuofu Chen, Hongyi Liu, Arvind Krishnamurthy, Ravi Netravali
Abstract: Diffusion Large Language Models (dLLMs) offer fast, parallel token generation, but their standalone use is plagued by an inherent efficiency-quality tradeoff. We show that, if carefully applied, the attributes of dLLMs can actually be a strength for drafters in speculative decoding with autoregressive (AR) verifiers. Our core insight is that dLLM's speed from parallel decoding drastically lowers the risk of costly rejections, providing a practical mechanism to effectively realize the (elusive) lengthy drafts that lead to large speedups with speculative decoding. We present FailFast, a dLLM-based speculative decoding framework that realizes this approach by dynamically adapting its speculation length. It "fails fast" by spending minimal compute in hard-to-speculate regions to shrink speculation latency and "wins big" by aggressively extending draft lengths in easier regions to reduce verification latency (in many cases, speculating and accepting 70 tokens at a time!). Without any fine-tuning, FailFast delivers lossless acceleration of AR LLMs and achieves up to 4.9$\times$ speedup over vanilla decoding, 1.7$\times$ over the best naive dLLM drafter, and 2.0$\times$ over EAGLE-3 across diverse models and workloads. We open-source FailFast at https://github.com/ruipeterpan/failfast.
Authors: Mikhail Lazarev, Andrey Ustyuzhanin
Abstract: Machine learning models have become firmly established across all scientific fields. Extracting features from data and making inferences based on them with neural network models often yields high accuracy; however, this approach has several drawbacks. Symbolic regression is a powerful technique for discovering analytical equations that describe data, providing interpretable and generalizable models capable of predicting unseen data. Symbolic regression methods have gained new momentum with the advancement of neural network technologies and offer several advantages, the main one being the interpretability of results. In this work, we examined the application of the deep symbolic regression algorithm SEGVAE to determine the properties of two-dimensional materials with defects. Comparing the results with state-of-the-art graph neural network-based methods shows comparable or, in some cases, even identical outcomes. We also discuss the applicability of this class of methods in natural sciences.
Authors: Andres M Bran, Tong Xie, Shai Pranesh, Jeffrey Meng, Xuan Vu Nguyen, Jeremy Goumaz, David Ming Segura, Ruizhi Xu, Dongzhan Zhou, Wenjie Zhang, Bram Hoex, Philippe Schwaller
Abstract: Large Language Models can develop reasoning capabilities through online fine-tuning with rule-based rewards. However, recent studies reveal a critical constraint: reinforcement learning succeeds only when the base model already assigns non-negligible probability to correct answers -- a property we term 'latent solvability'. This work investigates the emergence of chemical reasoning capabilities and what these prerequisites mean for chemistry. We identify two necessary conditions for RL-based chemical reasoning: 1) Symbolic competence, and 2) Latent chemical knowledge. We propose mid-stage scientific training (MiST): a set of mid-stage training techniques to satisfy these, including data-mixing with SMILES/CIF-aware pre-processing, continued pre-training on 2.9B tokens, and supervised fine-tuning on 1B tokens. These steps raise the latent-solvability score on 3B and 7B models by up to 1.8x, and enable RL to lift top-1 accuracy from 10.9 to 63.9% on organic reaction naming, and from 40.6 to 67.4% on inorganic material generation. Similar results are observed for other challenging chemical tasks, while producing interpretable reasoning traces. Our results define clear prerequisites for chemical reasoning training and highlight the broader role of mid-stage training in unlocking reasoning capabilities.
Authors: Sravan Karthick T
Abstract: Bitcoin price forecasting is characterized by extreme volatility and non-stationarity, often defying traditional univariate time-series models over long horizons. This paper addresses a critical gap by integrating Global M2 Liquidity, aggregated from 18 major economies, as a leading exogenous variable with a 12-week lag structure. Using the TimeXer architecture, we compare a liquidity-conditioned forecasting model (TimeXer-Exog) against state-of-the-art benchmarks including LSTM, N-BEATS, PatchTST, and a standard univariate TimeXer. Experiments conducted on daily Bitcoin price data from January 2020 to August 2025 demonstrate that explicit macroeconomic conditioning significantly stabilizes long-horizon forecasts. At a 70-day forecast horizon, the proposed TimeXer-Exog model achieves a mean squared error (MSE) 1.08e8, outperforming the univariate TimeXer baseline by over 89 percent. These results highlight that conditioning deep learning models on global liquidity provides substantial improvements in long-horizon Bitcoin price forecasting.
Authors: Advait Gadhikar, Riccardo Grazzi, James Hensman
Abstract: The presence of outliers in Large Language Models (LLMs) weights and activations makes them difficult to quantize. Recent work has leveraged rotations to mitigate these outliers. In this work, we propose methods that learn fusible rotations by minimizing principled and cheap proxy objectives to the weight quantization error. We primarily focus on GPTQ as the quantization method. Our main method is OptRot, which reduces weight outliers simply by minimizing the element-wise fourth power of the rotated weights. We show that OptRot outperforms both Hadamard rotations and more expensive, data-dependent methods like SpinQuant and OSTQuant for weight quantization. It also improves activation quantization in the W4A8 setting. We also propose a data-dependent method, OptRot$^{+}$, that further improves performance by incorporating information on the activation covariance. In the W4A4 setting, we see that both OptRot and OptRot$^{+}$ perform worse, highlighting a trade-off between weight and activation quantization.
Authors: Raul D. Steleac, Mohan Sridharan, David Abel
Abstract: Temporally extended actions improve the ability to explore and plan in single-agent settings. In multi-agent settings, the exponential growth of the joint state space with the number of agents makes coordinated behaviours even more valuable. Yet, this same exponential growth renders the design of multi-agent options particularly challenging. Existing multi-agent option discovery methods often sacrifice coordination by producing loosely coupled or fully independent behaviours. Toward addressing these limitations, we describe a novel approach for multi-agent option discovery. Specifically, we propose a joint-state abstraction that compresses the state space while preserving the information necessary to discover strongly coordinated behaviours. Our approach builds on the inductive bias that synchronisation over agent states provides a natural foundation for coordination in the absence of explicit objectives. We first approximate a fictitious state of maximal alignment with the team, the \textit{Fermat} state, and use it to define a measure of \textit{spreadness}, capturing team-level misalignment on each individual state dimension. Building on this representation, we then employ a neural graph Laplacian estimator to derive options that capture state synchronisation patterns between agents. We evaluate the resulting options across multiple scenarios in two multi-agent domains, showing that they yield stronger downstream coordination capabilities compared to alternative option discovery methods.
Authors: Xingsheng Chen, Regina Zhang, Bo Gao, Xingwei He, Xiaofeng Liu, Pietro Lio, Kwok-Yan Lam, Siu-Ming Yiu
Abstract: Time series prediction plays a pivotal role across diverse domains such as finance, healthcare, energy systems, and environmental modeling. However, existing approaches often struggle to balance efficiency, scalability, and accuracy, particularly when handling long-range dependencies and irregularly sampled data. To address these challenges, we propose MODE, a unified framework that integrates Low-Rank Neural Ordinary Differential Equations (Neural ODEs) with an Enhanced Mamba architecture. As illustrated in our framework, the input sequence is first transformed by a Linear Tokenization Layer and then processed through multiple Mamba Encoder blocks, each equipped with an Enhanced Mamba Layer that employs Causal Convolution, SiLU activation, and a Low-Rank Neural ODE enhancement to efficiently capture temporal dynamics. This low-rank formulation reduces computational overhead while maintaining expressive power. Furthermore, a segmented selective scanning mechanism, inspired by pseudo-ODE dynamics, adaptively focuses on salient subsequences to improve scalability and long-range sequence modeling. Extensive experiments on benchmark datasets demonstrate that MODE surpasses existing baselines in both predictive accuracy and computational efficiency. Overall, our contributions include: (1) a unified and efficient architecture for long-term time series modeling, (2) integration of Mamba's selective scanning with low-rank Neural ODEs for enhanced temporal representation, and (3) substantial improvements in efficiency and scalability enabled by low-rank approximation and dynamic selective scanning.
Authors: Nobuyuki Ota
Abstract: Understanding cellular mechanisms requires integrating information across DNA, RNA, and protein - the three molecular systems linked by the Central Dogma of molecular biology. While domain-specific foundation models have achieved success for each modality individually, they remain isolated, limiting our ability to model integrated cellular processes. Here we present the Central Dogma Transformer (CDT), an architecture that integrates pre-trained language models for DNA, RNA, and protein following the directional logic of the Central Dogma. CDT employs directional cross-attention mechanisms - DNA-to-RNA attention models transcriptional regulation, while RNA-to-Protein attention models translational relationships - producing a unified Virtual Cell Embedding that integrates all three modalities. We validate CDT v1 - a proof-of-concept implementation using fixed (non-cell-specific) RNA and protein embeddings - on CRISPRi enhancer perturbation data from K562 cells, achieving a Pearson correlation of 0.503, representing 63% of the theoretical ceiling set by cross-experiment variability (r = 0.797). Attention and gradient analyses provide complementary interpretive windows: in detailed case studies, these approaches highlight largely distinct genomic regions, with gradient analysis identifying a CTCF binding site that Hi-C data showed as physically contacting both enhancer and target gene. These results suggest that AI architectures aligned with biological information flow can achieve both predictive accuracy and mechanistic interpretability.
Authors: Jian Feng, Zhihong Huang
Abstract: Fine-tuning large language models (LLMs) with zeroth-order (ZO) optimization reduces memory by approximating gradients through function evaluations. However, existing methods essentially perform updates in a one-dimensional space, and suffer from collapse or substantial performance degradation under low-precision training. We introduce BSZO, an adaptive \textbf{B}ayesian \textbf{S}ubspace \textbf{Z}eroth-Order \textbf{O}ptimizer, which applies Kalman filtering to combine finite-difference information across multiple perturbation directions within a subspace. By treating each finite-difference measurement as a noisy observation, BSZO builds a posterior distribution over the subspace-projected gradient and updates it through Bayesian inference, with a residual-based adaptive mechanism to adapt to noise variations. Theoretical analysis shows that BSZO improves the convergence rate by a factor of $k/\gamma$ compared to standard ZO methods. Experiments on RoBERTa, Mistral, and OPT models show that BSZO outperforms the baselines across various tasks, achieving up to 6.67\% absolute average improvement on OPT-13B while remaining robust under fp16/bf16 precision and keeping memory usage close to inference-only baselines (1.00$\times$--1.08$\times$ of MeZO).
Authors: Yash Thesia, Meera Suthar
Abstract: Model stealing attacks pose an existential threat to Machine Learning as a Service (MLaaS), allowing adversaries to replicate proprietary models for a fraction of their training cost. While Data-Free Model Extraction (DFME) has emerged as a stealthy vector, it remains fundamentally constrained by the "Cold Start" problem: GAN-based adversaries waste thousands of queries converging from random noise to meaningful data. We propose DiMEx, a framework that weaponizes the rich semantic priors of pre-trained Latent Diffusion Models to bypass this initialization barrier entirely. By employing Random Embedding Bayesian Optimization (REMBO) within the generator's latent space, DiMEx synthesizes high-fidelity queries immediately, achieving 52.1 percent agreement on SVHN with just 2,000 queries - outperforming state-of-the-art GAN baselines by over 16 percent. To counter this highly semantic threat, we introduce the Hybrid Stateful Ensemble (HSE) defense, which identifies the unique "optimization trajectory" of latent-space attacks. Our results demonstrate that while DiMEx evades static distribution detectors, HSE exploits this temporal signature to suppress attack success rates to 21.6 percent with negligible latency.
Authors: Zhuoyang Jiang, Yaosen Min, Peiran Jin, Lei Chen
Abstract: We present Connection-Aware Motif Sequencing (CamS), a graph-to-sequence representation that enables decoder-only Transformers to learn molecular graphs via standard next-token prediction (NTP). For molecular property prediction, SMILES-based NTP scales well but lacks explicit topology, whereas graph-native masked modeling captures connectivity but risks disrupting the pivotal chemical details (e.g., activity cliffs). CamS bridges this gap by serializing molecular graphs into structure-rich causal sequences. CamS first mines data-driven connection-aware motifs. It then serializes motifs via scaffold-rooted breadth-first search (BFS) to establish a stable core-to-periphery order. Crucially, CamS enables hierarchical modeling by concatenating sequences from fine to coarse motif scales, allowing the model to condition global scaffolds on dense, uncorrupted local structural evidence. We instantiate CamS-LLaMA by pre-training a vanilla LLaMA backbone on CamS sequences. It achieves state-of-the-art performance on MoleculeNet and the activity-cliff benchmark MoleculeACE, outperforming both SMILES-based language models and strong graph baselines. Interpretability analysis confirms that our multi-scale causal serialization effectively drives attention toward cliff-determining differences.
Authors: Kun Zhao, Siyuan Dai, Pan Wang, Jifeng Song, Hui Ji, Chenghua Lin, Liang Zhan, Haoteng Tang
Abstract: Multimodal Large Language Models (MLLMs) have shown strong potential for radiology report generation, yet their clinical translation is hindered by architectural heterogeneity and the prevalence of factual hallucinations. Standard supervised fine-tuning often fails to strictly align linguistic outputs with visual evidence, while existing reinforcement learning approaches struggle with either prohibitive computational costs or limited exploration. To address these challenges, we propose a comprehensive framework for self-consistent radiology report generation. First, we conduct a systematic evaluation to identify optimal vision encoder and LLM backbone configurations for medical imaging. Building on this foundation, we introduce a novel "Reason-then-Summarize" architecture optimized via Group Relative Policy Optimization (GRPO). This framework restructures generation into two distinct components: a think block for detailed findings and an answer block for structured disease labels. By utilizing a multi-dimensional composite reward function, we explicitly penalize logical discrepancies between the generated narrative and the final diagnosis. Extensive experiments on the MIMIC-CXR benchmark demonstrate that our method achieves state-of-the-art performance in clinical efficacy metrics and significantly reduces hallucinations compared to strong supervised baselines.
Authors: Hao Tang, Hao Chen, Hao Li, Chao Li
Abstract: Spherical deep learning has been widely applied to a broad range of real-world problems. Existing approaches often face challenges in balancing strong spherical geometric inductive biases with the need to model real-world heterogeneity. To solve this while retaining spherical geometry, we first introduce a designable Green's function framework (DGF) to provide new spherical operator solution strategy: Design systematic Green's functions under rotational group. Based on DGF, to model biomedical heterogeneity, we propose Green's-Function Spherical Neural Operator (GSNO) fusing 3 operator solutions: (1) Equivariant Solution derived from Equivariant Green's Function for symmetry-consistent modeling; (2) Invariant Solution derived from Invariant Green's Function to eliminate nuisance heterogeneity, e.g., consistent background field; (3) Anisotropic Solution derived from Anisotropic Green's Function to model anisotropic systems, especially fibers with preferred direction. Therefore, the resulting model, GSNO can adapt to real-world heterogeneous systems with nuisance variability and anisotropy while retaining spectral efficiency. Evaluations on spherical MNIST, Shallow Water Equation, diffusion MRI fiber prediction, cortical parcellation and molecule structure modeling demonstrate the superiority of GSNO.
Authors: Paul Pu Liang
Abstract: Our experience of the world is multisensory, spanning a synthesis of language, sight, sound, touch, taste, and smell. Yet, artificial intelligence has primarily advanced in digital modalities like text, vision, and audio. This paper outlines a research vision for multisensory artificial intelligence over the next decade. This new set of technologies can change how humans and AI experience and interact with one another, by connecting AI to the human senses and a rich spectrum of signals from physiological and tactile cues on the body, to physical and social signals in homes, cities, and the environment. We outline how this field must advance through three interrelated themes of sensing, science, and synergy. Firstly, research in sensing should extend how AI captures the world in richer ways beyond the digital medium. Secondly, developing a principled science for quantifying multimodal heterogeneity and interactions, developing unified modeling architectures and representations, and understanding cross-modal transfer. Finally, we present new technical challenges to learn synergy between modalities and between humans and AI, covering multisensory integration, alignment, reasoning, generation, generalization, and experience. Accompanying this vision paper are a series of projects, resources, and demos of latest advances from the Multisensory Intelligence group at the MIT Media Lab, see https://mit-mi.github.io/.
Authors: Sidney Shapiro, Burhanuddin Panvelwala
Abstract: Reproducibility remains a persistent challenge in forecasting research and practice, particularly in business and financial analytics, where forecasts inform high-stakes decisions. Traditional forecasting methods, while theoretically interpretable, often require extensive manual tuning and are difficult to replicate in proprietary environments. Machine learning approaches offer predictive flexibility but introduce challenges related to interpretability, stochastic training procedures, and cross-environment reproducibility. This paper examines Prophet, an open-source forecasting framework developed by Meta, as a reproducibility-enabling solution that balances interpretability, standardized workflows, and accessibility. Rather than proposing a new algorithm, this study evaluates how Prophet's additive structure, open-source implementation, and standardized workflow contribute to transparent and replicable forecasting practice. Using publicly available financial and retail datasets, we compare the performance and interpretability of Prophet with multiple ARIMA specifications (auto-selected, manually specified, and seasonal variants) and Random Forest, under a controlled and fully documented experimental design. This multi-model comparison provides a robust assessment of Prophet's relative performance and reproducibility advantages. Through concrete Python examples, we demonstrate how Prophet facilitates efficient forecasting workflows and integration with analytical pipelines. The study positions Prophet within the broader context of reproducible research. It highlights Prophet's role as a methodological building block that supports verification, auditability, and methodological rigor. This work provides researchers and practitioners with a practical reference framework for reproducible forecasting in Python-based research workflows.
Authors: Juaren Steiger, Bin Li
Abstract: The constrained combinatorial multi-armed bandit model has been widely employed to solve problems in wireless networking and related areas, including the problem of wireless scheduling for throughput optimization under unknown channel conditions. Most work in this area uses an algorithm design strategy that combines a bandit learning algorithm with the virtual queue technique to track the throughput constraint violation. These algorithms seek to minimize the virtual queue length in their algorithm design. However, in networks where channel conditions change abruptly, the resulting constraints may become infeasible, leading to unbounded growth in virtual queue lengths. In this paper, we make the key observation that the dynamics of the head-of-line age, i.e. the age of the oldest packet in the virtual queue, make it more robust when used in algorithm design compared to the virtual queue length. We therefore design a learning-based scheduling policy that uses the head-of-line age in place of the virtual queue length. We show that our policy matches state-of-the-art performance under i.i.d. network conditions. Crucially, we also show that the system remains stable even under abrupt changes in channel conditions and can rapidly recover from periods of constraint infeasibility.
Authors: Yifan Chen, Yun Yang
Abstract: With appropriately chosen sampling probabilities, sampling-based random projection can be used to implement large-scale statistical methods, substantially reducing computational cost while maintaining low statistical error. However, computing optimal sampling probabilities is often itself expensive, and in practice one typically resorts to suboptimal schemes. This generally leads to increased time and space costs, as more subsamples are required and the resulting projection matrices become larger, thereby making the inference procedure more computationally demanding. In this paper, we extend the framework of sampling-based random projection and propose a new projection method, \emph{accumulative sub-sampling}. By carefully accumulating multiple such projections, accumulative sub-sampling improves statistical efficiency while controlling the effective matrix size throughout the statistical computation. On the theoretical side, we quantify how the quality of the subsampling scheme affects the error in approximating matrix products and positive semidefinite matrices, and show how the proposed accumulation strategy mitigates this effect. Moreover, we apply our method to statistical models involving intensive matrix operations, such as eigendecomposition in spectral clustering and matrix inversion in kernel ridge regression, and demonstrate that reducing the effective matrix size leads to substantial computational savings. Numerical experiments across a range of problems further show that our approach consistently improves computational efficiency compared to existing random projection baselines under suboptimal sampling schemes.
Authors: Yueqi Cao, Anthea Monod
Abstract: Persistent homology is an important methodology in topological data analysis which adapts theory from algebraic topology to data settings. Computing persistent homology produces persistence diagrams, which have been successfully used in diverse domains. Despite its widespread use, persistent homology is simply impossible to compute when a dataset is very large. We study a statistical approach to the problem of computing persistent homology for massive datasets using a multiple subsampling framework and extend it to three summaries of persistent homology: H\"{o}lder continuous vectorizations of persistence diagrams; the alternative representation as persistence measures; and standard persistence diagrams. Specifically, we derive finite sample convergence rates for empirical means for persistent homology and practical guidance on interpreting and tuning parameters. We validate our approach through extensive experiments on both synthetic and real-world data. We demonstrate the performance of multiple subsampling in a permutation test to analyze the topological structure of Poincar\'{e} embeddings of large lexical databases.
Authors: {\L}ukasz D\k{e}bowski
Abstract: We construct multiperiodic processes -- a simple example of stationary ergodic (but not mixing) processes over natural numbers that enjoy the vanishing entropy rate under a mild condition. Multiperiodic processes are supported on randomly shifted deterministic sequences called multiperiodic sequences, which can be efficiently generated using an algorithm called the Infinite Clock. Under a suitable parameterization, multiperiodic sequences exhibit relative frequencies of particular numbers given by Zipf's law. Exactly in the same setting, the respective multiperiodic processes satisfy an asymptotic power-law growth of block entropy, called Hilberg's law. Hilberg's law is deemed to hold for statistical language models, in particular.
Authors: Yifan Peng, Yian Chen, E. Miles Stoudenmire, Yuehaw Khoo
Abstract: We propose a hierarchical tensor-network approach for approximating high-dimensional probability density via empirical distribution. This leverages randomized singular value decomposition (SVD) techniques and involves solving linear equations for tensor cores in this tensor network. The complexity of the resulting algorithm scales linearly in the dimension of the high-dimensional density. An analysis of estimation error demonstrates the effectiveness of this method through several numerical experiments.
Authors: Liam Hodgkinson, Chris van der Heide, Robert Salomone, Fred Roosta, Michael W. Mahoney
Abstract: The problem of model selection is considered for the setting of interpolating estimators, where the number of model parameters exceeds the size of the dataset. Classical information criteria typically consider the large-data limit, penalizing model size. However, these criteria are not appropriate in modern settings where overparameterized models tend to perform well. For any overparameterized model, we show that there exists a dual underparameterized model that possesses the same marginal likelihood, thus establishing a form of Bayesian duality. This enables more classical methods to be used in the overparameterized setting, revealing the Interpolating Information Criterion, a measure of model quality that naturally incorporates the choice of prior into the model selection. Our new information criterion accounts for prior misspecification, geometric and spectral properties of the model, and is numerically consistent with known empirical and theoretical behavior in this regime.
Authors: Kei Ishikawa, Niao He, Takafumi Kanamori
Abstract: We study policy evaluation of offline contextual bandits subject to unobserved confounders. Sensitivity analysis methods are commonly used to estimate the policy value under the worst-case confounding over a given uncertainty set. However, existing work often resorts to some coarse relaxation of the uncertainty set for the sake of tractability, leading to overly conservative estimation of the policy value. In this paper, we propose a general estimator that provides a sharp lower bound of the policy value using convex programming. The generality of our estimator enables various extensions such as sensitivity analysis with f-divergence, model selection with cross validation and information criterion, and robust policy learning with the sharp lower bound. Furthermore, our estimation method can be reformulated as an empirical risk minimization problem thanks to the strong duality, which enables us to provide strong theoretical guarantees of the proposed estimator using techniques of the M-estimation.
Authors: Lei Shi, Jia-Qi Yang
Abstract: This study investigates leveraging stochastic gradient descent (SGD) to learn operators between general Hilbert spaces. We propose weak and strong regularity conditions for the target operator to depict its intrinsic structure and complexity. Under these conditions, we establish upper bounds for convergence rates of the SGD algorithm and conduct a minimax lower bound analysis, further illustrating that our convergence analysis and regularity conditions quantitatively characterize the tractability of solving operator learning problems using the SGD algorithm. It is crucial to highlight that our convergence analysis is still valid for nonlinear operator learning. We show that the SGD estimator will converge to the best linear approximation of the nonlinear target operator. Moreover, applying our analysis to operator learning problems based on vector-valued and real-valued reproducing kernel Hilbert spaces yields new convergence results, thereby refining the conclusions of existing literature.
Authors: Philipp Liznerski, Saurabh Varshneya, Ece Calikus, Puyu Wang, Alexander Bartscher, Sebastian Josef Vollmer, Sophie Fellenz, Marius Kloft
Abstract: Deep learning-based methods have achieved a breakthrough in image anomaly detection, but their complexity introduces a considerable challenge to understanding why an instance is predicted to be anomalous. We introduce a novel explanation method that generates multiple alternative modifications for each anomaly, capturing diverse concepts of anomalousness. Each modification is trained to be perceived as normal by the anomaly detector. The method provides a semantic explanation of the mechanism that triggered the detector, allowing users to explore ``what-if scenarios.'' Qualitative and quantitative analyses across various image datasets demonstrate that applying this method to state-of-the-art detectors provides high-quality semantic explanations.
Authors: Benjamin Ellenberger, Paul Haider, Jakob Jordan, Kevin Max, Ismael Jaras, Laura Kriener, Federico Benitez, Mihai A. Petrovici
Abstract: How physical networks of neurons, bound by spatio-temporal locality constraints, can perform efficient credit assignment, remains, to a large extent, an open question. In machine learning, the answer is almost universally given by the error backpropagation algorithm, through both space and time. However, this algorithm is well-known to rely on biologically implausible assumptions, in particular with respect to spatio-temporal (non-)locality. Alternative forward-propagation models such as real-time recurrent learning only partially solve the locality problem, but only at the cost of scaling, due to prohibitive storage requirements. We introduce Generalized Latent Equilibrium (GLE), a computational framework for fully local spatio-temporal credit assignment in physical, dynamical networks of neurons. We start by defining an energy based on neuron-local mismatches, from which we derive both neuronal dynamics via stationarity and parameter dynamics via gradient descent. The resulting dynamics can be interpreted as a real-time, biologically plausible approximation of backpropagation through space and time in deep cortical networks with continuous-time neuronal dynamics and continuously active, local synaptic plasticity. In particular, GLE exploits the morphology of dendritic trees to enable more complex information storage and processing in single neurons, as well as the ability of biological neurons to phase-shift their output rate with respect to their membrane potential, which is essential in both directions of information propagation. For the forward computation, it enables the mapping of time-continuous inputs to neuronal space, effectively performing a spatio-temporal convolution. For the backward computation, it permits the temporal inversion of feedback signals, which consequently approximate the adjoint variables necessary for useful parameter updates.
Authors: Seong Jin Lee, Will Wei Sun, Yufeng Liu
Abstract: As e-commerce expands, delivering real-time personalized recommendations from vast catalogs poses a critical challenge for retail platforms. Maximizing revenue requires careful consideration of both individual customer characteristics and available item features to continuously optimize assortments over time. In this paper, we consider the dynamic assortment problem with dual contexts -- user and item features. In high-dimensional scenarios, the quadratic growth of dimensions complicates computation and estimation. To tackle this challenge, we introduce a new low-rank dynamic assortment model to transform this problem into a manageable scale. Then we propose an efficient algorithm that estimates the intrinsic subspaces and utilizes the upper confidence bound approach to address the exploration-exploitation trade-off in online decision making. Theoretically, we establish a regret bound of $\tilde{O}((d_1+d_2)r\sqrt{T})$, where $d_1, d_2$ represent the dimensions of the user and item features respectively, $r$ is the rank of the parameter matrix, and $T$ denotes the time horizon. This bound represents a substantial improvement over prior literature, achieved by leveraging the low-rank structure. Extensive simulations and an application to the Expedia hotel recommendation dataset further demonstrate the advantages of our proposed method.
Authors: Etienne Lempereur, St\'ephane Mallat
Abstract: Finding low-dimensional interpretable models of complex physical fields such as turbulence remains an open question, 80 years after the pioneer work of Kolmogorov. Estimating high-dimensional probability distributions from data samples suffers from an optimization and an approximation curse of dimensionality. It may be avoided by following a hierarchic probability flow from coarse to fine scales. This inverse renormalization group is defined by conditional probabilities across scales, renormalized in a wavelet basis. For a $\vvarphi^4$ scalar potential, sampling these hierarchic models avoids the critical slowing down at the phase transition. In a well chosen wavelet basis, conditional probabilities can be captured with low dimensional parametric models, because interactions between wavelet coefficients are local in space and scales. An outstanding issue is also to approximate non-Gaussian fields having long-range interactions in space and across scales. We introduce low-dimensional models of wavelet conditional probabilities with the scattering covariance. It is calculated with a second wavelet transform, which defines interactions over two hierarchies of scales. We estimate and sample these wavelet scattering models to generate 2D vorticity fields of turbulence, and images of dark matter densities.
Authors: Tim Gyger, Reinhard Furrer, Fabio Sigrist
Abstract: Gaussian processes are flexible probabilistic regression models which are widely used in statistics and machine learning. However, a drawback is their limited scalability to large data sets. To alleviate this, full-scale approximations (FSAs) combine predictive process methods and covariance tapering, thus approximating both global and local structures. We show how iterative methods can be used to reduce computational costs in calculating likelihoods, gradients, and predictive distributions with FSAs. In particular, we introduce a novel preconditioner and show theoretically and empirically that it accelerates the conjugate gradient method's convergence speed and mitigates its sensitivity with respect to the FSA parameters and the eigenvalue structure of the original covariance matrix, and we demonstrate empirically that it outperforms a state-of-the-art pivoted Cholesky preconditioner. Furthermore, we introduce an accurate and fast way to calculate predictive variances using stochastic simulation and iterative methods. In addition, we show how our newly proposed fully independent training conditional (FITC) preconditioner can also be used in iterative methods for Vecchia approximations. In our experiments, it outperforms existing state-of-the-art preconditioners for Vecchia approximations. All methods are implemented in a free C++ software library with high-level Python and R packages.
Authors: Xiaoxiong Zhang, Zhiwei Zeng, Xin Zhou, Chunyan Miao
Abstract: Federated Knowledge Graph Embedding (FKGE) aims to facilitate collaborative learning of entity and relation embeddings from distributed Knowledge Graphs (KGs) across multiple clients, while preserving data privacy. Training FKGE models with higher dimensions is typically favored due to their potential for achieving superior performance. However, high-dimensional embeddings present significant challenges in terms of storage resource and inference speed. Unlike traditional KG embedding methods, FKGE involves multiple client-server communication rounds, where communication efficiency is critical. Existing embedding compression methods for traditional KGs may not be directly applicable to FKGE as they often require multiple model trainings which potentially incur substantial communication costs. In this paper, we propose a light-weight component based on Knowledge Distillation (KD) which is titled FedKD and tailored specifically for FKGE methods. During client-side local training, FedKD facilitates the low-dimensional student model to mimic the score distribution of triples from the high-dimensional teacher model using KL divergence loss. Unlike traditional KD way, FedKD adaptively learns a temperature to scale the score of positive triples and separately adjusts the scores of corresponding negative triples using a predefined temperature, thereby mitigating teacher over-confidence issue. Furthermore, we dynamically adjust the weight of KD loss to optimize the training process. Extensive experiments on three datasets support the effectiveness of FedKD.
Authors: Xiuyuan Cheng, Tingnan Gong, Yao Xie
Abstract: Point processes are widely used statistical models for continuous-time discrete event data, such as medical records, crime reports, and social network interactions, to capture the influence of historical events on future occurrences. In many applications, however, event times are not observed exactly, motivating the need to incorporate time uncertainty into point process modeling. In this work, we introduce a framework for modeling time-uncertain self-exciting point processes, known as Hawkes processes, possibly defined over a network. We begin by formulating the model in continuous time under assumptions motivated by real-world scenarios. By imposing a time grid, we obtain a discrete-time model that facilitates inference and enables computation via first-order optimization methods such as gradient descent and variational inequality (VI). We establish a parameter recovery guarantee for VI inference with an $O(1/k)$ convergence rate using $k$ steps. Our framework accommodates non-stationary processes by representing the influence kernel as a matrix (or tensor on a network), while also encompassing stationary processes, such as the classical Hawkes process, as a special case. Empirically, we demonstrate that the proposed approach outperforms existing baselines on both simulated and real-world datasets, including the sepsis-associated derangement prediction challenge and the Atlanta Police Crime Dataset.
Authors: Matteo Lapucci, Davide Pucci
Abstract: In this work, we address unconstrained finite-sum optimization problems, with particular focus on instances originating in large scale deep learning scenarios. Our main interest lies in the exploration of the relationship between recent line search approaches for stochastic optimization in the overparametrized regime and momentum directions. First, we point out that combining these two elements with computational benefits is not straightforward. To this aim, we propose a solution based on mini-batch persistency. We then introduce an algorithmic framework that exploits a mix of data persistency, conjugate-gradient type rules for the definition of the momentum parameter and stochastic line searches. The resulting algorithm provably possesses convergence properties under suitable assumptions and is empirically shown to outperform other popular methods from the literature, obtaining state-of-the-art results in both convex and nonconvex large scale training problems.
Authors: Seyed Ali Farokh, Hossein Zeinali
Abstract: This paper presents our submission to the Iranian division of the Text-Dependent Speaker Verification Challenge (TdSV) 2024. Conventional TdSV approaches typically jointly model speaker and linguistic features, requiring unsegmented inputs during training and incurring high computational costs. Additionally, these methods often fine-tune large-scale pre-trained speaker embedding models on the target domain dataset, which may compromise the pre-trained models' original ability to capture speaker-specific characteristics. To overcome these limitations, we employ a TdSV system that utilizes two pre-trained models independently and demonstrate that, by leveraging pre-trained models with targeted domain adaptation, competitive results can be achieved while avoiding the substantial computational costs associated with joint fine-tuning on unsegmented inputs in conventional approaches. Our best system reached a MinDCF of 0.0358 on the evaluation subset and secured first place in the challenge.
Authors: Yuma Toji, Jun Takahashi, Vwani Roychowdhury, Hideyuki Miyahara
Abstract: Several power-law critical properties involving different statistics in natural languages -- reminiscent of scaling properties of physical systems at or near phase transitions -- have been documented for decades. The recent rise of large language models has added further evidence and excitement by providing intriguing similarities with notions in physics such as scaling laws and emergent abilities. However, specific instances of classes of generative language models that exhibit phase transitions, as understood by the statistical physics community, are lacking. In this work, inspired by the one-dimensional Potts model in statistical physics, we construct a simple probabilistic language model that falls under the class of context-sensitive grammars, which we call the context-sensitive random language model, and numerically demonstrate an unambiguous phase transition in the framework of a natural language model. We explicitly show that a precisely defined order parameter -- that captures symbol frequency biases in the sentences generated by the language model -- changes from strictly zero to a strictly nonzero value (in the infinite-length limit of sentences), implying a mathematical singularity arising when tuning the parameter of the stochastic language model we consider. Furthermore, we identify the phase transition as a variant of the Berezinskii--Kosterlitz--Thouless (BKT) transition, which is known to exhibit critical properties not only at the transition point but also in the entire phase. This finding leads to the possibility that critical properties in natural languages may not require careful fine-tuning nor self-organized criticality, but are generically explained by the underlying connection between language structures and the BKT phases.
Authors: Hongbo Zhao, Fei Zhu, Bolin Ni, Feng Zhu, Gaofeng Meng, Zhaoxiang Zhang
Abstract: For privacy and security concerns, the need to erase unwanted information from pre-trained vision models is becoming evident nowadays. In real-world scenarios, erasure requests originate at any time from both users and model owners, and these requests usually form a sequence. Therefore, under such a setting, selective information is expected to be continuously removed from a pre-trained model while maintaining the rest. We define this problem as continual forgetting and identify three key challenges. (i) For unwanted knowledge, efficient and effective deleting is crucial. (ii) For remaining knowledge, the impact brought by the forgetting procedure should be minimal. (iii) In real-world scenarios, the training samples may be scarce or partially missing during the process of forgetting. To address them, we first propose Group Sparse LoRA (GS-LoRA). Specifically, towards (i), we introduce Low-Rank Adaptation (LoRA) modules to fine-tune the Feed-Forward Network (FFN) layers in Transformer blocks for each forgetting task independently, and towards (ii), a simple group sparse regularization is adopted, enabling automatic selection of specific LoRA groups and zeroing out the others. To further extend GS-LoRA to more practical scenarios, we incorporate prototype information as additional supervision and introduce a more practical approach, GS-LoRA++. For each forgotten class, we move the logits away from its original prototype. For the remaining classes, we pull the logits closer to their respective prototypes. We conduct extensive experiments on face recognition, object detection, and image classification and demonstrate that our method manages to forget specific classes with minimal impact on other classes. Codes have been released on https://github.com/bjzhb666/GS-LoRA.
Authors: Hao Cheng, Erjia Xiao, Jing Shao, Yichi Wang, Le Yang, Chao Shen, Philip Torr, Jindong Gu, Renjing Xu
Abstract: Large Language Models (LLMs) demonstrate impressive zero-shot performance across a wide range of natural language processing tasks. Integrating various modality encoders further expands their capabilities, giving rise to Multimodal Large Language Models (MLLMs) that process not only text but also visual and auditory modality inputs. However, these advanced capabilities may also pose significant safety problems, as models can be exploited to generate harmful or inappropriate content through jailbreak attacks. While prior work has extensively explored how manipulating textual or visual modality inputs can circumvent safeguards in LLMs and MLLMs, the vulnerability of audio-specific jailbreak on Large Audio-Language Models (LALMs) remains largely underexplored. To address this gap, we introduce Jailbreak-AudioBench, which consists of the Toolbox, curated Dataset, and comprehensive Benchmark. The Toolbox supports not only text-to-audio conversion but also various editing techniques for injecting audio hidden semantics. The curated Dataset provides diverse explicit and implicit jailbreak audio examples in both original and edited forms. Utilizing this dataset, we evaluate multiple state-of-the-art LALMs and establish the most comprehensive Jailbreak benchmark to date for audio modality. Finally, Jailbreak-AudioBench establishes a foundation for advancing future research on LALMs safety alignment by enabling the in-depth exposure of more powerful jailbreak threats, such as query-based audio editing, and by facilitating the development of effective defense mechanisms.
Authors: Baoming Chang, Amin Kamali, Verena Kantere
Abstract: Although machine learning (ML) shows potential in improving query optimization by generating and selecting more efficient plans, ensuring the robustness of learning-based cost models (LCMs) remains challenging. These LCMs currently lack explainability, which undermines user trust and limits the ability to derive insights from their cost predictions to improve plan quality. Accurately converting tree-structured query plans into representations via tree models is also essential, as omitting any details may negatively impact subsequent cost model performance. Additionally, inherent uncertainty in cost estimation leads to inaccurate predictions, resulting in suboptimal plan selection. To address these challenges, we introduce Reqo, a Robust and Explainable Query Optimization cost model that comprehensively enhances three main stages in query optimization: plan generation, plan representation, and plan selection. Reqo integrates three innovations: the first explainability technique for LCMs that quantifies subgraph contributions and produces plan generation hints to enhance candidate plan quality; a novel tree model based on Bidirectional Graph Neural Networks (Bi-GNNs) with a Gated Recurrent Unit (GRU) aggregator to further capture both node-level and structural information and effectively strengthen plan representation; and an uncertainty-aware learning-to-rank cost estimator that adaptively integrates cost estimates with uncertainties to enhance plan selection robustness. Extensive experiments demonstrate that Reqo outperforms state-of-the-art approaches across all three stages.
Authors: Abdullahi Isa Ahmed, Jamal Bentahar, El Mehdi Amhoud
Abstract: As next-generation Internet of Things (NG-IoT) networks continue to grow, the number of connected devices is rapidly increasing, along with their energy demands, creating challenges for resource management and sustainability. Energy-efficient communication, particularly for power-limited IoT devices, is therefore a key research focus. In this paper, we study Long Range (LoRa) networks supported by multiple unmanned aerial vehicles (UAVs) in an uplink data collection scenario. Our objective is to maximize system energy efficiency by jointly optimizing transmission power, spreading factor, bandwidth, and user association. To address this challenging problem, we first model it as a partially observable stochastic game (POSG) to account for dynamic channel conditions, end device mobility, and partial observability at each UAV. We then propose a two-stage solution: a channel-aware matching algorithm for ED-UAV association and a cooperative multi-agent reinforcement learning (MARL) based multi-agent proximal policy optimization (MAPPO) framework for resource allocation under centralized training with decentralized execution (CTDE). Simulation results show that our proposed approach significantly outperforms conventional off-policy and on-policy MARL algorithms.
Authors: Dylan Zhang, Qirun Dai, Hao Peng
Abstract: High-quality supervised fine-tuning (SFT) data are crucial for eliciting strong capabilities from pretrained large language models (LLMs). Typically, instructions are paired with multiple responses sampled from other LLMs, which are often out of the distribution of the target model to be fine-tuned. This, at scale, can lead to diminishing returns and even hurt the models' performance and robustness. We propose **GRAPE**, a novel SFT framework that accounts for the unique characteristics of the target model. For each instruction, it gathers responses from various LLMs and selects the one with the highest probability measured by the target model, indicating that it aligns most closely with the target model's pretrained distribution; it then proceeds with standard SFT training. We first evaluate GRAPE with a controlled experiment, where we sample various solutions for each question in UltraInteract from multiple models and fine-tune commonly used LMs like LLaMA3.1-8B, Mistral-7B, and Qwen2.5-7B on GRAPE-selected data. GRAPE significantly outperforms strong baselines, including distilling from the strongest model with an absolute gain of up to 13.8%, averaged across benchmarks, and training on 3x more data with a maximum performance improvement of 17.3%. GRAPE's strong performance generalizes to realistic settings. We experiment with the post-training data used for Tulu3 and Olmo-2. GRAPE outperforms strong baselines trained on 4.5 times more data by 6.1% and a state-of-the-art data selection approach by 3% on average performance. Remarkably, using 1/3 of the data and half the number of epochs, GRAPE enables LLaMA3.1-8B to surpass the performance of Tulu3-SFT by 3.5%.
Authors: Runyao Yu, Yuchen Tao, Fabian Leimgruber, Tara Esterl, Jochen Stiasny, Derek W. Bunn, Qingsong Wen, Hongye Guo, Jochen L. Cremer
Abstract: Probabilistic intraday electricity price forecasting is becoming increasingly important with the growth of renewable generation and the rise in demand-side engagement. Their uncertainties have increased the trading risks closer to delivery and the subsequent imbalance settlement costs. As a consequence, intraday trading has emerged to mitigate these risks. Unlike auction markets, intraday trading in many jurisdictions is characterized by the continuous posting of buy and sell orders on power exchange platforms. This dynamic orderbook microstructure of price formation presents special challenges for price forecasting. Conventional methods represent the orderbook via domain features aggregated from buy and sell trades, or by treating it as a multivariate time series, but such representations neglect the full buy-sell interaction structure of the orderbook. This research therefore develops a new order fusion methodology, which is an end-to-end and parameter-efficient probabilistic forecasting model that learns a full interaction-aware representation of the buy-sell dynamics. Furthermore, as quantile crossing is often a problem in probabilistic forecasting, this approach hierarchically estimates the quantiles with non-crossing constraints. Extensive experiments on the market price indices across high-liquidity (German) and low-liquidity (Austrian) markets demonstrate consistent improvements over conventional baselines, and ablation studies highlight the contributions of the main modeling components. The methodology is available at: https://runyao-yu.github.io/OrderFusion/.
Authors: Mohammad Pivezhandi, Abusayeed Saifullah, Prashant Modekurthy
Abstract: Optimizing task-to-core allocation can substantially reduce power consumption in multi-core platforms without degrading user experience. However, existing approaches overlook critical factors such as parallelism, compute intensity, and heterogeneous core types. In this paper, we introduce a statistical learning approach for feature selection that identifies the most influential features-such as core type, speed, temperature, and application-level parallelism or memory intensity-for accurate environment modeling and efficient energy minimization, a critical consideration for embedded systems. Our experiments, conducted with state-of-the-art Linux governors and thermal modeling techniques, show that correlation-aware task-to-core allocation lowers energy consumption by up to 10% and reduces core temperature by up to 5C compared to random core selection. Furthermore, our compressed, bootstrapped regression model improves thermal prediction accuracy by 6% while cutting model parameters by 16%, yielding an overall mean square error reduction of 61.6% relative to existing approaches. We provided results based on superscalar Intel Core i7 12th Gen processors with 14 cores, and validated our method across a diverse set of hardware platforms and effectively balanced performance, power, and thermal demands through statistical feature evaluation.
Authors: Matteo Vilucchio, Yatin Dandi, Mat\'eo Pirio Rossignol, Cedric Gerbelot, Florent Krzakala
Abstract: The analytic characterization of the high-dimensional behavior of optimization for Generalized Linear Models (GLMs) with Gaussian data has been a central focus in statistics and probability in recent years. While convex cases, such as the LASSO, ridge regression, and logistic regression, have been extensively studied using a variety of techniques, the non-convex case remains far less understood despite its significance. A non-rigorous statistical physics framework has provided remarkable predictions for the behavior of high-dimensional optimization problems, but rigorously establishing their validity for non-convex problems has remained a fundamental challenge. In this work, we address this challenge by developing a systematic framework that rigorously proves replica-symmetric formulas for non-convex GLMs and precisely determines the conditions under which these formulas are valid. Remarkably, the rigorous replica-symmetric predictions align exactly with the conjectures made by physicists, and the so-called replicon condition. The originality of our approach lies in connecting two powerful theoretical tools: the Gaussian Min-Max Theorem, which we use to provide precise lower bounds, and Approximate Message Passing (AMP), which is shown to achieve these bounds algorithmically. We demonstrate the utility of this framework through significant applications: (i) by proving the optimality of the Tukey loss over the more commonly used Huber loss under a $\varepsilon$ contaminated data model, (ii) establishing the optimality of negative regularization in high-dimensional non-convex regression and (iii) characterizing the performance limits of linearized AMP algorithms. By rigorously validating statistical physics predictions in non-convex settings, we aim to open new pathways for analyzing increasingly complex optimization landscapes beyond the convex regime.
Authors: Patrice Bechard, Chao Wang, Amirhossein Abaskohi, Juan Rodriguez, Christopher Pal, David Vazquez, Spandana Gella, Sai Rajeswar, Perouz Taslakian
Abstract: Workflows are a fundamental component of automation in enterprise platforms, enabling the orchestration of tasks, data processing, and system integrations. Despite being widely used, building workflows can be complex, often requiring manual configuration through low-code platforms or visual programming tools. To simplify this process, we explore the use of generative foundation models, particularly vision-language models (VLMs), to automatically generate structured workflows from visual inputs. Translating hand-drawn sketches or computer-generated diagrams into executable workflows is challenging due to the ambiguity of free-form drawings, variations in diagram styles, and the difficulty of inferring execution logic from visual elements. To address this, we introduce StarFlow, a framework for generating structured workflow outputs from sketches using vision-language models. We curate a diverse dataset of workflow diagrams -- including synthetic, manually annotated, and real-world samples -- to enable robust training and evaluation. We finetune and benchmark multiple vision-language models, conducting a series of ablation studies to analyze the strengths and limitations of our approach. Our results show that finetuning significantly enhances structured workflow generation, outperforming large vision-language models on this task.
Authors: Arthur M. Faria, Ignacio F. Gra\~na, Savvas Varsamopoulos
Abstract: Quantum Graph Neural Networks (QGNNs) offer a promising approach to combining quantum computing with graph-structured data processing. While classical Graph Neural Networks (GNNs) are scalable and robust, existing QGNNs often lack flexibility due to graph-specific quantum circuit designs, limiting their applicability to diverse real-world problems. To address this, we propose a versatile QGNN framework inspired by GraphSAGE, using quantum models as aggregators. We integrate inductive representation learning techniques with parameterized quantum convolutional and pooling layers, bridging classical and quantum paradigms. The convolutional layer is flexible, allowing tailored designs for specific tasks. Benchmarked on a node regression task with the QM9 dataset, our framework, using a single minimal circuit for all aggregation steps, handles molecules with varying numbers of atoms without changing qubits or circuit architecture. While classical GNNs achieve higher training performance, our quantum approach remains competitive and often shows stronger generalization as molecular complexity increases. We also observe faster learning in early training epochs. To mitigate trainability limitations of a single-circuit setup, we extend the framework with multiple quantum aggregators on QM9. Assigning distinct circuits to each hop substantially improves training performance across all cases. Additionally, we numerically demonstrate the absence of barren plateaus as qubit numbers increase, suggesting that the proposed model can scale to larger, more complex graph-based problems.
Authors: Stefano Riva, Andrea Missaglia, Carolina Introini, In Cheol Bang, Antonio Cammi
Abstract: In recent years, algorithms aiming at learning models from available data have become quite popular due to two factors: 1) the significant developments in Artificial Intelligence techniques and 2) the availability of large amounts of data. Nevertheless, this topic has already been addressed by methodologies belonging to the Reduced Order Modelling framework, of which perhaps the most famous equation-free technique is Dynamic Mode Decomposition. This algorithm aims to learn the best linear model that represents the physical phenomena described by a time series dataset: its output is a best state operator of the underlying dynamical system that can be used, in principle, to advance the original dataset in time even beyond its span. However, in its standard formulation, this technique cannot deal with parametric time series, meaning that a different linear model has to be derived for each parameter realization. Research on this is ongoing, and some versions of a parametric Dynamic Mode Decomposition already exist. This work contributes to this research field by comparing the different algorithms presently deployed and assessing their advantages and shortcomings compared to each other. To this aim, three different thermal-hydraulics problems are considered: two benchmark 'flow over cylinder' test cases at diverse Reynolds numbers, whose datasets are, respectively, obtained with the FEniCS finite element solver and retrieved from the CFDbench dataset, and the DYNASTY experimental facility operating at Politecnico di Milano, which studies the natural circulation established by internally heated fluids for Generation IV nuclear applications, whose dataset was generated using the RELAP5 nodal solver.
Authors: Xingshan Zeng, Weiwen Liu, Xu Huang, Zezhong Wang, Lingzhi Wang, Liangyou Li, Yasheng Wang, Lifeng Shang, Xin Jiang, Ruiming Tang, Qun Liu
Abstract: Tool learning, which allows Large Language Models (LLMs) to leverage external tools for solving complex user tasks, has emerged as a promising avenue for extending model capabilities. However, existing approaches primarily focus on data synthesis for fine-tuning LLMs to invoke tools effectively, largely ignoring how to fully stimulate the potential of the model. In this paper, we propose ToolACE-R, a novel framework that includes both model-aware iterative training and adaptive refinement for tool learning. ToolACE-R features a model-aware iterative training procedure that progressively adjust training samples based on the model's evolving capabilities to maximize its potential. Additionally, it incorporates self-refinement training corpus which emphasizes LLM's ability to iteratively refine their tool calls, optimizing performance without requiring external feedback. Furthermore, we introduce adaptive self-refinement mechanism for efficient test-time scaling, where the trained model can autonomously determine when to stop the process based on iterative self-refinement. We conduct extensive experiments across several benchmark datasets, showing that ToolACE-R achieves competitive performance compared to advanced API-based models. The performance of tool invocation can be further improved efficiently through adaptive self-refinement. These results highlight the effectiveness and generalizability of ToolACE-R, offering a promising direction for more efficient and scalable tool learning.
Authors: Minshuo Chen, Renyuan Xu, Yumin Xu, Ruixun Zhang
Abstract: Financial scenario simulation is essential for risk management and portfolio optimization, yet it remains challenging especially in high-dimensional and small data settings common in finance. We propose a diffusion factor model that integrates latent factor structure into generative diffusion processes, bridging econometrics with modern generative AI to address the challenges of the curse of dimensionality and data scarcity in financial simulation. By exploiting the low-dimensional factor structure inherent in asset returns, we decompose the score function--a key component in diffusion models--using time-varying orthogonal projections, and this decomposition is incorporated into the design of neural network architectures. We derive rigorous statistical guarantees, establishing nonasymptotic error bounds for both score estimation at O(d^{5/2} n^{-2/(k+5)}) and generated distribution at O(d^{5/4} n^{-1/2(k+5)}), primarily driven by the intrinsic factor dimension k rather than the number of assets d, surpassing the dimension-dependent limits in the classical nonparametric statistics literature and making the framework viable for markets with thousands of assets. Numerical studies confirm superior performance in latent subspace recovery under small data regimes. Empirical analysis demonstrates the economic significance of our framework in constructing mean-variance optimal portfolios and factor portfolios. This work presents the first theoretical integration of factor structure with diffusion models, offering a principled approach for high-dimensional financial simulation with limited data. Our code is available at https://github.com/xymmmm00/diffusion_factor_model.
Authors: Jia-Qi Yang, Lei Shi
Abstract: We consider a class of statistical inverse problems involving the estimation of a regression operator from a Polish space to a separable Hilbert space, where the target lies in a vector-valued reproducing kernel Hilbert space induced by an operator-valued kernel. To address the associated ill-posedness, we analyze regularized stochastic gradient descent (SGD) algorithms in both online and finite-horizon settings. The former uses polynomially decaying step sizes and regularization parameters, while the latter adopts fixed values. Under suitable structural and distributional assumptions, we establish dimension-independent bounds for prediction and estimation errors. The resulting convergence rates are near-optimal in expectation, and we also derive high-probability estimates that imply almost sure convergence. Our analysis introduces a general technique for obtaining high-probability guarantees in infinite-dimensional settings. Possible extensions to broader kernel classes and encoder-decoder structures are briefly discussed.
Authors: Chen Wu, Qian Ma, Prasenjit Mitra, Sencun Zhu
Abstract: Knowledge distillation has become a cornerstone in modern machine learning systems, celebrated for its ability to transfer knowledge from a large, complex teacher model to a more efficient student model. Traditionally, this process is regarded as secure, assuming the teacher model is clean. This belief stems from conventional backdoor attacks relying on poisoned training data with backdoor triggers and attacker-chosen labels, which are not involved in the distillation process. Instead, knowledge distillation uses the outputs of a clean teacher model to guide the student model, inherently preventing recognition or response to backdoor triggers as intended by an attacker. In this paper, we challenge this assumption by introducing a novel attack methodology that strategically poisons the distillation dataset with adversarial examples embedded with backdoor triggers. This technique allows for the stealthy compromise of the student model while maintaining the integrity of the teacher model. Our innovative approach represents the first successful exploitation of vulnerabilities within the knowledge distillation process using clean teacher models. Through extensive experiments conducted across various datasets and attack settings, we demonstrate the robustness, stealthiness, and effectiveness of our method. Our findings reveal previously unrecognized vulnerabilities and pave the way for future research aimed at securing knowledge distillation processes against backdoor attacks.
Authors: St\'ephane Lhaut, Holger Rootz\'en, Johan Segers
Abstract: Economically responsible mitigation of multivariate extreme risks-such as extreme rainfall over large areas, large simultaneous variations in many stock prices, or widespread breakdowns in transportation systems-requires assessing the resilience of the systems under plausible stress scenarios. This paper uses Extreme Value Theory (EVT) to develop a new approach to simulating such multivariate extreme events. Specifically, we assume that after transformation to a standard scale the distribution of the random phenomenon of interest is multivariate regular varying and use this to provide a sampling procedure for extremes on the original scale. Our procedure combines a Wasserstein-Aitchison Generative Adversarial Network (WA-GAN) to simulate the tail dependence structure on the standard scale with joint modeling of the univariate marginal tails on the original scale. The WA-GAN procedure relies on the angular measure-encoding the distribution on the unit simplex of the angles of extreme observations-after transformation to Aitchison coordinates, which allows the Wasserstein-GAN algorithm to be run in a linear space. Our method is applied both to simulated data under various tail dependence scenarios and to a financial data set from the Kenneth French Data Library. The proposed algorithm demonstrates strong performance compared to existing alternatives in the literature, both in capturing tail dependence structures and in generating accurate new extreme observations.
Authors: Jaehyun Jeon, Min Soo Kim, Jang Han Yoon, Sumin Shim, Yejin Choi, Hanbin Kim, Dae Hyun Kim, Youngjae Yu
Abstract: User interface (UI) design goes beyond visuals to shape user experience (UX), underscoring the shift toward UI/UX as a unified concept. While recent studies have explored UI evaluation using Multimodal Large Language Models (MLLMs), they largely focus on surface-level features, overlooking how design choices influence user behavior at scale. To fill this gap, we introduce WiserUI-Bench, a novel benchmark for multimodal understanding of how UI/UX design affects user behavior, built on 300 real-world UI image pairs from industry A/B tests, with empirically validated winners that induced more user actions. For future design progress in practice, post-hoc understanding of why such winners succeed with mass users is also required; we support this via expert-curated key interpretations for each instance. Experiments across multiple MLLMs on WiserUI-Bench for two main tasks, (1) predicting the more effective UI image between an A/B-tested pair, and (2) explaining it post-hoc in alignment with expert interpretations, show that models exhibit limited understanding of the behavioral impact of UI/UX design. We believe our work will foster research on leveraging MLLMs for visual design in user behavior contexts.
Authors: Haowei Wang, Jingyi Wang, Zhongxiang Dai, Nai-Yuan Chiang, Szu Hui Ng, Cosmin G. Petra
Abstract: Constrained Bayesian optimization (CBO) methods have seen significant success in black-box optimization with constraints. One of the most commonly used CBO methods is the constrained expected improvement (CEI) algorithm. CEI is a natural extension of expected improvement (EI) when constraints are incorporated. However, the theoretical convergence rate of CEI has not been established. In this work, we study the convergence rate of CEI by analyzing its simple regret upper bound. First, we show that when the objective function $f$ and constraint function $c$ are assumed to each lie in a reproducing kernel Hilbert space (RKHS), CEI achieves the convergence rates of $\mathcal{O} \left(t^{-\frac{1}{2}}\log^{\frac{d+1}{2}}(t) \right) \ \text{and }\ \mathcal{O}\left(t^{\frac{-\nu}{2\nu+d}} \log^{\frac{\nu}{2\nu+d}}(t)\right)$ for the commonly used squared exponential and Mat\'{e}rn kernels ($\nu>\frac{1}{2}$), respectively. Second, we show that when $f$ is assumed to be sampled from Gaussian processes (GPs), CEI achieves similar convergence rates with a high probability. Numerical experiments are performed to validate the theoretical analysis.
Authors: Pengrui Quan, Brian Wang, Kang Yang, Liying Han, Mani Srivastava
Abstract: Spatiotemporal reasoning plays a key role in Cyber-Physical Systems (CPS). Despite advances in Large Language Models (LLMs) and Large Reasoning Models (LRMs), their capacity to reason about complex spatiotemporal signals remains underexplored. This paper proposes a hierarchical SpatioTemporal reAsoning benchmaRK, STARK, to systematically evaluate LLMs across three levels of reasoning complexity: state estimation (e.g., predicting field variables, localizing and tracking events in space and time), spatiotemporal reasoning over states (e.g., inferring spatial-temporal relationships), and world-knowledge-aware reasoning that integrates contextual and domain knowledge (e.g., intent prediction, landmark-aware navigation). We curate 26 distinct spatiotemporal tasks with diverse sensor modalities, comprising 14,552 challenges where models answer directly or by Python Code Interpreter. Evaluating 3 LRMs and 8 LLMs, we find LLMs achieve limited success in tasks requiring geometric reasoning (e.g., multilateration or triangulation), particularly as complexity increases. Surprisingly, LRMs show robust performance across tasks with various levels of difficulty, often competing or surpassing traditional first-principle-based methods. Our results show that in reasoning tasks requiring world knowledge, the performance gap between LLMs and LRMs narrows, with some LLMs even surpassing LRMs. However, the LRM o3 model continues to achieve leading performance across all evaluated tasks, a result attributed primarily to the larger size of the reasoning models. STARK motivates future innovations in model architectures and reasoning paradigms for intelligent CPS by providing a structured framework to identify limitations in the spatiotemporal reasoning of LLMs and LRMs.
Authors: Arjhun Swaminathan, Mete Akg\"un
Abstract: Deep neural networks for image classification remain vulnerable to adversarial examples -- small, imperceptible perturbations that induce misclassifications. In black-box settings, where only the final prediction is accessible, crafting targeted attacks that aim to misclassify into a specific target class is particularly challenging due to narrow decision regions. Current state-of-the-art methods often exploit the geometric properties of the decision boundary separating a source image and a target image rather than incorporating information from the images themselves. In contrast, we propose Targeted Edge-informed Attack (TEA), a novel attack that utilizes edge information from the target image to carefully perturb it, thereby producing an adversarial image that is closer to the source image while still achieving the desired target classification. Our approach consistently outperforms current state-of-the-art methods across different models in low query settings (nearly 70% fewer queries are used), a scenario especially relevant in real-world applications with limited queries and black-box access. Furthermore, by efficiently generating a suitable adversarial example, TEA provides an improved target initialization for established geometry-based attacks.
Authors: Chun-Yi Kuan, Hung-yi Lee
Abstract: Audio-aware large language models (ALLMs) have recently made great strides in understanding and processing audio inputs. These models are typically adapted from text-based large language models (LLMs) through additional training on audio-related tasks. This adaptation process presents two major limitations. First, ALLMs often suffer from catastrophic forgetting, where crucial textual capabilities like instruction-following are lost after training on audio data. In some cases, models may even hallucinate sounds that are not present in the input audio, raising concerns about reliability. Second, achieving cross-modal alignment between audio and language typically relies on large collections of task-specific question-answer pairs for instruction tuning, making it resource-intensive. To address these issues, previous works have leveraged the backbone LLMs to synthesize general-purpose, caption-style alignment data. In this paper, we propose a data generation framework that produces contrastive-like training data, designed to enhance ALLMs' ability to differentiate between present and absent sounds. We further extend our approach to multi-audio scenarios, enabling the model to either explain differences between audio inputs or produce unified captions that describe all inputs, thereby enhancing audio-language alignment. We refer to the entire ALLM training framework as bootstrapping audio-language alignment via synthetic data generation from backbone LLMs (BALSa). Experimental results indicate that our method effectively mitigates audio hallucinations while reliably maintaining strong performance on audio understanding and reasoning benchmarks, as well as instruction-following skills. Moreover, incorporating multi-audio training further enhances the model's comprehension and reasoning capabilities. Overall, BALSa offers an efficient and scalable approach to developing ALLMs.
Authors: Meiyan Kang, Shizuo Kaji, Sang-Yun Lee, Taegon Kim, Hee-Hwan Ryu, Suyoung Choi
Abstract: Ground Penetrating Radar (GPR) is a widely used Non-Destructive Testing (NDT) technique for subsurface exploration, particularly in infrastructure inspection and maintenance. However, conventional interpretation methods are often limited by noise sensitivity and a lack of structural awareness. This study presents a novel framework that enhances the detection of underground utilities, especially pipelines, by integrating shape-aware topological features derived from B-scan GPR images using Topological Data Analysis (TDA), with the spatial detection capabilities of the YOLOv5 deep neural network (DNN). We propose a novel shape-aware topological representation that amplifies structural features in the input data, thereby improving the model's responsiveness to the geometrical features of buried objects. To address the scarcity of annotated real-world data, we employ a Sim2Real strategy that generates diverse and realistic synthetic datasets, effectively bridging the gap between simulated and real-world domains. Experimental results demonstrate significant improvements in mean Average Precision (mAP), validating the robustness and efficacy of our approach. This approach underscores the potential of TDA-enhanced learning in achieving reliable, real-time subsurface object detection, with broad applications in urban planning, safety inspection, and infrastructure management.
Authors: Boris Sedlak, Alireza Furutanpey, Zihang Wang, V\'ictor Casamayor Pujol, Schahram Dustdar
Abstract: Edge computing breaks with traditional autoscaling due to strict resource constraints, thus, motivating more flexible scaling behaviors using multiple elasticity dimensions. This work introduces an agent-based autoscaling framework that dynamically adjusts both hardware resources and internal service configurations to maximize requirements fulfillment in constrained environments. We compare four types of scaling agents: Active Inference, Deep Q Network, Analysis of Structural Knowledge, and Deep Active Inference, using two real-world processing services running in parallel: YOLOv8 for visual recognition and OpenCV for QR code detection. Results show all agents achieve acceptable SLO performance with varying convergence patterns. While the Deep Q Network benefits from pre-training, the structural analysis converges quickly, and the deep active inference agent combines theoretical foundations with practical scalability advantages. Our findings provide evidence for the viability of multi-dimensional agent-based autoscaling for edge environments and encourage future work in this research direction.
Authors: Tongtong Liang, Dan Qiao, Yu-Xiang Wang, Rahul Parhi
Abstract: We study the implicit bias of flatness / low (loss) curvature and its effects on generalization in two-layer overparameterized ReLU networks with multivariate inputs -- a problem well motivated by the minima stability and edge-of-stability phenomena in gradient-descent training. Existing work either requires interpolation or focuses only on univariate inputs. This paper presents new and somewhat surprising theoretical results for multivariate inputs. On two natural settings (1) generalization gap for flat solutions, and (2) mean-squared error (MSE) in nonparametric function estimation by stable minima, we prove upper and lower bounds, which establish that while flatness does imply generalization, the resulting rates of convergence necessarily deteriorate exponentially as the input dimension grows. This gives an exponential separation between the flat solutions compared to low-norm solutions (i.e., weight decay), which are known not to suffer from the curse of dimensionality. In particular, our minimax lower bound construction, based on a novel packing argument with boundary-localized ReLU neurons, reveals how flat solutions can exploit a kind of "neural shattering" where neurons rarely activate, but with high weight magnitudes. This leads to poor performance in high dimensions. We corroborate these theoretical findings with extensive numerical simulations. To the best of our knowledge, our analysis provides the first systematic explanation for why flat minima may fail to generalize in high dimensions.
Authors: Shirong Xu, Jingnan Zhang, Junhui Wang
Abstract: Paired comparison data, where users evaluate items in pairs, play a central role in ranking and preference learning tasks. While ordinal comparison data intuitively offer richer information than binary comparisons, this paper challenges that conventional wisdom. We propose a general parametric framework for modeling ordinal paired comparisons without ties. The model adopts a generalized additive structure, featuring a link function that quantifies the preference difference between two items and a pattern function that governs the distribution over ordinal response levels. This framework encompasses classical binary comparison models as special cases, by treating binary responses as binarized versions of ordinal data. Within this framework, we show that binarizing ordinal data can significantly improve the accuracy of ranking recovery. Specifically, we prove that under the counting algorithm, the ranking error associated with binary comparisons exhibits a faster exponential convergence rate than that of ordinal data. Furthermore, we characterize a substantial performance gap between binary and ordinal data in terms of a signal-to-noise ratio (SNR) determined by the pattern function. We identify the pattern function that minimizes the SNR and maximizes the benefit of binarization. Extensive simulations and a real application on the MovieLens dataset further corroborate our theoretical findings.
Authors: Yun Qu, Qi Wang, Yixiu Mao, Vincent Tao Hu, Bj\"orn Ommer, Xiangyang Ji
Abstract: Recent advances have witnessed the effectiveness of reinforcement learning (RL) finetuning in enhancing the reasoning capabilities of large language models (LLMs). The optimization process often requires numerous iterations to achieve satisfactory performance, resulting in high computational costs due to the need for frequent prompt evaluations under intensive LLM interactions and repeated policy updates. Appropriate online prompt selection methods reduce iteration steps by prioritizing informative prompts during training, while the pipeline's reliance on exhaustive prompt evaluation and subset selection for optimization still incurs substantial computational overhead due to frequent LLM inference calls. Distinguished from these direct evaluate-then-select schemes, this work investigates iterative approximate evaluation for arbitrary prompts and introduces Model Predictive Prompt Selection (MoPPS), a Bayesian risk-predictive framework that online estimates prompt difficulty without requiring costly LLM interactions. Technically, MoPPS models each prompt's success rate as a latent variable, performs streaming Bayesian inference, and employs posterior sampling in a constructed multi-armed bandit machine, enabling sample efficient and adaptive prompt selection. Extensive experiments across mathematics, planning, and vision-based geometry tasks show that MoPPS reliably predicts prompt difficulty and accelerates training with significantly reduced LLM rollouts. Our code is available at https://github.com/thu-rllab/MoPPS.
Authors: Zhou Li, Xiang Zhang, Jiawen Lv, Jihao Fan, Haiqiang Chen, Giuseppe Caire
Abstract: In hierarchical secure aggregation (HSA), a server communicates with clustered users through an intermediate layer of relays to compute the sum of users' inputs under two security requirements -- server security and relay security. Server security requires that the server learns nothing beyond the desired sum even when colluding with a subset of users, while relay security requires that each relay remains oblivious to the users' inputs under collusion. Existing work on HSA enforces homogeneous security where \tit{all} inputs must be protected against \tit{any} subset of potential colluding users with sizes up to a predefined threshold. Such a \homo formulation cannot capture scenarios with \tit{\het} \secty \reqs where \diff users may demand various levels of protection. In this paper, we study hierarchical secure aggregation (HSA) with heterogeneous security requirements and arbitrary user collusion. Specifically, we consider scenarios where the inputs of certain groups of users must remain information-theoretically secure against inference by the server or any relay, even if the server or any relay colludes with an arbitrary subset of other users. Under server security, the server learns nothing about these protected inputs beyond the prescribed aggregate sum, despite any such collusion. Under relay security, each relay similarly obtains no information about the protected inputs under the same collusion model. We characterize the optimal communication rates achievable across all layers for all parameter regimes. Furthermore, we study the minimum source keys required at the users to ensure security. For this source key requirement, we provide tight characterizations in two broad regimes determined by the security and collusion constraints, and establish a general information-theoretic lower bound together with a bounded-gap achievable scheme for the remaining regime.
Authors: Md Obyedullahil Mamun, Md Adyelullahil Mamun, Arif Ahmad, Md. Imran Hossain Emu
Abstract: Punctuation restoration enhances the readability of text and is critical for post-processing tasks in Automatic Speech Recognition (ASR), especially for low-resource languages like Bangla. In this study, we explore the application of transformer-based models, specifically XLM-RoBERTa-large, to automatically restore punctuation in unpunctuated Bangla text. We focus on predicting four punctuation marks: period, comma, question mark, and exclamation mark across diverse text domains. To address the scarcity of annotated resources, we constructed a large, varied training corpus and applied data augmentation techniques. Our best-performing model, trained with an augmentation factor of alpha = 0.20%, achieves an accuracy of 97.1% on the News test set, 91.2% on the Reference set, and 90.2% on the ASR set. Results show strong generalization to reference and ASR transcripts, demonstrating the model's effectiveness in real-world, noisy scenarios. This work establishes a strong baseline for Bangla punctuation restoration and contributes publicly available datasets and code to support future research in low-resource NLP.
Authors: Agnideep Aich, Sameera Hewage, Md Monzur Murshed, Bruce Wade, Ashit Baran Aich
Abstract: Modern neural networks often produce miscalibrated confidence scores and struggle to detect out-of-distribution (OOD) inputs, while most existing methods post-process outputs without testing internal consistency. We introduce the Bag-of-Coins (BoC) probe, a non-parametric diagnostic of logit coherence that compares softmax confidence $\hat p$ to an aggregate of pairwise Luce-style dominance probabilities $\bar q$, yielding a deterministic coherence score and a p-value-based structural score. Across ViT, ResNet, and RoBERTa with ID/OOD test sets, the coherence gap $\Delta=\bar q-\hat p$ reveals clear ID/OOD separation for ViT (ID ${\sim}0.1$-$0.2$, OOD ${\sim}0.5$-$0.6$) but substantial overlap for ResNet and RoBERTa (both ${\sim}0$), indicating architecture-dependent uncertainty geometry. As a practical method, BoC improves calibration only when the base model is poorly calibrated (ViT: ECE $0.024$ vs.\ $0.180$) and underperforms standard calibrators (ECE ${\sim}0.005$), while for OOD detection it fails across architectures (AUROC $0.020$-$0.253$) compared to standard scores ($0.75$-$0.99$). We position BoC as a research diagnostic for interrogating how architectures encode uncertainty in logit geometry rather than a production calibration or OOD detection method.
Authors: Agnideep Aich, Ashit Baran Aich, Bruce Wade
Abstract: Gradient descent (GD) on deep neural network loss landscapes is non-convex, yet often converges far faster in practice than classical guarantees suggest. Prior work shows that within locally quasi-convex regions (LQCRs), GD converges to stationary points at sublinear rates, leaving the commonly observed near-exponential training dynamics unexplained. We show that, under a mild local Neural Tangent Kernel (NTK) stability assumption, the loss satisfies a PL-type error bound within these regions, yielding a Locally Polyak-Lojasiewicz Region (LPLR) in which the squared gradient norm controls the suboptimality gap. For properly initialized finite-width networks, we show that under local NTK stability this PL-type mechanism holds around initialization and establish linear convergence of GD as long as the iterates remain within the resulting LPLR. Empirically, we observe PL-like scaling and linear-rate loss decay in controlled full-batch training and in a ResNet-style CNN trained with mini-batch SGD on a CIFAR-10 subset, indicating that LPLR signatures can persist under modern architectures and stochastic optimization. Overall, the results connect local geometric structure, local NTK stability, and fast optimization rates in a finite-width setting.
Authors: Nicola Apollonio, Giovanni Franzina, Giovanni Luca Torrisi
Abstract: We consider fully connected and feedforward deep neural networks with dependent and possibly heavy-tailed weights, as introduced in Lee et al. 2023, to address limitations of the standard Gaussian prior. It has been proved in Lee et al. 2023 that, as the number of nodes in the hidden layers grows large, according to a sequential and ordered limit, the law of the output converges weakly to a Gaussian mixture. Among our results, we present sufficient conditions on the model parameters (the activation function and the associated L\'evy measures) which ensure that the sequential limit is independent of the order. Next, we study the neural network through the lens of the posterior distribution with a Gaussian likelihood. If the random covariance matrix of the infinite-width limit is positive definite under the prior, we identify the posterior distribution of the output in the wide-width limit according to a sequential regime. Remarkably, we provide mild sufficient conditions to ensure the aforementioned invertibility of the random covariance matrix under the prior, thereby extending the results in L. Carvalho et al. 2025. We illustrate our findings using numerical simulations.
Authors: Anubhav Kataria, Surbhi Madan, Shreya Ghosh, Tom Gedeon, Abhinav Dhall
Abstract: Understanding individual, group and event level emotions along with contextual information is crucial for analyzing a multi-person social situation. To achieve this, we frame emotion comprehension as the task of predicting fine-grained individual emotion to coarse grained group and event level emotion. We introduce GEMS that leverages a multimodal swin-transformer and S3Attention based architecture, which processes an input scene, group members, and context information to generate joint predictions. Existing multi-person emotion related benchmarks mainly focus on atomic interactions primarily based on emotion perception over time and group level. To this end, we extend and propose VGAF-GEMS to provide more fine grained and holistic analysis on top of existing group level annotation of VGAF dataset. GEMS aims to predict basic discrete and continuous emotions (including valence and arousal) as well as individual, group and event level perceived emotions. Our benchmarking effort links individual, group and situational emotional responses holistically. The quantitative and qualitative comparisons with adapted state-of-the-art models demonstrate the effectiveness of GEMS framework on VGAF-GEMS benchmarking. We believe that it will pave the way of further research. The code and data is available at: https://github.com/katariaak579/GEMS
Authors: Jun Qi, Chao-Han Huck Yang, Pin-Yu Chen, Min-Hsiu Hsieh
Abstract: Variational Quantum Computing (VQC) faces fundamental scalability barriers, primarily due to barren plateaus and sensitivity to quantum noise. To address these challenges, we introduce TensorHyper-VQC, a novel tensor-train (TT)-guided hypernetwork framework that significantly improves the robustness and scalability of VQC. Our framework fully delegates the generation of quantum circuit parameters to a classical TT network, effectively decoupling optimization from quantum hardware. This innovative parameterization mitigates gradient vanishing, enhances noise resilience through structured low-rank representations, and facilitates efficient gradient propagation. Grounded in Neural Tangent Kernel and statistical learning theory, our rigorous theoretical analyses establish strong guarantees on approximation capability, optimization stability, and generalization performance. Extensive empirical results across quantum dot classification, Max-Cut optimization, and molecular quantum simulation tasks demonstrate that TensorHyper-VQC consistently achieves superior performance and robust noise tolerance, including hardware-level validation on a 156-qubit IBM Heron processor. These results position TensorHyper-VQC as a scalable and noise-resilient framework for advancing practical quantum machine learning on near-term devices.
Authors: Chengshuai Zhao, Zhen Tan, Pingchuan Ma, Dawei Li, Bohan Jiang, Yancheng Wang, Yingzhen Yang, Huan Liu
Abstract: Chain-of-Thought (CoT) prompting has been shown to be effective in eliciting structured reasoning (i.e., CoT reasoning) from large language models (LLMs). Regardless of its popularity, recent studies expose its failures in some reasoning tasks, raising fundamental questions about the nature of CoT reasoning. In this work, we propose a data distribution lens to understand when and why CoT reasoning succeeds or fails. We hypothesize that CoT reasoning reflects a structured inductive bias learned from in-distribution data, enabling models to conditionally generate reasoning trajectories that approximate those observed during training. As such, the effectiveness of CoT reasoning is fundamentally governed by the nature and degree of distribution discrepancy between training data and test queries. Guided by this lens, we dissect CoT reasoning via three dimensions: task, length, and format. To test the hypothesis, we introduce DataAlchemy, an abstract and fully controllable environment that trains LLMs from scratch and systematically probes them under various distribution conditions. Through rigorous controlled experiments, we reveal that CoT reasoning is a brittle mirage when it is pushed beyond training distributions, emphasizing the ongoing challenge of achieving genuine and generalizable reasoning.
Authors: Linyu Li, Zhi Jin, Yuanpeng He, Dongming Jin, Yichi Zhang, Haoran Duan, Xuan Zhang, Zhengwei Tao, Nyima Tash
Abstract: As social media and the World Wide Web become hubs for information dissemination, effectively organizing and understanding the vast amounts of dynamically evolving Web content is crucial. Knowledge graphs (KGs) provide a powerful framework for structuring this information. However, the rapid emergence of new hot topics, user relationships, and events in social media renders traditional static knowledge graph embedding (KGE) models rapidly outdated. Continual Knowledge Graph Embedding (CKGE) aims to address this issue, but existing methods commonly suffer from catastrophic forgetting, whereby older, but still valuable, information is lost when learning new knowledge (such as new memes or trending events). This means the model cannot effectively learn the evolution of the data. We propose a novel CKGE framework, BAKE. Unlike existing methods, BAKE formulates CKGE as a sequential Bayesian inference problem and utilizes the Bayesian posterior update principle as a natural continual learning strategy. This principle is insensitive to data order and provides theoretical guarantees to preserve prior knowledge as much as possible. Specifically, we treat each batch of new data as a Bayesian update to the model's prior. By maintaining the posterior distribution, the model effectively preserves earlier knowledge even as it evolves over multiple snapshots. Furthermore, to constrain the evolution of knowledge across snapshots, we introduce a continual clustering method that maintains the compact cluster structure of entity embeddings through a regularization term, ensuring semantic consistency while allowing controlled adaptation to new knowledge. We conduct extensive experiments on multiple CKGE benchmarks, which demonstrate that BAKE achieves the top performance in the vast majority of cases compared to existing approaches.
Authors: Zhuohao Yu, Xingru Jiang, Weizheng Gu, Yidong Wang, Qingsong Wen, Shikun Zhang, Wei Ye
Abstract: Watermarking LLM-generated text is critical for content attribution and misinformation prevention. However, existing methods compromise text quality, require white-box model access and logit manipulation. These limitations exclude API-based models and multilingual scenarios. We propose SAEMark, a general framework for post-hoc multi-bit watermarking that embeds personalized messages solely via inference-time, feature-based rejection sampling without altering model logits or requiring training. Our approach operates on deterministic features extracted from generated text, selecting outputs whose feature statistics align with key-derived targets. This framework naturally generalizes across languages and domains while preserving text quality through sampling LLM outputs instead of modifying. We provide theoretical guarantees relating watermark success probability and compute budget that hold for any suitable feature extractor. Empirically, we demonstrate the framework's effectiveness using Sparse Autoencoders (SAEs), achieving superior detection accuracy and text quality. Experiments across 4 datasets show SAEMark's consistent performance, with 99.7% F1 on English and strong multi-bit detection accuracy. SAEMark establishes a new paradigm for scalable watermarking that works out-of-the-box with closed-source LLMs while enabling content attribution.
Authors: Alexandre Brown, Glen Berseth
Abstract: Visual reinforcement learning (RL) is challenging due to the need to extract useful representations from high-dimensional inputs while learning effective control from sparse and noisy rewards. Although large perception models exist, integrating them effectively into RL for visual generalization and improved sample efficiency remains difficult. We propose SegDAC, a Segmentation-Driven Actor-Critic method. SegDAC uses Segment Anything (SAM) for object-centric decomposition and YOLO-World to ground the image segmentation process via text inputs. It includes a novel transformer-based architecture that supports a dynamic number of segments at each time step and effectively learns which segments to focus on using online RL, without using human labels. By evaluating SegDAC over a challenging visual generalization benchmark using Maniskill3, which covers diverse manipulation tasks under strong visual perturbations, we demonstrate that SegDAC achieves significantly better visual generalization, doubling prior performance on the hardest setting and matching or surpassing prior methods in sample efficiency across all evaluated tasks. Project Page: https://segdac.github.io/
Authors: Vy Tuong Dang, An Vo, Emilio Villa-Cueva, Quang Tau, Duc Dm, Thamar Solorio, Daeyoung Kim
Abstract: We introduce VMMU, a Vietnamese Multitask Multimodal Understanding and Reasoning Benchmark designed to evaluate how vision-language models (VLMs) interpret and reason over visual and textual information beyond English. VMMU consists of 2.5k multimodal questions across 7 tasks, covering a diverse range of problem contexts, including STEM problem solving, data interpretation, rule-governed visual reasoning, and abstract visual reasoning. All questions require genuine multimodal integration, rather than reliance on text-only cues or OCR-based shortcuts. We evaluate a diverse set of state-of-the-art proprietary and open-source VLMs on VMMU. Despite strong Vietnamese OCR performance, proprietary models achieve only 66% mean accuracy. Further analysis shows that the primary source of failure is not OCR, but instead multimodal grounding and reasoning over text and visual evidence. Code and data are available at https://vmmu.github.io.
URLs: https://vmmu.github.io.
Authors: Yuntao Du, Yuetian Chen, Hanshen Xiao, Bruno Ribeiro, Ninghui Li
Abstract: A Membership Inference Attack (MIA) assesses how much a target machine learning model reveals about its training data by determining whether specific query instances were part of the training set. State-of-the-art MIAs rely on training hundreds of shadow models that are independent of the target model, leading to significant computational overhead. In this paper, we introduce Imitative Membership Inference Attack (IMIA), which employs a novel imitative training technique to strategically construct a small number of target-informed imitative models that closely replicate the target model's behavior for inference. Extensive experimental results demonstrate that IMIA substantially outperforms existing MIAs in various attack settings while only requiring less than 5% of the computational cost of state-of-the-art approaches.
Authors: Agapi Rissaki, Ilias Fountalis, Wolfgang Gatterbauer, Benny Kimelfeld
Abstract: In recent years, there has been significant progress in the development of deep learning models over relational databases, including architectures based on heterogeneous graph neural networks (hetero-GNNs) and heterogeneous graph transformers. In effect, such architectures state how the database records and links (e.g., foreign-key references) translate into a large, complex numerical expression, involving numerous learnable parameters. This complexity makes it hard to explain, in human-understandable terms, how a model uses the available data to arrive at a given prediction. We present a novel framework for explaining machine-learning models over relational databases, where explanations are view definitions that highlight focused parts of the database that mostly contribute to the model's prediction. We establish such global abductive explanations by adapting the classic notion of determinacy by Nash, Segoufin, and Vianu (2010). In addition to tuning the tradeoff between determinacy and conciseness, the framework allows controlling the level of granularity by adopting different fragments of view definitions, such as ones highlighting whole columns, foreign keys between tables, relevant groups of tuples, and so on. We investigate the realization of the framework in the case of hetero-GNNs, and develop a model-specific approach via the notion of learnable masks. For comparison, we propose model-agnostic heuristic baselines and show that our approach is both more efficient and achieves better explanation quality in most cases. Our extensive empirical evaluation on the RelBench collection across diverse domains and record-level tasks demonstrates both the usefulness of our explanations and the efficiency of their generation.
Authors: Yikuan Xia, Jiazun Chen, Yirui Zhan, Suifeng Zhao, Weipeng Jiang, Chaorui Zhang, Wei Han, Bo Bai, Jun Gao
Abstract: This paper presents the db3 team's winning solution for the Meta CRAG-MM Challenge 2025 at KDD Cup'25. Addressing the challenge's unique multi-modal, multi-turn question answering benchmark (CRAG-MM), we developed a comprehensive framework that integrates tailored retrieval pipelines for different tasks with a unified LLM-tuning approach for hallucination control. Our solution features (1) domain-specific retrieval pipelines handling image-indexed knowledge graphs, web sources, and multi-turn conversations; and (2) advanced refusal training using SFT, DPO, and RL. The system achieved 2nd place in Task 1, 2nd place in Task 2, and 1st place in Task 3, securing the grand prize for excellence in ego-centric queries through superior handling of first-person perspective challenges.
Authors: Jia-Qi Yang, Lei Shi
Abstract: We develop a stochastic approximation framework for learning nonlinear operators between infinite-dimensional spaces utilizing general Mercer operator-valued kernels. Our framework encompasses two key classes: (i) compact kernels, which admit discrete spectral decompositions, and (ii) diagonal kernels of the form $K(x,x')=k(x,x')T$, where $k$ is a scalar-valued kernel and $T$ is a positive operator on the output space. This broad setting induces expressive vector-valued reproducing kernel Hilbert spaces (RKHSs) that generalize the classical $K=kI$ paradigm, thereby enabling rich structural modeling with rigorous theoretical guarantees. To address target operators lying outside the RKHS, we introduce vector-valued interpolation spaces to precisely quantify misspecification error. Within this framework, we establish dimension-free polynomial convergence rates, demonstrating that nonlinear operator learning can overcome the curse of dimensionality. The use of general operator-valued kernels further allows us to derive rates for intrinsically nonlinear operator learning, going beyond the linear-type behavior inherent in diagonal constructions of $K=kI$. Importantly, this framework accommodates a wide range of operator learning tasks, ranging from integral operators such as Fredholm operators to architectures based on encoder-decoder representations. Moreover, we validate its effectiveness through numerical experiments on the two-dimensional Navier-Stokes equations.
Authors: Xin Lei Lin, Soroush Mehraban, Abhishek Moturu, Babak Taati
Abstract: Automated pain assessment from facial expressions is crucial for non-communicative patients, such as those with dementia. Progress has been limited by two challenges: (i) existing datasets exhibit severe demographic and label imbalance due to ethical constraints, and (ii) current generative models cannot precisely control facial action units (AUs), facial structure, or clinically validated pain levels. We present 3DPain, a large-scale synthetic dataset specifically designed for automated pain assessment, featuring unprecedented annotation richness and demographic diversity. Our three-stage framework generates diverse 3D meshes, textures them with diffusion models, and applies AU-driven face rigging to synthesize multi-view faces with paired neutral and pain images, AU configurations, PSPI scores, and the first dataset-level annotations of pain-region heatmaps. The dataset comprises 82,500 samples across 25,000 pain expression heatmaps and 2,500 synthetic identities balanced by age, gender, and ethnicity. We further introduce ViTPain, a Vision Transformer based cross-modal distillation framework in which a heatmap-trained teacher guides a student trained on RGB images, enhancing accuracy, interpretability, and clinical reliability. Together, 3DPain and ViTPain establish a controllable, diverse, and clinically grounded foundation for generalizable automated pain assessment.
Authors: Huu Nguyen (Sonny), Victor May (Sonny), Harsh Raj (Sonny), Marianna Nezhurina (Sonny), Yishan Wang (Sonny), Yanqi Luo (Sonny), Minh Chien Vu (Sonny), Taishi Nakamura (Sonny), Ken Tsui (Sonny), Van Khue Nguyen (Sonny), David Salinas (Sonny), Aleksandra Krasnod\k{e}bska (Sonny), Christoph Schuhmann (Sonny), Mats Leon Richter (Sonny), Xuan-Son (Sonny), Vu, Jenia Jitsev
Abstract: We present MixtureVitae, an open-access pretraining corpus built to minimize legal risk while providing strong downstream performance. MixtureVitae follows a permissive-first, risk-mitigated sourcing strategy that combines public-domain and permissively licensed text (e.g., CC-BY/Apache) with carefully justified low-risk additions (e.g., government works and EU TDM-eligible sources). MixtureVitae adopts a simple, single-stage pretraining recipe that integrates a large proportion of permissive synthetic instruction and reasoning data-signals typically introduced during post-training and generally scarce in permissive web corpora. We categorize all sources into a three-tier scheme that reflects varying risk levels and provide shard-level provenance metadata to enable risk-aware usage. In controlled experiments using the open-sci-ref training protocol (fixed architectures and hyperparameters; 50B and 300B token budgets across 130M-1.7B parameters), models trained on MixtureVitae consistently outperform other permissive datasets across a suite of standard benchmarks, and at the 1.7B-parameters/300B-tokens setting, they surpass FineWeb-Edu and approach DCLM late in training. Performance is particularly strong on MMLU and on math and code benchmarks: a 1.7B model pretrained on 300B MixtureVitae tokens matches or exceeds a strong 1.7B instruction-tuned baseline on GSM8K, HumanEval, and MBPP, despite using over 36 times fewer tokens (300B vs. ~11T). Supported by a thorough decontamination analysis, these results show that permissive-first data with high instruction and reasoning density, tiered by licensing and provenance-related risk, can provide a practical and risk-mitigated foundation for training capable LLMs, reducing reliance on broad web scrapes without sacrificing competitiveness. Code: https://github.com/ontocord/mixturevitae
Authors: Rina Foygel Barber, Ashwin Pananjady
Abstract: We consider the problem of uncertainty quantification for prediction in a time series: if we use past data to forecast the next time point, can we provide valid prediction intervals around our forecasts? To avoid placing distributional assumptions on the data, in recent years the conformal prediction method has been a popular approach for predictive inference, since it provides distribution-free coverage for any iid or exchangeable data distribution. However, in the time series setting, the strong empirical performance of conformal prediction methods is not well understood, since even short-range temporal dependence is a strong violation of the exchangeability assumption. Using predictors with "memory" -- i.e., predictors that utilize past observations, such as autoregressive models -- further exacerbates this problem. In this work, we examine the theoretical properties of split conformal prediction in the time series setting, including the case where predictors may have memory. Our results bound the loss of coverage of these methods in terms of a new "switch coefficient", measuring the extent to which temporal dependence within the time series creates violations of exchangeability. Our characterization of the coverage probability is sharp over the class of stationary, $\beta$-mixing processes. Along the way, we introduce tools that may prove useful in analyzing other predictive inference methods for dependent data.
Authors: Zhiying Jiang, Raihan Seraj, Marcos Villagra, Bidhan Roy
Abstract: We present Paris, the first publicly released diffusion model pre-trained entirely through decentralized computation. Paris demonstrates that high-quality text-to-image generation can be achieved without centrally coordinated infrastructure. Paris is open for research and commercial use. Paris required implementing our Distributed Diffusion Training framework from scratch. The model consists of 8 expert diffusion models (129M-605M parameters each) trained in complete isolation with no gradient, parameter, or intermediate activation synchronization. Rather than requiring synchronized gradient updates across thousands of GPUs, we partition data into semantically coherent clusters where each expert independently optimizes its subset while collectively approximating the full distribution. A lightweight transformer router dynamically selects appropriate experts at inference, achieving generation quality comparable to centrally coordinated baselines. Eliminating synchronization enables training on heterogeneous hardware without specialized interconnects. Empirical validation confirms that Paris's decentralized training maintains generation quality while removing the dedicated GPU cluster requirement for large-scale diffusion models. Paris achieves this using 14$\times$ less training data and 16$\times$ less compute than the prior decentralized baseline.
Authors: Marc Brinner, Sina Zarrie{\ss}
Abstract: We propose SemCSE-Multi, a novel unsupervised framework for generating multifaceted embeddings of scientific abstracts, evaluated in the domains of invasion biology and medicine. These embeddings capture distinct, individually specifiable aspects in isolation, thus enabling fine-grained and controllable similarity assessments as well as adaptive, user-driven visualizations of scientific domains. Our approach relies on an unsupervised procedure that produces aspect-specific summarizing sentences and trains embedding models to map semantically related summaries to nearby positions in the embedding space. We then distill these aspect-specific embedding capabilities into a unified embedding model that directly predicts multiple aspect embeddings from a scientific abstract in a single, efficient forward pass. In addition, we introduce an embedding decoding pipeline that decodes embeddings back into natural language descriptions of their associated aspects. Notably, we show that this decoding remains effective even for unoccupied regions in low-dimensional visualizations, thus offering vastly improved interpretability in user-centric settings.
Authors: Xue Bin Peng
Abstract: MimicKit is an open-source framework for training motion controllers using motion imitation and reinforcement learning. The codebase provides implementations of commonly-used motion-imitation techniques and RL algorithms. This framework is intended to support research and applications in computer graphics and robotics by providing a unified training framework, along with standardized environment, agent, and data structures. The codebase is designed to be modular and easily configurable, enabling convenient modification and extension to new characters and tasks. The open-source codebase is available at: https://github.com/xbpeng/MimicKit.
Authors: Agnideep Aich, Ashit Baran Aich
Abstract: Kernel Stein discrepancies (KSDs) are widely used for goodness-of-fit testing, but standard KSDs can be insensitive to higher-order dependence features such as tail dependence. We introduce the Copula-Stein Discrepancy (CSD), which defines a Stein operator directly on the copula density to target dependence geometry rather than the joint score. For Archimedean copulas, CSD admits a closed-form Stein kernel derived from the scalar generator. We prove that CSD metrizes weak convergence of copula distributions, admits an empirical estimator with minimax-optimal rate $O_P(n^{-1/2})$, and is sensitive to differences in tail dependence coefficients. We further extend the framework beyond Archimedean families to general copulas, including elliptical and vine constructions. Computationally, exact CSD kernel evaluation is linear in dimension, and a random-feature approximation reduces the quadratic $O(n^2)$ sample scaling to near-linear $\tilde{O}(n)$; experiments show near-nominal Type~I error, increasing power, and rapid concentration of the approximation toward the exact $\widehat{\mathrm{CSD}}_n^2$ as the number of features grows.
Authors: Jacob Carlson
Abstract: Social scientists are increasingly turning to unstructured datasets to unlock new empirical insights, e.g., estimating descriptive statistics of or causal effects on quantitative measures derived from text, audio, or video data. In many such settings, unsupervised analysis is of primary interest, in that the researcher does not want to (or cannot) pre-specify all important aspects of the unstructured data to measure; they are interested in "discovery." This paper proposes a general and flexible framework for pursuing discovery from unstructured data in a statistically principled way. The framework leverages recent methods from the literature on machine learning interpretability to map unstructured data points to high-dimensional, sparse, and interpretable "dictionaries" of concepts; computes statistics of dictionary entries for testing relevant concept-level hypotheses; performs selective inference on these hypotheses using algorithms validated by new results in high-dimensional central limit theory, producing a selected set ("discoveries"); and both generates and evaluates human-interpretable natural language descriptions of these discoveries. The proposed framework has few researcher degrees of freedom, is fully replicable, and is cheap to implement -- both in terms of financial cost and researcher time. Applications to recent descriptive and causal analyses of unstructured data in empirical economics are explored. An open source Jupyter notebook is provided for researchers to implement the framework in their own projects.
Authors: Renren Jin, Pengzhi Gao, Yuqi Ren, Zhuowen Han, Tongxuan Zhang, Wuwei Huang, Wei Liu, Jian Luan, Deyi Xiong
Abstract: Reinforcement learning with verifiable rewards (RLVR) has emerged as a prominent paradigm for enhancing the reasoning capabilities of large language models (LLMs). However, the entropy of LLMs usually collapses during RLVR training, leading to premature convergence to suboptimal local minima and hindering further performance improvement. Although various approaches have been proposed to mitigate entropy collapse, a comprehensive study of entropy in RLVR remains lacking. To bridge this gap, we conduct extensive experiments to investigate the entropy dynamics of LLMs trained with RLVR and analyze how model entropy correlates with response diversity, calibration, and performance across various benchmarks. Our results identify three key factors that influence entropy: the clipping thresholds in the optimization objective, the number of off-policy updates, and the diversity of the training data. Furthermore, through both theoretical analysis and empirical validation, we demonstrate that tokens with positive advantages are the primary drivers of entropy collapse. Motivated by this insight, we propose Positive-Advantage Reweighting, a simple yet effective approach that regulates model entropy by adjusting the loss weights assigned to tokens with positive advantages during RLVR training, while maintaining competitive performance.
Authors: Gordon Ma, Dimitris G. Angelakis
Abstract: Qubit-efficient optimization studies how large combinatorial problems can be addressed with quantum circuits whose width is far smaller than the number of logical variables. In quadratic unconstrained binary optimization (QUBO), objective values depend only on one- and two-body statistics, yet standard variational algorithms explore exponentially large Hilbert spaces. We recast qubit-efficient optimization as a geometric question: what is the minimal representation the objective itself requires? Focusing on QUBO problems, we show that enforcing mutual consistency among pairwise statistics defines a convex body -- the level-2 Sherali-Adams polytope -- that captures the information on which quadratic objectives depend. We operationalize this geometry in a minimal variational pipeline that separates representation, consistency, and decoding: a logarithmic-width circuit produces pairwise moments, a differentiable information projection enforces local feasibility, and a maximum-entropy ensemble provides a principled global decoder. This information-minimal construction achieves near-optimal approximation ratios on large unweighted Max-Cut instances (up to N=2000) at shallow depth, indicating that pairwise polyhedral geometry already captures the relevant structure in this regime. By making the information-minimal geometry explicit, this work establishes a clean baseline for qubit-efficient optimization and sharpens the question of where genuinely quantum structure becomes necessary.
Authors: Ryoma Yataka, Pu Perry Wang, Petros Boufounos, Ryuhei Takahashi
Abstract: Multi-view indoor radar perception has drawn attention due to its cost-effectiveness and low privacy risks. Existing methods often rely on {implicit} cross-view radar feature association, such as proposal pairing in RFMask or query-to-feature cross-attention in RETR, which can lead to ambiguous feature matches and degraded detection in complex indoor scenes. To address these limitations, we propose \textbf{REXO} (multi-view Radar object dEtection with 3D bounding boX diffusiOn), which lifts the 2D bounding box (BBox) diffusion process of DiffusionDet into the 3D radar space. REXO utilizes these noisy 3D BBoxes to guide an {explicit} cross-view radar feature association, enhancing the cross-view radar-conditioned denoising process. By accounting for prior knowledge that the person is in contact with the ground, REXO reduces the number of diffusion parameters by determining them from this prior. Evaluated on two open indoor radar datasets, our approach surpasses state-of-the-art methods by a margin of +4.22 AP on the HIBER dataset and +11.02 AP on the MMVR dataset. The REXO implementation is available at https://github.com/merlresearch/radar-bbox-diffusion.
Authors: Alexander Goslin
Abstract: For decades, procedural worlds have been built on procedural noise functions such as Perlin noise, which are fast and infinite, yet fundamentally limited in realism and large-scale coherence. We introduce Terrain Diffusion, a generative framework that bridges the fidelity of diffusion models with the properties that made procedural noise indispensable: seamless infinite extent, seed-consistency, and constant-time random access. At its core is InfiniteDiffusion, a novel algorithm for infinite generation that reformulates standard diffusion sampling for unbounded domains. While noise functions remain near-instant, our framework outpaces orbital velocity by 9 times on a consumer GPU, enabling realistic terrain generation at interactive rates. We integrate a hierarchical stack of diffusion models to couple planetary context with local detail, a compact Laplacian encoding to stabilize outputs across Earth-scale dynamic ranges, and an open-source infinite-tensor framework for constant-memory manipulation of unbounded tensors. Together, these components position diffusion models as a practical, scalable foundation for the next generation of infinite virtual worlds.
Authors: Brandon Robinson
Abstract: Neural Network Field Theories (NN-FTs) can realize global conformal symmetries via embedding space architectures. These models describe Generalized Free Fields (GFFs) in the infinite width limit. However, they typically lack a local stress-energy tensor satisfying conformal Ward identities. This presents an obstruction to realizing infinite-dimensional, local conformal symmetry typifying 2d Conformal Field Theories (CFTs). We present the first construction of an NN-FT that encodes the full Virasoro symmetry of a 2d CFT. We formulate a neural free boson theory with a local stress tensor $T(z)$ by properly choosing the architecture and prior distribution of network parameters. We verify the analytical results through numerical simulation; computing the central charge and the scaling dimensions of vertex operators. We then construct an NN realization of a Majorana Fermion and an $\mathcal{N}=(1,1)$ scalar multiplet, which then enables an extension of the formalism to include super-Virasoro symmetry. Finally, we extend the framework by constructing boundary NN-FTs that preserve (super-)conformal symmetry via the method of images.
Authors: Osvaldo Simeone
Abstract: The rapid growth of artificial intelligence (AI) has brought novel data processing and generative capabilities but also escalating energy requirements. This challenge motivates renewed interest in neuromorphic computing principles, which promise brain-like efficiency through discrete and sparse activations, recurrent dynamics, and non-linear feedback. In fact, modern AI architectures increasingly embody neuromorphic principles through heavily quantized activations, state-space dynamics, and sparse attention mechanisms. This paper elaborates on the connections between neuromorphic models, state-space models, and transformer architectures through the lens of the distinction between intra-token processing and inter-token processing. Most early work on neuromorphic AI was based on spiking neural networks (SNNs) for intra-token processing, i.e., for transformations involving multiple channels, or features, of the same vector input, such as the pixels of an image. In contrast, more recent research has explored how neuromorphic principles can be leveraged to design efficient inter-token processing methods, which selectively combine different information elements depending on their contextual relevance. Implementing associative memorization mechanisms, these approaches leverage state-space dynamics or sparse self-attention. Along with a systematic presentation of modern neuromorphic AI models through the lens of intra-token and inter-token processing, training methodologies for neuromorphic AI models are also reviewed. These range from surrogate gradients leveraging parallel convolutional processing to local learning rules based on reinforcement learning mechanisms.
Authors: Jinniao Qiu
Abstract: In this work, we investigate a stochastic control framework for global optimization over both Euclidean spaces and the Wasserstein space of probability measures, where the objective function may be non-convex and/or non-differentiable. In the Euclidean setting, the original minimization problem is approximated by a family of regularized stochastic control problems; using dynamic programming, we analyze the associated Hamilton--Jacobi--Bellman equations and obtain tractable representations via the Cole--Hopf transformation and the Feynman--Kac formula. For optimization over probability measures, we formulate a regularized mean-field control problem characterized by a master equation, and further approximate it by controlled $N$-particle systems. We establish that, as the regularization parameter tends to zero (and as the particle number tends to infinity for the optimization over probability measures), the value of the control problem converges to the global minimum of the original objective. Building on the resulting probabilistic representations, Monte Carlo-based numerical schemes are proposed and numerical experiments are reported to illustrate the effectiveness of the methods and to support the theoretical convergence rates.
Authors: Viacheslav Siniaev, Iaroslav Chelombitko, Aleksey Komissarov
Abstract: Large Language Models (LLMs) have demonstrated exceptional code generation capabilities, yet their token-level mechanisms remain underexplored, particularly in compressed models. Through systematic analysis of programming language token representations, we characterize how programming languages are encoded in LLM tokenizers by analyzing their vocabulary distribution and keyword coverage patterns. We introduce a novel cold-start probability analysis method that provides insights into model behavior without requiring explicit prompts. Additionally, we present a comprehensive evaluation of how different model optimization techniques - including quantization, distillation, model scaling, and task-specific fine-tuning - affect token-level representations and code generation quality. Our experiments, supported by comprehensive probability distribution analysis and evaluation metrics, reveal critical insights into token-level behavior and provide empirically-validated guidelines for maintaining code generation quality under various optimization constraints. These findings advance both theoretical understanding of LLM code generation and practical implementation of optimized models in production environments.
Authors: Ali Kashefi
Abstract: We present two novel generative geometric deep learning frameworks, termed Flow Matching PointNet and Diffusion PointNet, for predicting fluid flow variables on irregular geometries by incorporating PointNet into flow matching and diffusion models, respectively. In these frameworks, a reverse generative process reconstructs physical fields from standard Gaussian noise conditioned on unseen geometries. The proposed approaches operate directly on point-cloud representations of computational domains (e.g., grid vertices of finite-volume meshes) and therefore avoid the limitations of pixelation used to project geometries onto uniform lattices, as is common in U-Net-based flow matching and diffusion models. In contrast to graph neural network-based diffusion models, Flow Matching PointNet and Diffusion PointNet do not exhibit high-frequency noise artifacts in the predicted fields. Moreover, unlike such approaches, which require auxiliary intermediate networks to condition geometry, the proposed frameworks rely solely on PointNet, resulting in a simple and unified architecture. The performance of the proposed frameworks is evaluated on steady incompressible flow past a cylinder, using a geometric dataset constructed by varying the cylinder's cross-sectional shape and orientation across samples. The results demonstrate that Flow Matching PointNet and Diffusion PointNet achieve more accurate predictions of velocity and pressure fields, as well as lift and drag forces, and exhibit greater robustness to incomplete geometries compared to a vanilla PointNet with the same number of trainable parameters.
Authors: Zhibo Hu, Chen Wang, Yanfeng Shu, Hye-young Paik, Liming Zhu
Abstract: Earlier research has shown that metaphors influence human's decision making, which raises the question of whether metaphors also influence large language models (LLMs)' reasoning pathways, considering their training data contain a large number of metaphors. In this work, we investigate the problem in the scope of the emergent misalignment problem where LLMs can generalize patterns learned from misaligned content in one domain to another domain. We discover a strong causal relationship between metaphors in training data and the misalignment degree of LLMs' reasoning contents. With interventions using metaphors in pre-training, fine-tuning and re-alignment phases, models' cross-domain misalignment degrees change significantly. As we delve deeper into the causes behind this phenomenon, we observe that there is a connection between metaphors and the activation of global and local latent features of large reasoning models. By monitoring these latent features, we design a detector that predict misaligned content with high accuracy.
Authors: Nikhil Anand, Shwetha Somasundaram, Anirudh Phukan, Apoorv Saxena, Koyel Mukherjee
Abstract: Large Language Models (LLMs) encode vast amounts of parametric knowledge during pre-training. As world knowledge evolves, effective deployment increasingly depends on their ability to faithfully follow externally retrieved context. When such evidence conflicts with the model's internal knowledge, LLMs often default to memorized facts, producing unfaithful outputs. In this work, we introduce ContextFocus, a lightweight activation steering approach that improves context faithfulness in such knowledge-conflict settings while preserving fluency and efficiency. Unlike prior approaches, our solution requires no model finetuning and incurs minimal inference-time overhead, making it highly efficient. We evaluate ContextFocus on the ConFiQA benchmark, comparing it against strong baselines including ContextDPO, COIECD, and prompting-based methods. Furthermore, we show that our method is complementary to prompting strategies and remains effective on larger models. Extensive experiments show that ContextFocus significantly improves contextual-faithfulness. Our results highlight the effectiveness, robustness, and efficiency of ContextFocus in improving contextual-faithfulness of LLM outputs.
Authors: Dongqi Liu, Hang Ding, Qiming Feng, Jian Li, Xurong Xie, Zhucun Xue, Chengjie Wang, Jiangning Zhang, Yabiao Wang
Abstract: Retrieval-Augmented Generation (RAG) has emerged as an important means of enhancing the performance of large language models (LLMs) in knowledge-intensive tasks. However, most existing RAG strategies treat retrieved passages in a flat and unstructured way, which prevents the model from capturing structural cues and constrains its ability to synthesize knowledge from dispersed evidence across documents. To overcome these limitations, we propose Disco-RAG, a discourse-aware framework that explicitly injects discourse signals into the generation process. Our method constructs intra-chunk discourse trees to capture local hierarchies and builds inter-chunk rhetorical graphs to model cross-passage coherence. These structures are jointly integrated into a planning blueprint that conditions the generation. Experiments on question answering and long-document summarization benchmarks show the efficacy of our approach. Disco-RAG achieves state-of-the-art results on the benchmarks without fine-tuning. These findings underscore the important role of discourse structure in advancing RAG systems.
Authors: Osman Goni Ridwan, Gilles Frapper, Hongfei Xue, Qiang Zhu
Abstract: We present a materials generation framework that couples a symmetry-conditioned variational autoencoder (CVAE) with a differentiable SO(3) power spectrum objective to steer candidates toward a specified local environment under the crystallographic constraints. In particular, we implement a fully differentiable pipeline to enable batch-wise optimization on both direct and latent crystallographic representations. Using the GPU acceleration, this implementation achieves about fivefold speed compared to our previous CPU workflow, while yielding comparable outcomes. In addition, we introduce the optimization strategy that alternatively performs optimization on the direct and latent crystal representations. This dual-level relaxation approach can effectively escape local minima defined by different objective gradients, thus increasing the success rate of generating complex structures satisfying the target local environments. This framework can be extended to systems consisting of multi-components and multi-environments, providing a scalable route to generate material structures with the target local environment.
Authors: Zuhair Ahmed Khan Taha, Mohammed Mudassir Uddin, Shahnawaz Alam
Abstract: Large language models hold considerable promise for various applications, but their computational requirements create a barrier that many institutions cannot overcome. A single session using a 70-billion-parameter model can cost around $127 in cloud computing fees, which puts these tools out of reach for organizations operating on limited budgets. We present AgentCompress, a framework that tackles this problem through task-aware dynamic compression. The idea comes from a simple observation: not all tasks require the same computational effort. Complex reasoning, for example, is far more demanding than text reformatting, yet conventional compression applies the same reduction to both. Our approach uses a lightweight neural controller that looks at the first few tokens of each request, estimates how complex the task will be, and sends it to an appropriately quantized version of the model. This routing step adds only about 12 milliseconds of overhead. We tested the framework on 290 multi-stage workflows from domains including computer science, physics, chemistry, and biology. The results show a 68.3% reduction in computational costs while preserving 96.2% of the original success rate. These findings suggest that routing queries intelligently can make powerful language models substantially more affordable without sacrificing output quality
Authors: Yoav Evron, Michal Bar-Asher Siegal, Michael Fire
Abstract: The recent Artificial Intelligence (AI) revolution has opened transformative possibilities for the humanities, particularly in unlocking the visual-artistic content embedded in historical illuminated manuscripts. While digital archives now offer unprecedented access to these materials, the ability to systematically locate, extract, and analyze illustrations at scale remains a major challenge. We present a general and scalable AI-based pipeline for large-scale visual analysis of illuminated manuscripts. The framework integrates modern deep-learning models for page-level illustration detection, illustration extraction, and multimodal description, enabling scholars to search, cluster, and study visual materials and artistic trends across entire corpora. We demonstrate the applicability of this approach on large heterogeneous collections, including the Vatican Library and richly illuminated manuscripts such as the Bible of Borso d'Este. The system reveals meaningful visual patterns and cross-manuscript relationships by embedding illustrations into a shared representation space and analyzing their similarity structure (see figure 4). By harnessing recent advances in computer vision and vision-language models, our framework enables new forms of large-scale visual scholarship in historical studies, art history, and cultural heritage making it possible to explore iconography, stylistic trends, and cultural connections in ways that were previously impractical.