Authors: Minseo Kwak, Jaehyung Kim
Abstract: The opacity of massive pretraining corpora in Large Language Models (LLMs) raises significant privacy and copyright concerns, making pretraining data detection a critical challenge. Existing state-of-the-art methods typically rely on token likelihoods, yet they often overlook the divergence from the model's top-1 prediction and local correlation between adjacent tokens. In this work, we propose Gap-K%, a novel pretraining data detection method grounded in the optimization dynamics of LLM pretraining. By analyzing the next-token prediction objective, we observe that discrepancies between the model's top-1 prediction and the target token induce strong gradient signals, which are explicitly penalized during training. Motivated by this, Gap-K% leverages the log probability gap between the top-1 predicted token and the target token, incorporating a sliding window strategy to capture local correlations and mitigate token-level fluctuations. Extensive experiments on the WikiMIA and MIMIR benchmarks demonstrate that Gap-K% achieves state-of-the-art performance, consistently outperforming prior baselines across various model sizes and input lengths.
Authors: Adnan Ahmad, Chiara Boldrini, Lorenzo Valerio, Andrea Passarella, Marco Conti
Abstract: Decentralized Federated Learning (DFL) is a serverless collaborative machine learning paradigm where devices collaborate directly with neighbouring devices to exchange model information for learning a generalized model. However, variations in individual experiences and different levels of device interactions lead to data and model initialization heterogeneities across devices. Such heterogeneities leave variations in local model parameters across devices that leads to slower convergence. This paper tackles the data and model heterogeneity by explicitly addressing the parameter level varying evidential credence across local models. A novel aggregation approach is introduced that captures these parameter variations in local models and performs robust aggregation of neighbourhood local updates. Specifically, consensus weights are generated via approximation of second-order information of local models on their local datasets. These weights are utilized to scale neighbourhood updates before aggregating them into global neighbourhood representation. In extensive experiments with computer vision tasks, the proposed approach shows strong generalizability of local models at reduced communication costs.
Authors: Hyunmin Kim, Yukun Zhou, Rahul A. Jonas, Lie Ju, Sunjin Hwang, Pearse A. Keane, Siegfried K. Wagner
Abstract: Oculomics - the concept of predicting systemic diseases, such as cardiovascular disease and dementia, through retinal imaging - has advanced rapidly due to the data efficiency of transformer-based foundation models like RETFound. Image-level mixed sample data augmentations, such as CutMix and MixUp, are frequently used for training transformers, yet these techniques perturb patient-specific attributes, such as medical comorbidity and clinical factors, since they only account for images and labels. To address this limitation, we propose a hierarchical sampling strategy, Oculomix, for mixed sample augmentations. Our method is based on two clinical priors. First (exam level), images acquired from the same patient at the same time point share the same attributes. Second (patient level), images acquired from the same patient at different time points have a soft temporal trend, as morbidity generally increases over time. Guided by these priors, our method constrains the mixing space to the patient and exam levels to better preserve patient-specific characteristics and leverages their hierarchical relationships. The proposed method is validated using ViT models on a five-year prediction of major adverse cardiovascular events (MACE) in a large ethnically diverse population (Alzeye). We show that Oculomix consistently outperforms image-level CutMix and MixUp by up to 3% in AUROC, demonstrating the necessity and value of the proposed method in oculomics.
Authors: Tobias Habermann, Michael Mecik, Zhenyu Wang, C\'esar David Vera, Martin Kumm, Mario Garrido
Abstract: Among hardware accelerators for deep-learning inference, data flow implementations offer low latency and high throughput capabilities. In these architectures, each neuron is mapped to a dedicated hardware unit, making them well-suited for field-programmable gate array (FPGA) implementation. Previous unrolled implementations mostly focus on fully connected networks because of their simplicity, although it is well known that convolutional neural networks (CNNs) require fewer computations for the same accuracy. When observing the data flow in CNNs, pooling layers and convolutional layers with a stride larger than one, the number of data at their output is reduced with respect to their input. This data reduction strongly affects the data rate in a fully parallel implementation, making hardware units heavily underutilized unless it is handled properly. This work addresses this issue by analyzing the data flow of CNNs and presents a novel approach to designing data-rate-aware, continuous-flow CNN architectures. The proposed approach ensures a high hardware utilization close to 100% by interleaving low data rate signals and sharing hardware units, as well as using the right parallelization to achieve the throughput of a fully parallel implementation. The results show that a significant amount of the arithmetic logic can be saved, which allows implementing complex CNNs like MobileNet on a single FPGA with high throughput.
Authors: Faruk Alpay, Bugra Kilictas
Abstract: We study the emergence of multi-step reasoning in deep Transformer language models through a geometric and statistical-physics lens. Treating the hidden-state trajectory as a flow on an implicit Riemannian manifold, we analyze the layerwise covariance spectrum of activations, where $C^{(\ell)}=\mathbb{E}[h^{(\ell)}h^{(\ell)\top}]$, and track deviations from a random-matrix bulk. Across model scales (1.5B--30B), we observe a sharp reduction in effective dimensionality consistent with a phase transition: an order parameter based on sparsity/localization, $\Omega(h)=1-\|h\|_1/(\sqrt{d}\|h\|_2)$, exhibits a discontinuity near a critical normalized depth $\gamma_c\approx 0.42$ in sufficiently large models. We formalize the forward pass as a discrete coarse-graining map and relate the appearance of stable "concept basins" to fixed points of this renormalization-like dynamics. The resulting low-entropy regime is characterized by a spectral tail collapse and by the formation of transient, reusable object-like structures in representation space, which we call Transient Class Objects (TCOs). We provide theoretical conditions connecting logical separability to spectral decay and validate the predicted signatures with layerwise probes on multiple open-weight model families.
Authors: Yuhao Li
Abstract: How can populations of learners develop coordinated, diverse behaviors without explicit communication or diversity incentives? We demonstrate that competition alone is sufficient to induce emergent specialization -- learners spontaneously partition into specialists for different environmental regimes through competitive dynamics, consistent with ecological niche theory. We introduce the NichePopulation algorithm, a simple mechanism combining competitive exclusion with niche affinity tracking. Validated across six real-world domains (cryptocurrency trading, commodity prices, weather forecasting, solar irradiance, urban traffic, and air quality), our approach achieves a mean Specialization Index of 0.75 with effect sizes of Cohen's d > 20. Key findings: (1) At lambda=0 (no niche bonus), learners still achieve SI > 0.30, proving specialization is genuinely emergent; (2) Diverse populations outperform homogeneous baselines by +26.5% through method-level division of labor; (3) Our approach outperforms MARL baselines (QMIX, MAPPO, IQL) by 4.3x while being 4x faster.
Authors: Valery Manokhin, Daniel Gr{\o}nhaug
Abstract: We study model-agnostic post-hoc calibration methods intended to improve probabilistic predictions in supervised binary classification on real i.i.d. tabular data, with particular emphasis on conformal and Venn-based approaches that provide distribution-free validity guarantees under exchangeability. We benchmark 21 widely used classifiers, including linear models, SVMs, tree ensembles (CatBoost, XGBoost, LightGBM), and modern tabular neural and foundation models, on binary tasks from the TabArena-v0.1 suite using randomized, stratified five-fold cross-validation with a held-out test fold. Five calibrators; Isotonic regression, Platt scaling, Beta calibration, Venn-Abers predictors, and Pearsonify are trained on a separate calibration split and applied to test predictions. Calibration is evaluated using proper scoring rules (log-loss and Brier score) and diagnostic measures (Spiegelhalter's Z, ECE, and ECI), alongside discrimination (AUC-ROC) and standard classification metrics. Across tasks and architectures, Venn-Abers predictors achieve the largest average reductions in log-loss, followed closely by Beta calibration, while Platt scaling exhibits weaker and less consistent effects. Beta calibration improves log-loss most frequently across tasks, whereas Venn-Abers displays fewer instances of extreme degradation and slightly more instances of extreme improvement. Importantly, we find that commonly used calibration procedures, most notably Platt scaling and isotonic regression, can systematically degrade proper scoring performance for strong modern tabular models. Overall classification performance is often preserved, but calibration effects vary substantially across datasets and architectures, and no method dominates uniformly. In expectation, all methods except Pearsonify slightly increase accuracy, but the effect is marginal, with the largest expected gain about 0.008%.
Authors: Jiayu Xu, Junbiao Pang
Abstract: Learning from Noisy Labels (LNL) presents a fundamental challenge in deep learning, as real-world datasets often contain erroneous or corrupted annotations, \textit{e.g.}, data crawled from Web. Current research focuses on sophisticated label correction mechanisms. In contrast, this paper adopts a novel perspective by establishing a theoretical analysis the relationship between flatness of the loss landscape and the presence of label noise. In this paper, we theoretically demonstrate that carefully simulated label noise synergistically enhances both the generalization performance and robustness of label noises. Consequently, we propose Noise-Compensated Sharpness-aware Minimization (NCSAM) to leverage the perturbation of Sharpness-Aware Minimization (SAM) to remedy the damage of label noises. Our analysis reveals that the testing accuracy exhibits a similar behavior that has been observed on the noise-clear dataset. Extensive experimental results on multiple benchmark datasets demonstrate the consistent superiority of the proposed method over existing state-of-the-art approaches on diverse tasks.
Authors: Ibrahim Albulushi, Saleh Bunaiyan, Suraj S. Cheema, Hesham ElSawy, Feras Al-Dirini
Abstract: Extending the intelligence of sensors to the data-acquisition process - deciding whether to sample or not - can result in transformative energy-efficiency gains. However, making such a decision in a deterministic manner involves risk of losing information. Here we present a sensing paradigm that enables making such a decision in a probabilistic manner. The paradigm takes inspiration from the autonomous nervous system and employs a probabilistic neuron (p-neuron) driven by an analog feature extraction circuit. The response time of the system is on the order of microseconds, over-coming the sub-sampling-rate response time limit and enabling real-time intelligent autonomous activation of data-sampling. Validation experiments on active seismic survey data demonstrate lossless probabilistic data acquisition, with a normalized mean squared error of 0.41%, and 93% saving in the active operation time of the system and the number of generated samples.
Authors: Huanlin Gao, Ping Chen, Fuyuan Shi, Ruijia Wu, Li YanTao, Qiang Hui, Yuren You, Ting Lu, Chao Tan, Shaoan Zhao, Zhaoxiang Liu, Fang Zhao, Kai Wang, Shiguo Lian
Abstract: We present MeanCache, a training-free caching framework for efficient Flow Matching inference. Existing caching methods reduce redundant computation but typically rely on instantaneous velocity information (e.g., feature caching), which often leads to severe trajectory deviations and error accumulation under high acceleration ratios. MeanCache introduces an average-velocity perspective: by leveraging cached Jacobian--vector products (JVP) to construct interval average velocities from instantaneous velocities, it effectively mitigates local error accumulation. To further improve cache timing and JVP reuse stability, we develop a trajectory-stability scheduling strategy as a practical tool, employing a Peak-Suppressed Shortest Path under budget constraints to determine the schedule. Experiments on FLUX.1, Qwen-Image, and HunyuanVideo demonstrate that MeanCache achieves 4.12X and 4.56X and 3.59X acceleration, respectively, while consistently outperforming state-of-the-art caching baselines in generation quality. We believe this simple yet effective approach provides a new perspective for Flow Matching inference and will inspire further exploration of stability-driven acceleration in commercial-scale generative models.
Authors: Canyang Zhao, Bolin Peng, J. Patrick Mayo, Ce Ju, Bing Liu
Abstract: Cross-session nonstationarity in neural activity recorded by implanted electrodes is a major challenge for invasive Brain-computer interfaces (BCIs), as decoders trained on data from one session often fail to generalize to subsequent sessions. This issue is further exacerbated in practice, as retraining or adapting decoders becomes particularly challenging when only limited data are available from a new session. To address this challenge, we propose a Task-Conditioned Latent Alignment framework (TCLA) for cross-session neural decoding. Building upon an autoencoder architecture, TCLA first learns a low-dimensional representation of neural dynamics from a source session with sufficient data. For target sessions with limited data, TCLA then aligns target latent representations to the source in a task-conditioned manner, enabling effective transfer of learned neural dynamics. We evaluate TCLA on the macaque motor and oculomotor center-out dataset. Compared to baseline methods trained solely on target-session data, TCLA consistently improves decoding performance across datasets and decoding settings, with gains in the coefficient of determination of up to 0.386 for y coordinate velocity decoding in a motor dataset. These results suggest that TCLA provides an effective strategy for transferring knowledge from source to target sessions, enabling more robust neural decoding under conditions with limited data.
Authors: Mingxuan Luo, Guipeng Xv, Sishuo Chen, Xinyu Li, Li Zhang, Zhangming Chan, Xiang-Rong Sheng, Han Zhu, Jian Xu, Bo Zheng, Chen Lin
Abstract: In industrial recommender systems, conversion rate (CVR) is widely used for traffic allocation, but it fails to fully reflect recommendation effectiveness because it ignores refund behavior. To better capture true user satisfaction and business value, net conversion rate (NetCVR), defined as the probability that a clicked item is purchased and not refunded, has been proposed.Unlike CVR, NetCVR prediction involves a more complex multi-stage cascaded delayed feedback process. The two cascaded delays from click to conversion and from conversion to refund have opposite effects, making traditional CVR modeling methods inapplicable. Moreover, the lack of open-source datasets and online continuous training schemes further hinders progress in this area.To address these challenges, we introduce CASCADE (Cascaded Sequences of Conversion and Delayed Refund), the first large-scale open dataset derived from the Taobao app for online continuous NetCVR prediction. Through an in-depth analysis of CASCADE, we identify three key insights: (1) NetCVR exhibits strong temporal dynamics, necessitating online continuous modeling; (2) cascaded modeling of CVR and refund rate outperforms direct NetCVR modeling; and (3) delay time, which correlates with both CVR and refund rate, is an important feature for NetCVR prediction.Based on these insights, we propose TESLA, a continuous NetCVR modeling framework featuring a CVR-refund-rate cascaded architecture, stage-wise debiasing, and a delay-time-aware ranking loss. Extensive experiments demonstrate that TESLA consistently outperforms state-of-the-art methods on CASCADE, achieving absolute improvements of 12.41 percent in RI-AUC and 14.94 percent in RI-PRAUC on NetCVR prediction. The code and dataset are publicly available at https://github.com/alimama-tech/NetCVR.
Authors: Jinlin Liu, Wei Chen, Xiaojin Zhang
Abstract: Collecting web data to train deep models has become increasingly common, raising concerns about unauthorized data usage. To mitigate this issue, unlearnable examples introduce imperceptible perturbations into data, preventing models from learning effectively. However, existing methods typically rely on deep neural networks as surrogate models for perturbation generation, resulting in significant computational costs. In this work, we propose Perturbation-Induced Linearization (PIL), a computationally efficient yet effective method that generates perturbations using only linear surrogate models. PIL achieves comparable or better performance than existing surrogate-based methods while reducing computational time dramatically. We further reveal a key mechanism underlying unlearnable examples: inducing linearization to deep models, which explains why PIL can achieve competitive results in a very short time. Beyond this, we provide an analysis about the property of unlearnable examples under percentage-based partial perturbation. Our work not only provides a practical approach for data protection but also offers insights into what makes unlearnable examples effective.
Authors: Soham Sarkar, Tanmay Sen, Sayantan Banerjee
Abstract: Industrial image anomaly detection is a challenging problem owing to extreme class imbalance and the scarcity of labeled defective samples, particularly in few-shot settings. We propose BayPrAnoMeta, a Bayesian generalization of Proto-MAML for few-shot industrial image anomaly detection. Unlike existing Proto-MAML approaches that rely on deterministic class prototypes and distance-based adaptation, BayPrAnoMeta replaces prototypes with task-specific probabilistic normality models and performs inner-loop adaptation via a Bayesian posterior predictive likelihood. We model normal support embeddings with a Normal-Inverse-Wishart (NIW) prior, producing a Student-$t$ predictive distribution that enables uncertainty-aware, heavy-tailed anomaly scoring and is essential for robustness in extreme few-shot settings. We further extend BayPrAnoMeta to a federated meta-learning framework with supervised contrastive regularization for heterogeneous industrial clients and prove convergence to stationary points of the resulting nonconvex objective. Experiments on the MVTec AD benchmark demonstrate consistent and significant AUROC improvements over MAML, Proto-MAML, and PatchCore-based methods in few-shot anomaly detection settings.
Authors: Chiraag Kaushik, Davis Barch, Andrea Fanelli
Abstract: The Linear Representation Hypothesis asserts that the embeddings learned by neural networks can be understood as linear combinations of features corresponding to high-level concepts. Based on this ansatz, sparse autoencoders (SAEs) have recently become a popular method for decomposing embeddings into a sparse combination of linear directions, which have been shown empirically to often correspond to human-interpretable semantics. However, recent attempts to apply SAEs to multimodal embedding spaces (such as the popular CLIP embeddings for image/text data) have found that SAEs often learn "split dictionaries", where most of the learned sparse features are essentially unimodal, active only for data of a single modality. In this work, we study how to effectively adapt SAEs for the setting of multimodal embeddings while ensuring multimodal alignment. We first argue that the existence of a split dictionary decomposition on an aligned embedding space implies the existence of a non-split dictionary with improved modality alignment. Then, we propose a new SAE-based approach to multimodal embedding decomposition using cross-modal random masking and group-sparse regularization. We apply our method to popular embeddings for image/text (CLIP) and audio/text (CLAP) data and show that, compared to standard SAEs, our approach learns a more multimodal dictionary while reducing the number of dead neurons and improving feature semanticity. We finally demonstrate how this improvement in alignment of concepts between modalities can enable improvements in the interpretability and control of cross-modal tasks.
Authors: Fang Li
Abstract: Despite the ubiquity of tabular data in high-stakes domains, traditional deep learning architectures often struggle to match the performance of gradient-boosted decision trees while maintaining scientific interpretability. Standard neural networks typically treat features as independent entities, failing to exploit the inherent manifold structural dependencies that define tabular distributions. We propose Structural Compositional Function Networks (StructuralCFN), a novel architecture that imposes a Relation-Aware Inductive Bias via a differentiable structural prior. StructuralCFN explicitly models each feature as a mathematical composition of its counterparts through Differentiable Adaptive Gating, which automatically discovers the optimal activation physics (e.g., attention-style filtering vs. inhibitory polarity) for each relationship. Our framework enables Structured Knowledge Integration, allowing domain-specific relational priors to be injected directly into the architecture to guide discovery. We evaluate StructuralCFN across a rigorous 10-fold cross-validation suite on 18 benchmarks, demonstrating statistically significant improvements (p < 0.05) on scientific and clinical datasets (e.g., Blood Transfusion, Ozone, WDBC). Furthermore, StructuralCFN provides Intrinsic Symbolic Interpretability: it recovers the governing "laws" of the data manifold as human-readable mathematical expressions while maintaining a compact parameter footprint (300--2,500 parameters) that is over an order of magnitude (10x--20x) smaller than standard deep baselines.
Authors: Shih-Hsuan Chiu, Ming-Syan Chen
Abstract: Personalized virtual assistants powered by large language models (LLMs) on edge devices are attracting growing attention, with Retrieval-Augmented Generation (RAG) emerging as a key method for personalization by retrieving relevant profile data and generating tailored responses. However, deploying RAG on edge devices faces efficiency hurdles due to the rapid growth of profile data, such as user-LLM interactions and recent updates. While Computing-in-Memory (CiM) architectures mitigate this bottleneck by eliminating data movement between memory and processing units via in-situ operations, they are susceptible to environmental noise that can degrade retrieval precision. This poses a critical issue in dynamic, multi-domain edge-based scenarios (e.g., travel, medicine, and law) where both accuracy and adaptability are paramount. To address these challenges, we propose Task-Oriented Noise-resilient Embedding Learning (TONEL), a framework that improves noise robustness and domain adaptability for RAG in noisy edge environments. TONEL employs a noise-aware projection model to learn task-specific embeddings compatible with CiM hardware constraints, enabling accurate retrieval under noisy conditions. Extensive experiments conducted on personalization benchmarks demonstrate the effectiveness and practicality of our methods relative to strong baselines, especially in task-specific noisy scenarios.
Authors: Yan Zhang, Xuefeng Liu, Sipeng Chen, Sascha Ranftl, Chong Liu, Shibo Li
Abstract: Standard Bayesian Optimization (BO) assumes uniform smoothness across the search space an assumption violated in multi-regime problems such as molecular conformation search through distinct energy basins or drug discovery across heterogeneous molecular scaffolds. A single GP either oversmooths sharp transitions or hallucinates noise in smooth regions, yielding miscalibrated uncertainty. We propose RAMBO, a Dirichlet Process Mixture of Gaussian Processes that automatically discovers latent regimes during optimization, each modeled by an independent GP with locally-optimized hyperparameters. We derive collapsed Gibbs sampling that analytically marginalizes latent functions for efficient inference, and introduce adaptive concentration parameter scheduling for coarse-to-fine regime discovery. Our acquisition functions decompose uncertainty into intra-regime and inter-regime components. Experiments on synthetic benchmarks and real-world applications, including molecular conformer optimization, virtual screening for drug discovery, and fusion reactor design, demonstrate consistent improvements over state-of-the-art baselines on multi-regime objectives.
Authors: Javier Mencia-Ledo, Mohammad Noaeen, Zahra Shakeri
Abstract: Metastatic castration-resistant prostate cancer (mCRPC) is a highly aggressive disease with poor prognosis and heterogeneous treatment response. In this work, we developed and externally validated a visit-level 180-day mortality risk model using longitudinal data from two Phase III cohorts (n=526 and n=640). Only visits with observable 180-day outcomes were labeled; right-censored cases were excluded from analysis. We compared five candidate architectures: Long Short-Term Memory, Gated Recurrent Unit (GRU), Cox Proportional Hazards, Random Survival Forest (RSF), and Logistic Regression. For each dataset, we selected the smallest risk-threshold that achieved an 85% sensitivity floor. The GRU and RSF models showed high discrimination capabilities initially (C-index: 87% for both). In external validation, the GRU obtained a higher calibration (slope: 0.93; intercept: 0.07) and achieved an PR-AUC of 0.87. Clinical impact analysis showed a median time-in-warning of 151.0 days for true positives (59.0 days for false positives) and 18.3 alerts per 100 patient-visits. Given late-stage frailty or cachexia and hemodynamic instability, permutation importance ranked BMI and systolic blood pressure as the strongest associations. These results suggest that longitudinal routine clinical markers can estimate short-horizon mortality risk in mCRPC and support proactive care planning over a multi-month window.
Authors: Chi-Yao Huang, Khoa Vo, Aayush Atul Verma, Duo Lu, Yezhou Yang
Abstract: Training a single network with multiple objectives often leads to conflicting gradients that degrade shared representations, forcing them into a compromised state that is suboptimal for any single task--a problem we term latent representation collapse. We introduce Domain Expansion, a framework that prevents these conflicts by restructuring the latent space itself. Our framework uses a novel orthogonal pooling mechanism to construct a latent space where each objective is assigned to a mutually orthogonal subspace. We validate our approach across diverse benchmarks--including ShapeNet, MPIIGaze, and Rotated MNIST--on challenging multi-objective problems combining classification with pose and gaze estimation. Our experiments demonstrate that this structure not only prevents collapse but also yields an explicit, interpretable, and compositional latent space where concepts can be directly manipulated.
Authors: Baptiste Debes, Tinne Tuytelaars
Abstract: Gradient-regularized value learning methods improve sample efficiency by leveraging learned models of transition dynamics and rewards to estimate return gradients. However, existing approaches, such as MAGE, struggle in stochastic or noisy environments, limiting their applicability. In this work, we address these limitations by extending distributional reinforcement learning on continuous state-action spaces to model not only the distribution over scalar state-action value functions but also over their gradients. We refer to this approach as Distributional Sobolev Training. Inspired by Stochastic Value Gradients (SVG), our method utilizes a one-step world model of reward and transition distributions implemented via a conditional Variational Autoencoder (cVAE). The proposed framework is sample-based and employs Max-sliced Maximum Mean Discrepancy (MSMMD) to instantiate the distributional Bellman operator. We prove that the Sobolev-augmented Bellman operator is a contraction with a unique fixed point, and highlight a fundamental smoothness trade-off underlying contraction in gradient-aware RL. To validate our method, we first showcase its effectiveness on a simple stochastic reinforcement learning toy problem, then benchmark its performance on several MuJoCo environments.
Authors: Paul Seurin, Dean Price
Abstract: Heat-pipe microreactors (HPMRs) are compact and transportable nuclear power systems exhibiting inherent safety, well-suited for deployment in remote regions where access is limited and reliance on costly fossil fuels is prevalent. In prior work, we developed a design optimization framework that incorporates techno-economic considerations through surrogate modeling and reinforcement learning (RL)-based optimization, focusing solely on minimizing the levelized cost of electricity (LCOE) by using a bottom-up cost estimation approach. In this study, we extend that framework to a multi-objective optimization that uses the Pareto Envelope Augmented with Reinforcement Learning (PEARL) algorithm. The objectives include minimizing both the rod-integrated peaking factor ($F_{\Delta h}$) and LCOE -- subject to safety and operational constraints. We evaluate three cost scenarios: (1) a high-cost axial and drum reflectors, (2) a low-cost axial reflector, and (3) low-cost axial and drum reflectors. Our findings indicate that reducing the solid moderator radius, pin pitch, and drum coating angle -- all while increasing the fuel height -- effectively lowers $F_{\Delta h}$. Across all three scenarios, four key strategies consistently emerged for optimizing LCOE: (1) minimizing the axial reflector contribution when costly, (2) reducing control drum reliance, (3) substituting expensive tri-structural isotropic (TRISO) fuel with axial reflector material priced at the level of graphite, and (4) maximizing fuel burnup. While PEARL demonstrates promise in navigating trade-offs across diverse design scenarios, discrepancies between surrogate model predictions and full-order simulations remain. Further improvements are anticipated through constraint relaxation and surrogate development, constituting an ongoing area of investigation.
Authors: Meng Xin, Sweta Priyadarshi, Jingyu Xin, Bilal Kartal, Aditya Vavre, Asma Kuriparambil Thekkumpate, Zijia Chen, Ameya Sunil Mahabaleshwarkar, Ido Shahaf, Akhiad Bercovich, Kinjal Patel, Suguna Varshini Velury, Chenjie Luo, Zhiyu Cheng, Jenny Chen, Chen-Han Yu, Wei Ping, Oleg Rybakov, Nima Tajbakhsh, Oluwatobi Olabiyi, Dusan Stosic, Di Wu, Song Han, Eric Chung, Sharath Turuvekere Sreenivas, Bryan Catanzaro, Yoshi Suhara, Tijmen Blankevoort, Huizi Mao
Abstract: This technical report presents quantization-aware distillation (QAD) and our best practices for recovering accuracy of NVFP4-quantized large language models (LLMs) and vision-language models (VLMs). QAD distills a full-precision teacher model into a quantized student model using a KL divergence loss. While applying distillation to quantized models is not a new idea, we observe key advantages of QAD for today's LLMs: 1. It shows remarkable effectiveness and stability for models trained through multi-stage post-training pipelines, including supervised fine-tuning (SFT), reinforcement learning (RL), and model merging, where traditional quantization-aware training (QAT) suffers from engineering complexity and training instability; 2. It is robust to data quality and coverage, enabling accuracy recovery without full training data. We evaluate QAD across multiple post-trained models including AceReason Nemotron, Nemotron 3 Nano, Nemotron Nano V2, Nemotron Nano V2 VL (VLM), and Llama Nemotron Super v1, showing consistent recovery to near-BF16 accuracy.
Authors: Juncheng Dong, Moyang Guo, Ethan X. Fang, Zhuoran Yang, Vahid Tarokh
Abstract: Transformer models have achieved remarkable empirical successes, largely due to their in-context learning capabilities. Inspired by this, we explore training an autoregressive transformer for in-context reinforcement learning (ICRL). In this setting, we initially train a transformer on an offline dataset consisting of trajectories collected from various RL tasks, and then fix and use this transformer to create an action policy for new RL tasks. Notably, we consider the setting where the offline dataset contains trajectories sampled from suboptimal behavioral policies. In this case, standard autoregressive training corresponds to imitation learning and results in suboptimal performance. To address this, we propose the Decision Importance Transformer(DIT) framework, which emulates the actor-critic algorithm in an in-context manner. In particular, we first train a transformer-based value function that estimates the advantage functions of the behavior policies that collected the suboptimal trajectories. Then we train a transformer-based policy via a weighted maximum likelihood estimation loss, where the weights are constructed based on the trained value function to steer the suboptimal policies to the optimal ones. We conduct extensive experiments to test the performance of DIT on both bandit and Markov Decision Process problems. Our results show that DIT achieves superior performance, particularly when the offline dataset contains suboptimal historical data.
Authors: David Kin Wai Ho, Arman Fazeli, Mohamad M. Mansour, Louay M. A. Jalloul
Abstract: To advance Polar code design for 6G applications, we develop a reinforcement learning-based universal sequence design framework that is extensible and adaptable to diverse channel conditions and decoding strategies. Crucially, our method scales to code lengths up to $2048$, making it suitable for use in standardization. Across all $(N,K)$ configurations supported in 5G, our approach achieves competitive performance relative to the NR sequence adopted in 5G and yields up to a 0.2 dB gain over the beta-expansion baseline at $N=2048$. We further highlight the key elements that enabled learning at scale: (i) incorporation of physical law constrained learning grounded in the universal partial order property of Polar codes, (ii) exploitation of the weak long term influence of decisions to limit lookahead evaluation, and (iii) joint multi-configuration optimization to increase learning efficiency.
Authors: Jovan Krajevski, Biljana Tojtovska Ribarski
Abstract: Since its introduction, Facebook Prophet has attracted positive attention from both classical statisticians and the Bayesian statistics community. The model provides two built-in inference methods: maximum a posteriori estimation using the L-BFGS-B algorithm, and Markov Chain Monte Carlo (MCMC) sampling via the No-U-Turn Sampler (NUTS). While exploring various time-series forecasting problems using Bayesian inference with Prophet, we encountered limitations stemming from the inability to apply alternative inference techniques beyond those provided by default. Additionally, the fluent API design of Facebook Prophet proved insufficiently flexible for implementing our custom modeling ideas. To address these shortcomings, we developed a complete reimplementation of the Prophet model in PyMC, which enables us to extend the base model and evaluate and compare multiple Bayesian inference methods. In this paper, we present our PyMC-based implementation and analyze in detail the implementation of different Bayesian inference techniques. We consider full MCMC techniques, MAP estimation and Variational inference techniques on a time-series forecasting problem. We discuss in details the sampling approach, convergence diagnostics, forecasting metrics as well as their computational efficiency and detect possible issues which will be addressed in our future work.
Authors: Yuetian Chen, Kaiyuan Zhang, Yuntao Du, Edoardo Stoppa, Charles Fleming, Ashish Kundu, Bruno Ribeiro, Ninghui Li
Abstract: Diffusion Language Models (DLMs) represent a promising alternative to autoregressive language models, using bidirectional masked token prediction. Yet their susceptibility to privacy leakage via Membership Inference Attacks (MIA) remains critically underexplored. This paper presents the first systematic investigation of MIA vulnerabilities in DLMs. Unlike the autoregressive models' single fixed prediction pattern, DLMs' multiple maskable configurations exponentially increase attack opportunities. This ability to probe many independent masks dramatically improves detection chances. To exploit this, we introduce SAMA (Subset-Aggregated Membership Attack), which addresses the sparse signal challenge through robust aggregation. SAMA samples masked subsets across progressive densities and applies sign-based statistics that remain effective despite heavy-tailed noise. Through inverse-weighted aggregation prioritizing sparse masks' cleaner signals, SAMA transforms sparse memorization detection into a robust voting mechanism. Experiments on nine datasets show SAMA achieves 30% relative AUC improvement over the best baseline, with up to 8 times improvement at low false positive rates. These findings reveal significant, previously unknown vulnerabilities in DLMs, necessitating the development of tailored privacy defenses.
Authors: Richard Csaky
Abstract: We present a large autoregressive model for source-space MEG that scales next-token prediction to long context across datasets and scanners: handling a corpus of over 500 hours and thousands of sessions across the three largest MEG datasets. A modified SEANet-style vector-quantizer reduces multichannel MEG into a flattened token stream on which we train a Qwen2.5-VL backbone from scratch to predict the next brain token and to recursively generate minutes of MEG from up to a minute of context. To evaluate long-horizon generation, we introduce three task-matched tests: (i) on-manifold stability via generated-only drift compared to the time-resolved distribution of real sliding windows, and (ii) conditional specificity via correct context versus prompt-swap controls using a neurophysiologically grounded metric set. We train on CamCAN and Omega and run all analyses on held-out MOUS, establishing cross-dataset generalization. Across metrics, generations remain relatively stable over long rollouts and are closer to the correct continuation than swapped controls. Code available at: https://github.com/ricsinaruto/brain-gen.
Authors: Bo Dai, Na Li, Dale Schuurmans
Abstract: Self-supervised learning (SSL) have improved empirical performance by unleashing the power of unlabeled data for practical applications. Specifically, SSL extracts the representation from massive unlabeled data, which will be transferred to a plenty of down streaming tasks with limited data. The significant improvement on diverse applications of representation learning has attracted increasing attention, resulting in a variety of dramatically different self-supervised learning objectives for representation extraction, with an assortment of learning procedures, but the lack of a clear and unified understanding. Such an absence hampers the ongoing development of representation learning, leaving a theoretical understanding missing, principles for efficient algorithm design unclear, and the use of representation learning methods in practice unjustified. The urgency for a unified framework is further motivated by the rapid growth in representation learning methods. In this paper, we are therefore compelled to develop a principled foundation of representation learning. We first theoretically investigate the sufficiency of the representation from a spectral representation view, which reveals the spectral essence of the existing successful SSL algorithms and paves the path to a unified framework for understanding and analysis. Such a framework work also inspires the development of more efficient and easy-to-use representation learning algorithms with principled way in real-world applications.
Authors: Pedro Chumpitaz-Flores, My Duong, Ying Mao, Kaixun Hua
Abstract: Pairwise-constrained clustering augments unsupervised partitioning with side information by enforcing must-link (ML) and cannot-link (CL) constraints between specific samples, yielding labelings that respect known affinities and separations. However, ML and CL constraints add an extra layer of complexity to the clustering problem, with current methods struggling in data scalability, especially in niche applications like quantum or quantum-hybrid clustering. We propose PASS, a pairwise-constraints and ambiguity-driven subset selection framework that preserves ML and CL constraints satisfaction while allowing scalable, high-quality clustering solution. PASS collapses ML constraints into pseudo-points and offers two selectors: a constraint-aware margin rule that collects near-boundary points and all detected CL violations, and an information-geometric rule that scores points via a Fisher-Rao distance derived from soft assignment posteriors, then selects the highest-information subset under a simple budget. Across diverse benchmarks, PASS attains competitive SSE at substantially lower cost than exact or penalty-based methods, and remains effective in regimes where prior approaches fail.
Authors: Jim Maar, Denis Paperno, Callum Stuart McDougall, Neel Nanda
Abstract: Prior work suggests that language models, while trained on next token prediction, show implicit planning behavior: they may select the next token in preparation to a predicted future token, such as a likely rhyming word, as supported by a prior qualitative study of Claude 3.5 Haiku using a cross-layer transcoder. We propose much simpler techniques for assessing implicit planning in language models. With case studies on rhyme poetry generation and question answering, we demonstrate that our methodology easily scales to many models. Across models, we find that the generated rhyme (e.g. "-ight") or answer to a question ("whale") can be manipulated by steering at the end of the preceding line with a vector, affecting the generation of intermediate tokens leading up to the rhyme or answer word. We show that implicit planning is a universal mechanism, present in smaller models than previously thought, starting from 1B parameters. Our methodology offers a widely applicable direct way to study implicit planning abilities of LLMs. More broadly, understanding planning abilities of language models can inform decisions in AI safety and control.
Authors: Penghe Zhang, Naihua Xiu, Houduo Qi
Abstract: Due to the rise of cardinality minimization in optimization, sparse support vector machines (SSVMs) have attracted much attention lately and show certain empirical advantages over convex SVMs. A common way to derive an SSVM is to add a cardinality function such as $\ell_0$-norm to the dual problem of a convex SVM. However, this process lacks theoretical justification. This paper fills the gap by developing a local duality theory for such an SSVM formulation and exploring its relationship with the hinge-loss SVM (hSVM) and the ramp-loss SVM (rSVM). In particular, we prove that the derived SSVM is exactly the dual problem of the 0/1-loss SVM, and the linear representer theorem holds for their local solutions. The local solution of SSVM also provides guidelines on selecting hyperparameters of hSVM and rSVM. {Under specific conditions, we show that a sequence of global solutions of hSVM converges to a local solution of 0/1-loss SVM. Moreover, a local minimizer of 0/1-loss SVM is a local minimizer of rSVM.} This explains why a local solution induced by SSVM outperforms hSVM and rSVM in the prior empirical study. We further conduct numerical tests on real datasets and demonstrate potential advantages of SSVM by working with locally nice solutions proposed in this paper.
Authors: James Amarel, Robyn Miller, Nicolas Hengartner, Benjamin Migliori, Emily Casleton, Alexei Skurikhin, Earl Lawrence, Gerd J. Kunde
Abstract: We study how neural emulators of partial differential equation solution operators internalize physical symmetries by introducing an influence-based diagnostic that measures the propagation of parameter updates between symmetry-related states, defined as the metric-weighted overlap of loss gradients evaluated along group orbits. This quantity probes the local geometry of the learned loss landscape and goes beyond forward-pass equivariance tests by directly assessing whether learning dynamics couple physically equivalent configurations. Applying our diagnostic to autoregressive fluid flow emulators, we show that orbit-wise gradient coherence provides the mechanism for learning to generalize over symmetry transformations and indicates when training selects a symmetry compatible basin. The result is a novel technique for evaluating if surrogate models have internalized symmetry properties of the known solution operator.
Authors: Zeyang Huang, Takanori Fujiwara, Angelos Chatzimparmpas, Wandrille Duchemin, Andreas Kerren
Abstract: We present a new nonlinear dimensionality reduction method, MAPLE, that enhances UMAP by improving manifold modeling. MAPLE employs a self-supervised learning approach to more efficiently encode low-dimensional manifold geometry. Central to this approach are maximum manifold capacity representations (MMCRs), which help untangle complex manifolds by compressing variances among locally similar data points while amplifying variance among dissimilar data points. This design is particularly effective for high-dimensional data with substantial intra-cluster variance and curved manifold structures, such as biological or image data. Our qualitative and quantitative evaluations demonstrate that MAPLE can produce clearer visual cluster separations and finer subcluster resolution than UMAP while maintaining comparable computational cost.
Authors: Alexander Benanti, Xi Han, Hong Qin
Abstract: Numerical techniques for solving partial differential equations (PDEs) are integral for many fields across science and engineering. Such techniques usually involve solving large, sparse linear systems, where preconditioning methods are critical. In recent years, neural methods, particularly graph neural networks (GNNs), have demonstrated their potential through accelerated convergence. Nonetheless, to extract connective structures, existing techniques aggregate discretized system matrices into graphs, and suffer from rank inflation and a suboptimal convergence rate. In this paper, we articulate NeuraLSP, a novel neural preconditioner combined with a novel loss metric that leverages the left singular subspace of the system matrix's near-nullspace vectors. By compressing spectral information into a fixed low-rank operator, our method exhibits both theoretical guarantees and empirical robustness to rank inflation, affording up to a 53% speedup. Besides the theoretical guarantees for our newly-formulated loss function, our comprehensive experimental results across diverse families of PDEs also substantiate the aforementioned theoretical advances.
Authors: Chen Cheng, Ang Li
Abstract: Out-of-distribution (OOD) generalization remains a fundamental challenge in real-world classification, where test distributions often differ substantially from training data. Most existing approaches pursue domain-invariant representations, implicitly assuming that invariance implies reliability. However, features that are invariant across domains are not necessarily causally effective for prediction. In this work, we revisit OOD classification from a causal perspective and propose to evaluate learned representations based on their necessity and sufficiency under distribution shift. We introduce an explicit segment-level framework that directly measures causal effectiveness across domains, providing a more faithful criterion than invariance alone. Experiments on multi-domain benchmarks demonstrate consistent improvements in OOD performance, particularly under challenging domain shifts, highlighting the value of causal evaluation for robust generalization.
Authors: Ioannis Anagnostides, Rohan Chauhan, Ioannis Panageas, Tuomas Sandholm, Jingming Yan
Abstract: Performative prediction captures the phenomenon where deploying a predictive model shifts the underlying data distribution. While simple retraining dynamics are known to converge linearly when the performative effects are weak ($\rho < 1$), the complexity in the regime $\rho > 1$ was hitherto open. In this paper, we establish a sharp phase transition: computing an $\epsilon$-performatively stable point is PPAD-complete -- and thus polynomial-time equivalent to Nash equilibria in general-sum games -- even when $\rho = 1 + O(\epsilon)$. This intractability persists even in the ostensibly simple setting with a quadratic loss function and linear distribution shifts. One of our key technical contributions is to extend this PPAD-hardness result to general convex domains, which is of broader interest in the complexity of variational inequalities. Finally, we address the special case of strategic classification, showing that computing a strategic local optimum is PLS-hard.
Authors: Zhipeng Zhang, Wenting Ma, Kai Li, Meng Guo, Lei Yang, Wei Yu, Hongji Cui, Yichen Zhang, Mo Zhang, Jinzhe Lin, Zhenjie Yao
Abstract: Robust reinforcement learning methods typically focus on suppressing unreliable experiences or corrupted rewards, but they lack the ability to reason about the reliability of their own learning process. As a result, such methods often either overreact to noise by becoming overly conservative or fail catastrophically when uncertainty accumulates. In this work, we propose a meta-cognitive reinforcement learning framework that enables an agent to assess, regulate, and recover its learning behavior based on internally estimated reliability signals. The proposed method introduces a meta-trust variable driven by Value Prediction Error Stability (VPES), which modulates learning dynamics via fail-safe regulation and gradual trust recovery. Experiments on continuous-control benchmarks with reward corruption demonstrate that recovery-enabled meta-cognitive control achieves higher average returns and significantly reduces late-stage training failures compared to strong robustness baselines.
Authors: Ratnavibusena Don Shahain Manujith, Yang Zhang, Teoh Tze Tzun, Kenji Kawaguchi
Abstract: Recent advances align diffusion models with human preferences to increase aesthetic appeal and mitigate artifacts and biases. Such methods aim to maximize a conditional output distribution aligned with higher rewards whilst not drifting far from a pretrained prior. This is commonly enforced by KL (Kullback Leibler) regularization. As such, a central issue still remains: how does one choose the right regularization strength? Too high of a strength leads to limited alignment and too low of a strength leads to "reward hacking". This renders the task of choosing the correct regularization strength highly non-trivial. Existing approaches sweep over this hyperparameter by aligning a pretrained model at multiple regularization strengths and then choose the best strength. Unfortunately, this is prohibitively expensive. We introduce DeRaDiff, a denoising time realignment procedure that, after aligning a pretrained model once, modulates the regularization strength during sampling to emulate models trained at other regularization strengths without any additional training or finetuning. Extending decoding-time realignment from language to diffusion models, DeRaDiff operates over iterative predictions of continuous latents by replacing the reverse step reference distribution by a geometric mixture of an aligned and reference posterior, thus giving rise to a closed form update under common schedulers and a single tunable parameter, lambda, for on the fly control. Our experiments show that across multiple text image alignment and image-quality metrics, our method consistently provides a strong approximation for models aligned entirely from scratch at different regularization strengths. Thus, our method yields an efficient way to search for the optimal strength, eliminating the need for expensive alignment sweeps and thereby substantially reducing computational costs.
Authors: Zhiyang Chen, Hailong Yao, Xia Yin
Abstract: Recent work has shown that machine-learned predictions can provably improve the performance of classic algorithms. In this work, we propose the first minimum-cost network flow algorithm augmented with a dual prediction. Our method is based on a classic minimum-cost flow algorithm, namely $\varepsilon$-relaxation. We provide time complexity bounds in terms of the infinity norm prediction error, which is both consistent and robust. We also prove sample complexity bounds for PAC-learning the prediction. We empirically validate our theoretical results on two applications of minimum-cost flow, i.e., traffic networks and chip escape routing, in which we learn a fixed prediction, and a feature-based neural network model to infer the prediction, respectively. Experimental results illustrate $12.74\times$ and $1.64\times$ average speedup on two applications.
Authors: Tianze Jiang, Blake Bordelon, Cengiz Pehlevan, Boris Hanin
Abstract: Mixture-of-Experts (MoE) layers have emerged as an important tool in scaling up modern neural networks by decoupling total trainable parameters from activated parameters in the forward pass for each token. However, sparse MoEs add complexity to training due to (i) new trainable parameters (router weights) that, like all other parameter groups, require hyperparameter (HP) tuning; (ii) new architecture scale dimensions (number of and size of experts) that must be chosen and potentially taken large. To make HP selection cheap and reliable, we propose a new parameterization for transformer models with MoE layers when scaling model width, depth, number of experts, and expert (hidden) size. Our parameterization is justified by a novel dynamical mean-field theory (DMFT) analysis. When varying different model dimensions trained at a fixed token budget, we find empirically that our parameterization enables reliable HP transfer across models from 51M to over 2B total parameters. We further take HPs identified from sweeping small models on a short token horizon to train larger models on longer horizons and report performant model behaviors.
Authors: Jinyang Wu, Shuo Yang, Changpeng Yang, Yuhao Shen, Shuai Zhang, Zhengqi Wen, Jianhua Tao
Abstract: Reinforcement learning has empowered large language models to act as intelligent agents, yet training them for long-horizon tasks remains challenging due to the scarcity of high-quality trajectories, especially under limited resources. Existing methods typically scale up rollout sizes and indiscriminately allocate computational resources among intermediate steps. Such attempts inherently waste substantial computation budget on trivial steps while failing to guarantee sample quality. To address this, we propose \textbf{Spark} (\textbf{S}trategic \textbf{P}olicy-\textbf{A}ware explo\textbf{R}ation via \textbf{K}ey-state dynamic branching), a novel framework that selectively branches at critical decision states for resource-efficient exploration. Our key insight is to activate adaptive branching exploration at critical decision points to probe promising trajectories, thereby achieving precise resource allocation that prioritizes sampling quality over blind coverage. This design leverages the agent's intrinsic decision-making signals to reduce dependence on human priors, enabling the agent to autonomously expand exploration and achieve stronger generalization. Experiments across diverse tasks (e.g., embodied planning), demonstrate that \textsc{Spark} achieves superior success rates with significantly fewer training samples, exhibiting robust generalization even in unseen scenarios.
Authors: Hadi Elzayn, Jacob Goldin
Abstract: We derive an accounting identity for predictive models that links accuracy with common fairness criteria. The identity shows that for globally calibrated models, the weighted sums of miscalibration within groups and error imbalance across groups is equal to a "total unfairness budget." For binary outcomes, this budget is the model's mean-squared error times the difference in group prevalence across outcome classes. The identity nests standard impossibility results as special cases, while also describing inherent tradeoffs when one or more fairness measures are not perfectly satisfied. The results suggest that accuracy and fairness are best viewed as complements in binary prediction tasks: increasing accuracy necessarily shrinks the total unfairness budget and vice-versa. Experiments on benchmark data confirm the theory and show that many fairness interventions largely substitute between fairness violations, and when they reduce accuracy they tend to expand the total unfairness budget. The results extend naturally to prediction tasks with non-binary outcomes, illustrating how additional outcome information can relax fairness incompatibilities and identifying conditions under which the binary-style impossibility does and does not extend to regression tasks.
Authors: Julian Gutierrez, Redouane Silvente
Abstract: We present two machine learning frameworks for forecasting aggregated curves and optimizing storage in the EPEX SPOT day-ahead market. First, a fast parametric model forecasts hourly demand and supply curves in a low-dimensional and grid-robust representation, with minimum and maximum volumes combined with a Chebyshev polynomial for the elastic segment. The model enables daily use with low error and clear interpretability. Second, for a more comprehensive analysis, though less suited to daily operation, we employ generative models that learn the joint distribution of 24-hour order-level submissions given weather and fuel variables. These models generate synthetic daily scenarios of individual buy and sell orders, which, once aggregated, yield hourly supply and demand curves. Based on these forecasts, we optimize a price-making storage strategy, quantify revenue distributions, and highlight the price-compression effect with lower peaks, higher off-peak levels, and diminishing returns as capacity expands.
Authors: Zichao Yu, Ming Li, Wenyi Zhang, Difan Zou, Weiguo Gao
Abstract: Inferring physical fields from sparse observations while strictly satisfying partial differential equations (PDEs) is a fundamental challenge in computational physics. Recently, deep generative models offer powerful data-driven priors for such inverse problems, yet existing methods struggle to enforce hard physical constraints without costly retraining or disrupting the learned generative prior. Consequently, there is a critical need for a sampling mechanism that can reconcile strict physical consistency and observational fidelity with the statistical structure of the pre-trained prior. To this end, we present ProFlow, a proximal guidance framework for zero-shot physics-consistent sampling, defined as inferring solutions from sparse observations using a fixed generative prior without task-specific retraining. The algorithm employs a rigorous two-step scheme that alternates between: (\romannumeral1) a terminal optimization step, which projects the flow prediction onto the intersection of the physically and observationally consistent sets via proximal minimization; and (\romannumeral2) an interpolation step, which maps the refined state back to the generative trajectory to maintain consistency with the learned flow probability path. This procedure admits a Bayesian interpretation as a sequence of local maximum a posteriori (MAP) updates. Comprehensive benchmarks on Poisson, Helmholtz, Darcy, and viscous Burgers' equations demonstrate that ProFlow achieves superior physical and observational consistency, as well as more accurate distributional statistics, compared to state-of-the-art diffusion- and flow-based baselines.
Authors: Ibne Farabi Shihab, Sanjeda Akter, Anuj Sharma
Abstract: We study black-box optimization of Lipschitz functions under noisy evaluations. Existing adaptive discretization methods implicitly avoid suboptimal regions but do not provide explicit certificates of optimality or measurable progress guarantees. We introduce \textbf{Certificate-Guided Pruning (CGP)}, which maintains an explicit \emph{active set} $A_t$ of potentially optimal points via confidence-adjusted Lipschitz envelopes. Any point outside $A_t$ is certifiably suboptimal with high probability, and under a margin condition with near-optimality dimension $\alpha$, we prove $\Vol(A_t)$ shrinks at a controlled rate yielding sample complexity $\tildeO(\varepsilon^{-(2+\alpha)})$. We develop three extensions: CGP-Adaptive learns $L$ online with $O(\log T)$ overhead; CGP-TR scales to $d > 50$ via trust regions with local certificates; and CGP-Hybrid switches to GP refinement when local smoothness is detected. Experiments on 12 benchmarks ($d \in [2, 100]$) show CGP variants match or exceed strong baselines while providing principled stopping criteria via certificate volume.
Authors: Hari Krishna Sahoo, Mudit Gaur, Vaneet Aggarwal
Abstract: Recently, flow-based generative models have shown superior efficiency compared to diffusion models. In this paper, we study rectified flow models, which constrain transport trajectories to be linear from the base distribution to the data distribution. This structural restriction greatly accelerates sampling, often enabling high-quality generation with a single Euler step. Under standard assumptions on the neural network classes used to parameterize the velocity field and data distribution, we prove that rectified flows achieve sample complexity $\tilde{O}(\varepsilon^{-2})$. This improves on the best known $O(\varepsilon^{-4})$ bounds for flow matching model and matches the optimal rate for mean estimation. Our analysis exploits the particular structure of rectified flows: because the model is trained with a squared loss along linear paths, the associated hypothesis class admits a sharply controlled localized Rademacher complexity. This yields the improved, order-optimal sample complexity and provides a theoretical explanation for the strong empirical performance of rectified flow models.
Authors: Yueyang Wang, Jiawei Fu, Baolong Bi, Xili Wang, Xiaoqing Liu
Abstract: SWE-bench has emerged as the premier benchmark for evaluating Large Language Models on complex software engineering tasks. While these capabilities are fundamentally acquired during the mid-training phase and subsequently elicited during Supervised Fine-Tuning (SFT), there remains a critical deficit in metrics capable of guiding mid-training effectively. Standard metrics such as Perplexity (PPL) are compromised by the "Long-Context Tax" and exhibit weak correlation with downstream SWE performance. In this paper, we bridge this gap by first introducing a rigorous data filtering strategy. Crucially, we propose the Entropy Compression Hypothesis, redefining intelligence not by scalar Top-1 compression, but by the capacity to structure uncertainty into Entropy-Compressed States of low orders ("reasonable hesitation"). Grounded in this fine-grained entropy analysis, we formulate a novel metric, HE-SNR (High-Entropy Signal-to-Noise Ratio). Validated on industrial-scale Mixture-of-Experts (MoE) models across varying context windows (32K/128K), our approach demonstrates superior robustness and predictive power. This work provides both the theoretical foundation and practical tools for optimizing the latent potential of LLMs in complex engineering domains.
Authors: Jinren Ding, Xuejian Xu, Shen Jiang, Zhitong Hao, Jinhui Yang, Peng Jiang
Abstract: Decision Transformer (DT) shows promise for generative auto-bidding by capturing temporal dependencies, but suffers from two critical limitations: insufficient cross-correlation modeling among state, action, and return-to-go (RTG) sequences, and indiscriminate learning of optimal/suboptimal behaviors. To address these, we propose C2, a novel framework enhancing DT with two core innovations: (1) a Cross Learning Block (CLB) via cross-attention to strengthen inter-sequence correlation modeling; (2) a Constraint-aware Loss (CL) incorporating budget and Cost-Per-Acquisition (CPA) constraints for selective learning of optimal trajectories. Extensive offline evaluations on the AuctionNet dataset demonstrate consistent performance gains (up to 3.23\% over state-of-the-art GAVE) across diverse budget settings; ablation studies verify the complementary synergy of CLB and CL, confirming C2's superiority in auto-bidding. The code for reproducing our results is available at: https://github.com/Dingjinren/C2.
Authors: Long Van Tran, Truyen Tran, Phuoc Nguyen
Abstract: Recent advances in stochastic differential equations (SDEs) have enabled robust modeling of real-world dynamical processes across diverse domains, such as finance, health, and systems biology. However, parameter estimation for SDEs typically relies on accurately timestamped observational sequences. When temporal ordering information is corrupted, missing, or deliberately hidden (e.g., for privacy), existing estimation methods often fail. In this paper, we investigate the conditions under which temporal order can be recovered and introduce a novel framework that simultaneously reconstructs temporal information and estimates SDE parameters. Our approach exploits asymmetries between forward and backward processes, deriving a score-matching criterion to infer the correct temporal order between pairs of observations. We then recover the total order via a sorting procedure and estimate SDE parameters from the reconstructed sequence using maximum likelihood. Finally, we conduct extensive experiments on synthetic and real-world datasets to demonstrate the effectiveness of our method, extending parameter estimation to settings with missing temporal order and broadening applicability in sensitive domains.
Authors: Daojun Liang, Qi Li, Yinglong Wang, Jing Chen, Hu Zhang, Xiaoxiao Cui, Qizheng Wang, Shuo Li
Abstract: Time series forecasting has long been dominated by advances in model architecture, with recent progress driven by deep learning and hybrid statistical techniques. However, as forecasting models approach diminishing returns in accuracy, a critical yet underexplored opportunity emerges: the strategic use of post-processing. In this paper, we address the last-mile gap in time-series forecasting, which is to improve accuracy and uncertainty without retraining or modifying a deployed backbone. We propose $\delta$-Adapter, a lightweight, architecture-agnostic way to boost deployed time series forecasters without retraining. $\delta$-Adapter learns tiny, bounded modules at two interfaces: input nudging (soft edits to covariates) and output residual correction. We provide local descent guarantees, $O(\delta)$ drift bounds, and compositional stability for combined adapters. Meanwhile, it can act as a feature selector by learning a sparse, horizon-aware mask over inputs to select important features, thereby improving interpretability. In addition, it can also be used as a distribution calibrator to measure uncertainty. Thus, we introduce a Quantile Calibrator and a Conformal Corrector that together deliver calibrated, personalized intervals with finite-sample coverage. Our experiments across diverse backbones and datasets show that $\delta$-Adapter improves accuracy and calibration with negligible compute and no interface changes.
Authors: Viet Hung Dinh, Ming Ding, Youyang Qu, Kanchana Thilakarathna
Abstract: While explainable artificial intelligence (XAI) for large language models (LLMs) remains an evolving field with many unresolved questions, increasing regulatory pressures have spurred interest in its role in ensuring transparency, accountability, and privacy-preserving machine unlearning. Despite recent advances in XAI have provided some insights, the specific role of attention layers in transformer based LLMs remains underexplored. This study investigates the memory mechanisms instantiated by attention layers, drawing on prior research in psychology and computational psycholinguistics that links Transformer attention to cue based retrieval in human memory. In this view, queries encode the retrieval context, keys index candidate memory traces, attention weights quantify cue trace similarity, and values carry the encoded content, jointly enabling the construction of a context representation that precedes and facilitates memory retrieval. Guided by the Encoding Specificity Principle, we hypothesize that the cues used in the initial stage of retrieval are instantiated as keywords. We provide converging evidence for this keywords-as-cues hypothesis. In addition, we isolate neurons within attention layers whose activations selectively encode and facilitate the retrieval of context-defining keywords. Consequently, these keywords can be extracted from identified neurons and further contribute to downstream applications such as unlearning.
Authors: Kaiyi Yang, Seonyeong Park, Gangwon Jeong, Hsuan-Kai Huang, Alexander A. Oraevsky, Umberto Villa, Mark A. Anastasio
Abstract: Photoacoustic computed tomography (PACT) is a promising imaging modality that combines the advantages of optical contrast with ultrasound detection. Utilizing ultrasound transducers with larger surface areas can improve detection sensitivity. However, when computationally efficient analytic reconstruction methods that neglect the spatial impulse responses (SIRs) of the transducer are employed, the spatial resolution of the reconstructed images will be compromised. Although optimization-based reconstruction methods can explicitly account for SIR effects, their computational cost is generally high, particularly in three-dimensional (3D) applications. To address the need for accurate but rapid 3D PACT image reconstruction, this study presents a framework for establishing a learned SIR compensation method that operates in the data domain. The learned compensation method maps SIR-corrupted PACT measurement data to compensated data that would have been recorded by idealized point-like transducers. Subsequently, the compensated data can be used with a computationally efficient reconstruction method that neglects SIR effects. Two variants of the learned compensation model are investigated that employ a U-Net model and a specifically designed, physics-inspired model, referred to as Deconv-Net. A fast and analytical training data generation procedure is also a component of the presented framework. The framework is rigorously validated in virtual imaging studies, demonstrating resolution improvement and robustness to noise variations, object complexity, and sound speed heterogeneity. When applied to in-vivo breast imaging data, the learned compensation models revealed fine structures that had been obscured by SIR-induced artifacts. To our knowledge, this is the first demonstration of learned SIR compensation in 3D PACT imaging.
Authors: Yuxuan Bao, Jan Zajac, Megan Powers, Venkat Raman, J. Nathan Kutz
Abstract: Bridging the sim2real gap between computationally inexpensive models and complex physical systems remains a central challenge in machine learning applications to engineering problems, particularly in multi-scale settings where reduced-order models typically capture only dominant dynamics. In this work, we present Cheap2Rich, a multi-scale data assimilation framework that reconstructs high-fidelity state spaces from sparse sensor histories by combining a fast low-fidelity prior with learned, interpretable discrepancy corrections. We demonstrate the performance on rotating detonation engines (RDEs), a challenging class of systems that couple detonation-front propagation with injector-driven unsteadiness, mixing, and stiff chemistry across disparate scales. Our approach successfully reconstructs high-fidelity RDE states from sparse measurements while isolating physically meaningful discrepancy dynamics associated with injector-driven effects. The results highlight a general multi-fidelity framework for data assimilation and system identification in complex multi-scale systems, enabling rapid design exploration and real-time monitoring and control while providing interpretable discrepancy dynamics. Code for this project is is available at: github.com/kro0l1k/Cheap2Rich.
Authors: Tianyi Alex Qiu, Micah Carroll, Cameron Allen
Abstract: The evaluation and post-training of large language models (LLMs) rely on supervision, but strong supervision for difficult tasks is often unavailable, especially when evaluating frontier models. In such cases, models are demonstrated to exploit evaluations built on such imperfect supervision, leading to deceptive results. However, underutilized in LLM research, a wealth of mechanism design research focuses on game-theoretic incentive compatibility, i.e., eliciting honest and informative answers with weak supervision. Drawing from this literature, we introduce the peer prediction method for model evaluation and post-training. It rewards honest and informative answers over deceptive and uninformative ones, using a metric based on mutual predictability and without requiring ground truth labels. We demonstrate the method's effectiveness and resistance to deception, with both theoretical guarantees and empirical validation on models with up to 405B parameters. We show that training an 8B model with peer prediction-based reward recovers most of the drop in truthfulness due to prior malicious finetuning, even when the reward is produced by a 0.135B language model with no finetuning. On the evaluation front, in contrast to LLM-as-a-Judge which requires strong and trusted judges, we discover an inverse scaling property in peer prediction, where, surprisingly, resistance to deception is strengthened as the capability gap between the experts and participants widens, enabling reliable evaluation of strong models with weak supervision. In particular, LLM-as-a-Judge become worse than random guess when facing deceptive models 5-20x the judge's size, while peer prediction thrives when such gaps are large, including in cases with over 100x size difference.
Authors: Xinyu Li, Sishuo Chen, Guipeng Xv, Li Zhang, Mingxuan Luo, Zhangming Chan, Xiang-Rong Sheng, Han Zhu, Jian Xu, Chen Lin
Abstract: The prediction objectives of online advertisement ranking models are evolving from probabilistic metrics like conversion rate (CVR) to numerical business metrics like post-click gross merchandise volume (GMV). Unlike the well-studied delayed feedback problem in CVR prediction, delayed feedback modeling for GMV prediction remains unexplored and poses greater challenges, as GMV is a continuous target, and a single click can lead to multiple purchases that cumulatively form the label. To bridge the research gap, we establish TRACE, a GMV prediction benchmark containing complete transaction sequences rising from each user click, which supports delayed feedback modeling in an online streaming manner. Our analysis and exploratory experiments on TRACE reveal two key insights: (1) the rapid evolution of the GMV label distribution necessitates modeling delayed feedback under online streaming training; (2) the label distribution of repurchase samples substantially differs from that of single-purchase samples, highlighting the need for separate modeling. Motivated by these findings, we propose RepurchasE-Aware Dual-branch prEdictoR (READER), a novel GMV modeling paradigm that selectively activates expert parameters according to repurchase predictions produced by a router. Moreover, READER dynamically calibrates the regression target to mitigate under-estimation caused by incomplete labels. Experimental results show that READER yields superior performance on TRACE over baselines, achieving a 2.19% improvement in terms of accuracy. We believe that our study will open up a new avenue for studying online delayed feedback modeling for GMV prediction, and our TRACE benchmark with the gathered insights will facilitate future research and application in this promising direction. Our code and dataset are available at https://github.com/alimama-tech/OnlineGMV .
Authors: Fengrui Zuo, Zhiwei Ke, Yiming Liu, Wenqi Lou, Chao Wang, Xvehai Zhou
Abstract: Diffusion language models (DLMs) generate text through iterative denoising, but inference requires full-sequence attention at every iteration, resulting in substantial redundant computation on masked tokens. Block-wise diffusion can reduce this cost, yet it typically relies on retraining and constrained update orders, limiting its direct applicability to pretrained DLMs. Our token-level analysis reveals pronounced structural locality in DLM inference. Decoding is driven by a small set of prefix-localized active tokens; the influence of distant undecoded context diminishes rapidly, and decoded tokens exhibit stage-wise temporal stability, enabling reuse of intermediate representations except for a brief post-decode transient. Motivated by these observations, we propose \textbf{\placeholder}\footnote{The source code is available at https://github.com/vhicrgit/Window-Diffusion.}, a window-based token pruning and caching method for inference. We maintain a local computation window that slides rightward as denoising progresses, and partition undecoded tokens into: (i) \textit{active tokens} that are computed online, (ii) \textit{buffer tokens} whose KV states are cached and periodically refreshed, and (iii) \textit{far-field tokens} that are pruned outside the window. Computation is restricted to active and buffer tokens within the window, while far-field tokens are omitted at each stage. Experiments on LLaDA and Dream show that, under matched compute budgets, our method achieves up to $99\times$ inference speedup while largely preserving generation performance.
Authors: Minjae Lee, Wonjun Kang, Byeongkeun Ahn, Christian Classen, Kevin Galim, Seunghyuk Oh, Minghao Yan, Hyung Il Koo, Kangwook Lee
Abstract: Speculative decoding (SD) has proven effective for accelerating LLM inference by quickly generating draft tokens and verifying them in parallel. However, SD remains largely unexplored for Large Vision-Language Models (LVLMs), which extend LLMs to process both image and text prompts. To address this gap, we benchmark existing inference methods with small draft models on 11 datasets across diverse input scenarios and observe scenario-specific performance fluctuations. Motivated by these findings, we propose Test-time Adaptive Batched Ensemble Drafting (TABED), which dynamically ensembles multiple drafts obtained via batch inference by leveraging deviations from past ground truths available in the SD setting. The dynamic ensemble method achieves an average robust walltime speedup of 1.74x over autoregressive decoding and a 5% improvement over single drafting methods, while remaining training-free and keeping ensembling costs negligible through parameter sharing. With its plug-and-play compatibility, we further enhance TABED by integrating advanced verification and alternative drafting methods. Code and custom-trained models are available at https://github.com/furiosa-ai/TABED.
Authors: Chen-Yang Dai, Che-Chia Chang, Te-Sheng Lin, Ming-Chih Lai, Chieh-Hsin Lai
Abstract: Physics-informed neural networks (PINNs) solve time-dependent partial differential equations (PDEs) by learning a mesh-free, differentiable solution that can be evaluated anywhere in space and time. However, standard space--time PINNs take time as an input but reuse a single network with shared weights across all times, forcing the same features to represent markedly different dynamics. This coupling degrades accuracy and can destabilize training when enforcing PDE, boundary, and initial constraints jointly. We propose Time-Induced Neural Networks (TINNs), a novel architecture that parameterizes the network weights as a learned function of time, allowing the effective spatial representation to evolve over time while maintaining shared structure. The resulting formulation naturally yields a nonlinear least-squares problem, which we optimize efficiently using a Levenberg--Marquardt method. Experiments on various time-dependent PDEs show up to $4\times$ improved accuracy and $10\times$ faster convergence compared to PINNs and strong baselines.
Authors: Mariia Drozdova
Abstract: Can standard continuous-time generative models represent distributions whose support is an extremely sparse, globally constrained discrete set? We study this question using completed Sudoku grids as a controlled testbed, treating them as a subset of a continuous relaxation space. We train flow-matching and score-based models along a Gaussian probability path and compare deterministic (ODE) sampling, stochastic (SDE) sampling, and DDPM-style discretizations derived from the same continuous-time training. Unconditionally, stochastic sampling substantially outperforms deterministic flows; score-based samplers are the most reliable among continuous-time methods, and DDPM-style ancestral sampling achieves the highest validity overall. We further show that the same models can be repurposed for guided generation: by repeatedly sampling completions under clamped clues and stopping when constraints are satisfied, the model acts as a probabilistic Sudoku solver. Although far less sample-efficient than classical solvers and discrete-geometry-aware diffusion methods, these experiments demonstrate that classic diffusion/flow formulations can assign non-zero probability mass to globally constrained combinatorial structures and can be used for constraint satisfaction via stochastic search.
Authors: Qing Lyu, Zhe Fu, Alexandre Bayen
Abstract: Identifying safety-critical scenarios is essential for autonomous driving, but the rarity of such events makes supervised labeling impractical. Traditional rule-based metrics like Time-to-Collision are too simplistic to capture complex interaction risks, and existing methods lack a systematic way to verify whether statistical anomalies truly reflect physical danger. To address this gap, we propose an unsupervised anomaly detection framework based on a multi-agent Transformer that models normal driving and measures deviations through prediction residuals. A dual evaluation scheme has been proposed to assess both detection stability and physical alignment: Stability is measured using standard ranking metrics in which Kendall Rank Correlation Coefficient captures rank agreement and Jaccard index captures the consistency of the top-K selected items; Physical alignment is assessed through correlations with established Surrogate Safety Measures (SSM). Experiments on the NGSIM dataset demonstrate our framework's effectiveness: We show that the maximum residual aggregator achieves the highest physical alignment while maintaining stability. Furthermore, our framework identifies 388 unique anomalies missed by Time-to-Collision and statistical baselines, capturing subtle multi-agent risks like reactive braking under lateral drift. The detected anomalies are further clustered into four interpretable risk types, offering actionable insights for simulation and testing.
Authors: Wei Huang, Anda Cheng, Yinggui Wang, Lei Wang, Tao Wei
Abstract: Large Language Models (LLMs) can be fine-tuned on domain-specific data to enhance their performance in specialized fields. However, such data often contains numerous low-quality samples, necessitating effective data processing (DP). In practice, DP strategies are typically developed through iterative manual analysis and trial-and-error adjustment. These processes inevitably incur high labor costs and may lead to privacy issues in high-privacy domains like healthcare due to direct human access to sensitive data. Thus, achieving automated data processing without exposing the raw data has become a critical challenge. To address this challenge, we propose LLM-AutoDP, a novel framework that leverages LLMs as agents to automatically generate and optimize data processing strategies. Our method generates multiple candidate strategies and iteratively refines them using feedback signals and comparative evaluations. This iterative in-context learning mechanism enables the agent to converge toward high-quality processing pipelines without requiring direct human intervention or access to the underlying data. To further accelerate strategy search, we introduce three key techniques: Distribution Preserving Sampling, which reduces data volume while maintaining distributional integrity; Processing Target Selection, which uses a binary classifier to identify low-quality samples for focused processing; Cache-and-Reuse Mechanism}, which minimizes redundant computations by reusing prior processing results. Results show that models trained on data processed by our framework achieve over 80% win rates against models trained on unprocessed data. Compared to AutoML baselines based on LLM agents, LLM-AutoDP achieves approximately a 65% win rate. Moreover, our acceleration techniques reduce the total searching time by up to 10 times, demonstrating both effectiveness and efficiency.
Authors: Kaile Wang, Jiannong Cao, Yu Yang, Xiaoyin Li, Mingjin Zhang
Abstract: Heterogeneous federated learning (HFL) aims to ensure effective and privacy-preserving collaboration among different entities. As newly joined clients require significant adjustments and additional training to align with the existing system, the problem of generalizing federated learning models to unseen clients under heterogeneous data has become progressively crucial. Consequently, we highlight two unsolved challenging issues in federated domain generalization: Optimization Divergence and Performance Divergence. To tackle the above challenges, we propose FedRD, a novel heterogeneity-aware federated learning algorithm that collaboratively utilizes parameter-guided global generalization aggregation and local debiased classification to reduce divergences, aiming to obtain an optimal global model for participating and unseen clients. Extensive experiments on public multi-domain datasets demonstrate that our approach exhibits a substantial performance advantage over competing baselines in addressing this specific problem.
Authors: Wei Li
Abstract: Time series forecasting presents significant challenges due to the complex temporal dependencies at multiple time scales. This paper introduces ScatterFusion, a novel framework that synergistically integrates scattering transforms with hierarchical attention mechanisms for robust time series forecasting. Our approach comprises four key components: (1) a Hierarchical Scattering Transform Module (HSTM) that extracts multi-scale invariant features capturing both local and global patterns; (2) a Scale-Adaptive Feature Enhancement (SAFE) module that dynamically adjusts feature importance across different scales; (3) a Multi-Resolution Temporal Attention (MRTA) mechanism that learns dependencies at varying time horizons; and (4) a Trend-Seasonal-Residual (TSR) decomposition-guided structure-aware loss function. Extensive experiments on seven benchmark datasets demonstrate that ScatterFusion outperforms other common methods, achieving significant reductions in error metrics across various prediction horizons.
Authors: Wei Li
Abstract: Time series forecasting requires capturing patterns across multiple temporal scales while maintaining computational efficiency. This paper introduces AWGformer, a novel architecture that integrates adaptive wavelet decomposition with cross-scale attention mechanisms for enhanced multi-variate time series prediction. Our approach comprises: (1) an Adaptive Wavelet Decomposition Module (AWDM) that dynamically selects optimal wavelet bases and decomposition levels based on signal characteristics; (2) a Cross-Scale Feature Fusion (CSFF) mechanism that captures interactions between different frequency bands through learnable coupling matrices; (3) a Frequency-Aware Multi-Head Attention (FAMA) module that weights attention heads according to their frequency selectivity; (4) a Hierarchical Prediction Network (HPN) that generates forecasts at multiple resolutions before reconstruction. Extensive experiments on benchmark datasets demonstrate that AWGformer achieves significant average improvements over state-of-the-art methods, with particular effectiveness on multi-scale and non-stationary time series. Theoretical analysis provides convergence guarantees and establishes the connection between our wavelet-guided attention and classical signal processing principles.
Authors: Yuhang Liu, Erdun Gao, Dong Gong, Anton van den Hengel, Javen Qinfeng Shi
Abstract: Developing human understandable interpretation of large language models (LLMs) becomes increasingly critical for their deployment in essential domains. Mechanistic interpretability seeks to mitigate the issues through extracts human-interpretable process and concepts from LLMs' activations. Sparse autoencoders (SAEs) have emerged as a popular approach for extracting interpretable and monosemantic concepts by decomposing the LLM internal representations into a dictionary. Despite their empirical progress, SAEs suffer from a fundamental theoretical ambiguity: the well-defined correspondence between LLM representations and human-interpretable concepts remains unclear. This lack of theoretical grounding gives rise to several methodological challenges, including difficulties in principled method design and evaluation criteria. In this work, we show that, under mild assumptions, LLM representations can be approximated as a {linear mixture} of the log-posteriors over concepts given the input context, through the lens of a latent variable model where concepts are treated as latent variables. This motivates a principled framework for concept extraction, namely Concept Component Analysis (ConCA), which aims to recover the log-posterior of each concept from LLM representations through a {unsupervised} linear unmixing process. We explore a specific variant, termed sparse ConCA, which leverages a sparsity prior to address the inherent ill-posedness of the unmixing problem. We implement 12 sparse ConCA variants and demonstrate their ability to extract meaningful concepts across multiple LLMs, offering theory-backed advantages over SAEs.
Authors: S\"onke Beier, Paula Pirker-D\'iaz, Friedrich Pagenkopf, Karoline Wiesner
Abstract: Diffusion Map is a spectral dimensionality reduction technique which is able to uncover nonlinear submanifolds in high-dimensional data. And, it is increasingly applied across a wide range of scientific disciplines, such as biology, engineering, and social sciences. But data preprocessing, parameter settings and component selection have a significant influence on the resulting manifold, something which has not been comprehensively discussed in the literature so far. We provide a practice oriented review of the Diffusion Map technique, illustrate pitfalls and showcase a recently introduced technique for identifying the most relevant components. Our results show that the first components are not necessarily the most relevant ones.
Authors: Zhiyu Chen, Minhao Liu, Yanru Zhang
Abstract: Recent lightweight MLP-based models have achieved strong performance in time series forecasting by capturing stable trends and seasonal patterns. However, their effectiveness hinges on an implicit assumption of local stationarity assumption, making them prone to errors in long-term forecasting of highly non-stationary series, especially when abrupt fluctuations occur, a common challenge in domains like web traffic monitoring. To overcome this limitation, we propose TimeCatcher, a novel Volatility-Aware Variational Forecasting framework. TimeCatcher extends linear architectures with a variational encoder to capture latent dynamic patterns hidden in historical data and a volatility-aware enhancement mechanism to detect and amplify significant local variations. Experiments on nine real-world datasets from traffic, financial, energy, and weather domains show that TimeCatcher consistently outperforms state-of-the-art baselines, with particularly large improvements in long-term forecasting scenarios characterized by high volatility and sudden fluctuations. Our code is available at https://github.com/ColaPrinceCHEN/TimeCatcher.
Authors: Fatima Ezzeddine, Obaida Ammar, Silvia Giordano, Omran Ayoub
Abstract: Explainable Artificial Intelligence (XAI) is becoming increasingly essential for enhancing the transparency of machine learning (ML) models. Among the various XAI techniques, counterfactual explanations (CFs) hold a pivotal role due to their ability to illustrate how changes in input features can alter an ML model's decision, thereby offering actionable recourse to users. Ensuring that individuals with comparable attributes and those belonging to different protected groups (e.g., demographic) receive similar and actionable recourse options is essential for trustworthy and fair decision-making. In this work, we address this challenge directly by focusing on the generation of fair CFs. Specifically, we start by defining and formulating fairness at: 1) individual fairness, ensuring that similar individuals receive similar CFs, 2) group fairness, ensuring equitable CFs across different protected groups and 3) hybrid fairness, which accounts for both individual and broader group-level fairness. We formulate the problem as an optimization task and propose a novel model-agnostic, reinforcement learning based approach to generate CFs that satisfy fairness constraints at both the individual and group levels, two objectives that are usually treated as orthogonal. As fairness metrics, we extend existing metrics commonly used for auditing ML models, such as equal choice of recourse and equal effectiveness across individuals and groups. We evaluate our approach on three benchmark datasets, showing that it effectively ensures individual and group fairness while preserving the quality of the generated CFs in terms of proximity and plausibility, and quantify the cost of fairness in the different levels separately. Our work opens a broader discussion on hybrid fairness and its role and implications for XAI and beyond CFs.
Authors: Kadircan Aksoy, Peter Jung, Protim Bhattacharjee
Abstract: We study the supervised training dynamics of neural classifiers through the lens of binary hypothesis testing. We model classification as a set of binary tests between class-conditional distributions of representations and empirically show that, along training trajectories, well-generalizing networks increasingly align with Neyman-Pearson optimal decision rules via monotonic improvements in KL divergence that relate to error rate exponents. We finally discuss how this yields an explanation and possible training or regularization strategies for different classes of neural networks.
Authors: C. V\'azquez-Garc\'ia, F. J. Mart\'inez-Murcia, F. Segovia Rom\'an, A. Forte, J. Ram\'irez, I. Ill\'an, A. Hern\'andez-Segura, C. Jim\'enez-Mesa, Juan M. G\'orriz
Abstract: High-dimensional neuroimaging data presents challenges for assessing neurodegenerative diseases due to complex non-linear relationships. Variational Autoencoders (VAEs) can encode scans into lower-dimensional latent spaces capturing disease-relevant features. We propose a semi-supervised VAE framework with a flexible similarity regularization term that aligns selected latent variables with clinical or biomarker measures of dementia progression. This allows adapting the similarity metric and supervised variables to specific goals or available data. We demonstrate the approach using PET scans from the Alzheimer's Disease Neuroimaging Initiative (ADNI), guiding the first latent dimension to align with a cognitive score. Using this supervised latent variable, we generate average reconstructions across levels of cognitive impairment. Voxel-wise GLM analysis reveals reduced metabolism in key regions, mainly the hippocampus, and within major Resting State Networks, particularly the Default Mode and Central Executive Networks. The remaining latent variables encode affine transformations and intensity variations, capturing confounds such as inter-subject variability and site effects. Our framework effectively extracts disease-related patterns aligned with established Alzheimer's biomarkers, offering an interpretable and adaptable tool for studying neurodegenerative progression.
Authors: Jiawen Chen, Qi Shao, Mingtong Zhou, Duxin Chen, Wenwu Yu
Abstract: Topological deep learning has emerged for modeling higher-order relational structures beyond pairwise interactions that standard graph neural networks fail to capture. Although combinatorial complexes offer a unified topological framework, most existing topological deep learning methods rely on local message passing via attention mechanisms, which incur quadratic complexity and remain low-dimensional, limiting scalability and rank-aware information aggregation in higher-order complexes.We propose Combinatorial Complex Mamba (CCMamba), the first unified mamba-based neural framework for learning on combinatorial complexes. CCMamba reformulates message passing as a selective state-space modeling problem by organizing multi-rank incidence relations into structured sequences processed by rank-aware state-space models. This enables adaptive, directional, and long range information propagation in linear time without self attention. We further establish the theoretical analysis that the expressive power upper-bound of CCMamba message passing is the 1-Weisfeiler-Lehman test. Experiments on graph, hypergraph, and simplicial benchmarks demonstrate that CCMamba consistently outperforms existing methods while exhibiting improved scalability and robustness to depth.
Authors: Ariel Maymon, Yanir Buznah, Uri Shaham
Abstract: Unsupervised ensemble learning emerged to address the challenge of combining multiple learners' predictions without access to ground truth labels or additional data. This paradigm is crucial in scenarios where evaluating individual classifier performance or understanding their strengths is challenging due to limited information. We propose a novel deep energy-based method for constructing an accurate meta-learner using only the predictions of individual learners, potentially capable of capturing complex dependence structures between them. Our approach requires no labeled data, learner features, or problem-specific information, and has theoretical guarantees for when learners are conditionally independent. We demonstrate superior performance across diverse ensemble scenarios, including challenging mixture of experts settings. Our experiments span standard ensemble datasets and curated datasets designed to test how the model fuses expertise from multiple sources. These results highlight the potential of unsupervised ensemble learning to harness collective intelligence, especially in data-scarce or privacy-sensitive environments.
Authors: Efstratios Zaradoukas, Bardh Prenkaj, Gjergji Kasneci
Abstract: During pretraining, LLMs inadvertently memorize sensitive or copyrighted data, posing significant compliance challenges under legal frameworks like the GDPR and the EU AI Act. Fulfilling these mandates demands techniques that can remove information from a deployed model without retraining from scratch. Existing unlearning approaches attempt to address this need, but often leak the very data they aim to erase, sacrifice fluency and robustness, or depend on costly external reward models. We introduce PURGE (Policy Unlearning through Relative Group Erasure), a novel method grounded in the Group Relative Policy Optimization framework that formulates unlearning as a verifiable problem. PURGE uses an intrinsic reward signal that penalizes any mention of forbidden concepts, allowing safe and consistent unlearning. Our approach reduces token usage per target by up to a factor of 46 compared with SotA methods, while improving fluency by 5.48 percent and adversarial robustness by 12.02 percent over the base model. On the Real World Knowledge Unlearning (RWKU) benchmark, PURGE achieves 11 percent unlearning effectiveness while preserving 98 percent of original utility. PURGE shows that framing LLM unlearning as a verifiable task, enables more reliable, efficient, and scalable forgetting, suggesting a promising new direction for unlearning research that combines theoretical guarantees, improved safety, and practical deployment efficiency.
Authors: Anna van Elst, Igor Colin, Stephan Cl\'emen\c{c}on
Abstract: Specifications for decentralized learning on resource-constrained edge devices require algorithms that are communication-efficient, robust to data corruption, and lightweight in memory usage. While state-of-the-art gossip-based methods satisfy the first requirement, achieving robustness remains challenging. Asynchronous decentralized ADMM-based methods have been explored for estimating the median, a statistical centrality measure that is notoriously more robust than the mean. However, existing approaches require memory that scales with node degree, making them impractical when memory is limited. In this paper, we propose AsylADMM, a novel gossip algorithm for decentralized median and quantile estimation, primarily designed for asynchronous updates and requiring only two variables per node. We analyze a synchronous variant of AsylADMM to establish theoretical guarantees and empirically demonstrate fast convergence for the asynchronous algorithm. We then show that our algorithm enables quantile-based trimming, geometric median estimation, and depth-based trimming, with quantile-based trimming empirically outperforming existing rank-based methods. Finally, we provide a novel theoretical analysis of rank-based trimming via Markov chain theory.
Authors: Aiming Hao, Chen Zhu, Jiashu Zhu, Jiahong Wu, Xiangxiang Chu
Abstract: Ordinal regression and ranking are challenging due to inherent ordinal dependencies that conventional methods struggle to model. We propose Ranking-Aware Reinforcement Learning (RARL), a novel RL framework that explicitly learns these relationships. At its core, RARL features a unified objective that synergistically integrates regression and Learning-to-Rank (L2R), enabling mutual improvement between the two tasks. This is driven by a ranking-aware verifiable reward that jointly assesses regression precision and ranking accuracy, facilitating direct model updates via policy optimization. To further enhance training, we introduce Response Mutation Operations (RMO), which inject controlled noise to improve exploration and prevent stagnation at saddle points. The effectiveness of RARL is validated through extensive experiments on three distinct benchmarks.
Authors: Hyunjun Na, Donghwan Lee
Abstract: Gradient temporal-difference (GTD) learning algorithms are widely used for off-policy policy evaluation with function approximation. However, existing convergence analyses rely on the restrictive assumption that the so-called feature interaction matrix (FIM) is nonsingular. In practice, the FIM can become singular and leads to instability or degraded performance. In this paper, we propose a regularized optimization objective by reformulating the mean-square projected Bellman error (MSPBE) minimization. This formulation naturally yields a regularized GTD algorithms, referred to as R-GTD, which guarantees convergence to a unique solution even when the FIM is singular. We establish theoretical convergence guarantees and explicit error bounds for the proposed method, and validate its effectiveness through empirical experiments.
Authors: Junaid Sajid, Ivo M\"u\"ursepp, Luca Reggiani, Davide Scazzoli, Federico Francesco Luigi Mariani, Maurizio Magarini, Rizwan Ahmad, Muhammad Mahtab Alam
Abstract: Uncrewed Aerial Vehicles (UAVs) are increasingly used in civilian and industrial applications, making secure low-altitude operations crucial. In dense mmWave environments, accurately classifying low-altitude UAVs as either inside authorized or restricted airspaces remains challenging, requiring models that handle complex propagation and signal variability. This paper proposes a deep learning model, referred to as CoBA, which stands for integrated Convolutional Neural Network (CNN), Bidirectional Long Short-Term Memory (BiLSTM), and Attention which leverages Fifth Generation (5G) millimeter-wave (mmWave) radio measurements to classify UAV operations in authorized and restricted airspaces at low altitude. The proposed CoBA model integrates convolutional, bidirectional recurrent, and attention layers to capture both spatial and temporal patterns in UAV radio measurements. To validate the model, a dedicated dataset is collected using the 5G mmWave network at TalTech, with controlled low altitude UAV flights in authorized and restricted scenarios. The model is evaluated against conventional ML models and a fingerprinting-based benchmark. Experimental results show that CoBA achieves superior accuracy, significantly outperforming all baseline models and demonstrating its potential for reliable and regulated UAV airspace monitoring.
Authors: Xinyu Wang, Ruoyu Wang, Qiangwei Peng, Peijie Zhou, Tiejun Li
Abstract: Reconstructing dynamical evolution from limited observations is a fundamental challenge in single-cell biology, where dynamic unbalanced optimal transport provides a principled framework for modeling coupled transport and mass variation. However, existing approaches rely on trajectory simulation at inference time, making inference a key bottleneck for scalable applications. In this work, we propose a mean-flow framework for unbalanced flow matching that summarizes both transport and mass-growth dynamics over arbitrary time intervals using mean velocity and mass-growth fields, enabling fast one-step generation without trajectory simulation. To solve dynamic unbalanced optimal transport under the Wasserstein-Fisher-Rao geometry, we further build on this framework to develop Wasserstein-Fisher-Rao Mean Flow Matching (WFR-MFM). Across synthetic and real single-cell RNA sequencing datasets, WFR-MFM achieves orders-of-magnitude faster inference than a range of existing baselines while maintaining high predictive accuracy, and enables efficient perturbation response prediction on large synthetic datasets with thousands of conditions.
Authors: Gawon Lee, Hanbyeol Park, Minseop Kim, Dohee Kim, Hyerim Bae
Abstract: Time series forecasting (TSF) faces challenges in modeling complex intra-channel temporal dependencies and inter-channel correlations. Although recent research has highlighted the efficiency of linear architectures in capturing global trends, these models often struggle with non-linear signals. To address this gap, we conducted a systematic receptive field analysis of convolutional neural network (CNN) TSF models. We introduce the "individual receptive field" to uncover granular structural dependencies, revealing that convolutional layers act as feature extractors that mirror channel-wise attention while exhibiting superior robustness to non-linear fluctuations. Based on these insights, we propose ACFormer, an architecture designed to reconcile the efficiency of linear projections with the non-linear feature-extraction power of convolutions. ACFormer captures fine-grained information through a shared compression module, preserves temporal locality via gated attention, and reconstructs variable-specific temporal patterns using an independent patch expansion layer. Extensive experiments on multiple benchmark datasets demonstrate that ACFormer consistently achieves state-of-the-art performance, effectively mitigating the inherent drawbacks of linear models in capturing high-frequency components.
Authors: Gilles Eerlings, Brent Zoomers, Jori Liesenborgs, Gustavo Rovelo Ruiz, Kris Luyten
Abstract: We propose DIVERSE, a framework for systematically exploring the Rashomon set of deep neural networks, the collection of models that match a reference model's accuracy while differing in their predictive behavior. DIVERSE augments a pretrained model with Feature-wise Linear Modulation (FiLM) layers and uses Covariance Matrix Adaptation Evolution Strategy (CMA-ES) to search a latent modulation space, generating diverse model variants without retraining or gradient access. Across MNIST, PneumoniaMNIST, and CIFAR-10, DIVERSE uncovers multiple high-performing yet functionally distinct models. Our experiments show that DIVERSE offers a competitive and efficient exploration of the Rashomon set, making it feasible to construct diverse sets that maintain robustness and performance while supporting well-balanced model multiplicity. While retraining remains the baseline to generate Rashomon sets, DIVERSE achieves comparable diversity at reduced computational cost.
Authors: Leon G\"otz, Lars Frederik Peiss, Erik Sauer, Andreas Udo Sass, Thorsten Bagdonat, Stephan G\"unnemann, Leo Schwinn
Abstract: Virtual sensors use machine learning to predict target signals from available measurements, replacing expensive physical sensors in critical applications. Existing virtual sensor approaches require application-specific models with hand-selected inputs for each sensor, cannot leverage task synergies, and lack consistent benchmarks. At the same time, emerging time series foundation models are computationally expensive and limited to predicting their input signals, making them incompatible with virtual sensors. We introduce the first foundation model for virtual sensors addressing both limitations. Our unified model can simultaneously predict diverse virtual sensors exploiting synergies while maintaining computational efficiency. It learns relevant input signals for each virtual sensor, eliminating expert knowledge requirements while adding explainability. In our large-scale evaluation on a standard benchmark and an application-specific dataset with over 18 billion samples, our architecture achieves 415x reduction in computation time and 951x reduction in memory requirements, while maintaining or even improving predictive quality compared to baselines. Our model scales gracefully to hundreds of virtual sensors with nearly constant parameter count, enabling practical deployment in large-scale sensor networks.
Authors: Panayiotis Ioannou, Pietro Li\`o, Pietro Cicuta
Abstract: Accurately modelling the dynamics of complex systems and discovering their governing differential equations are critical tasks for accelerating scientific discovery. Using noisy, synthetic data from two damped oscillatory systems, we explore the extrapolation capabilities of Neural Ordinary Differential Equations (NODEs) and the ability of Symbolic Regression (SR) to recover the underlying equations. Our study yields three key insights. First, we demonstrate that NODEs can extrapolate effectively to new boundary conditions, provided the resulting trajectories share dynamic similarity with the training data. Second, SR successfully recovers the equations from noisy ground-truth data, though its performance is contingent on the correct selection of input variables. Finally, we find that SR recovers two out of the three governing equations, along with a good approximation for the third, when using data generated by a NODE trained on just 10% of the full simulation. While this last finding highlights an area for future work, our results suggest that using NODEs to enrich limited data and enable symbolic regression to infer physical laws represents a promising new approach for scientific discovery.
Authors: Rohan Asthana, Vasileios Belagiannis
Abstract: Diffusion-based image generative models produce high-fidelity images through iterative denoising but remain vulnerable to memorization, where they unintentionally reproduce exact copies or parts of training images. Recent memorization detection methods are primarily based on the norm of score difference as indicators of memorization. We prove that such norm-based metrics are mainly effective under the assumption of isotropic log-probability distributions, which generally holds at high or medium noise levels. In contrast, analyzing the anisotropic regime reveals that memorized samples exhibit strong angular alignment between the guidance vector and unconditional scores in the low-noise setting. Through these insights, we develop a memorization detection metric by integrating isotropic norm and anisotropic alignment. Our detection metric can be computed directly on pure noise inputs via two conditional and unconditional forward passes, eliminating the need for costly denoising steps. Detection experiments on Stable Diffusion v1.4 and v2 show that our metric outperforms existing denoising-free detection methods while being at least approximately 5x faster than the previous best approach. Finally, we demonstrate the effectiveness of our approach by utilizing a mitigation strategy that adapts memorized prompts based on our developed metric.
Authors: Alejandro Luque-Cerpa, Mengyuan Wang, Emil Carlsson, Sanjit A. Seshia, Devdatt Dubhashi, Hazem Torfah
Abstract: We introduce a novel framework for learning context-aware runtime monitors for AI-based control ensembles. Machine-learning (ML) controllers are increasingly deployed in (autonomous) cyber-physical systems because of their ability to solve complex decision-making tasks. However, their accuracy can degrade sharply in unfamiliar environments, creating significant safety concerns. Traditional ensemble methods aim to improve robustness by averaging or voting across multiple controllers, yet this often dilutes the specialized strengths that individual controllers exhibit in different operating contexts. We argue that, rather than blending controller outputs, a monitoring framework should identify and exploit these contextual strengths. In this paper, we reformulate the design of safe AI-based control ensembles as a contextual monitoring problem. A monitor continuously observes the system's context and selects the controller best suited to the current conditions. To achieve this, we cast monitor learning as a contextual learning task and draw on techniques from contextual multi-armed bandits. Our approach comes with two key benefits: (1) theoretical safety guarantees during controller selection, and (2) improved utilization of controller diversity. We validate our framework in two simulated autonomous driving scenarios, demonstrating significant improvements in both safety and performance compared to non-contextual baselines.
Authors: Stefano Bertolasi, Diego Carrera, Diego Stucchi, Pasqualina Fragneto, Luigi Amedeo Bianchi
Abstract: Change Point Detection (CPD) is a critical task in time series analysis, aiming to identify moments when the underlying data-generating process shifts. Traditional CPD methods often rely on unsupervised techniques, which lack adaptability to task-specific definitions of change and cannot benefit from user knowledge. To address these limitations, we propose MuRAL-CPD, a novel semi-supervised method that integrates active learning into a multiresolution CPD algorithm. MuRAL-CPD leverages a wavelet-based multiresolution decomposition to detect changes across multiple temporal scales and incorporates user feedback to iteratively optimize key hyperparameters. This interaction enables the model to align its notion of change with that of the user, improving both accuracy and interpretability. Our experimental results on several real-world datasets show the effectiveness of MuRAL-CPD against state-of-the-art methods, particularly in scenarios where minimal supervision is available.
Authors: Zhiqiang Kou, Junyang Chen, Xin-Qiang Cai, Xiaobo Xia, Ming-Kun Xie, Dong-Dong Wu, Biao Liu, Yuheng Jia, Xin Geng, Masashi Sugiyama, Tat-Seng Chua
Abstract: Due to constraints on privacy, cost, and latency, on-premise deployment of small models is increasingly common. However, most practical pipelines stop at supervised fine-tuning (SFT) and fail to reach the reinforcement learning (RL) alignment stage. The main reason is that RL alignment typically requires either expensive human preference annotation or heavy reliance on high-quality reward models with large-scale API usage and ongoing engineering maintenance, both of which are ill-suited to on-premise settings. To bridge this gap, we propose a positive-unlabeled (PU) RL distillation method for on-premise small-model deployment. Without human-labeled preferences or a reward model, our method distills the teacher's preference-optimization capability from black-box generations into a locally trainable student. For each prompt, we query the teacher once to obtain an anchor response, locally sample multiple student candidates, and perform anchor-conditioned self-ranking to induce pairwise or listwise preferences, enabling a fully local training loop via direct preference optimization or group relative policy optimization. Theoretical analysis justifies that the induced preference signal by our method is order-consistent and concentrates on near-optimal candidates, supporting its stability for preference optimization. Experiments demonstrate that our method achieves consistently strong performance under a low-cost setting.
Authors: Enrique Valero-Leal, Bernd Bischl, Pedro Larra\~naga, Concha Bielza, Giuseppe Casalicchio
Abstract: Group counterfactual explanations find a set of counterfactual instances to explain a group of input instances contrastively. However, existing methods either (i) optimize counterfactuals only for a fixed group and do not generalize to new group members, (ii) strictly rely on strong model assumptions (e.g., linearity) for tractability or/and (iii) poorly control the counterfactual group geometry distortion. We instead learn an explicit optimal transport map that sends any group instance to its counterfactual without re-optimization, minimizing the group's total transport cost. This enables generalization with fewer parameters, making it easier to interpret the common actionable recourse. For linear classifiers, we prove that functions representing group counterfactuals are derived via mathematical optimization, identifying the underlying convex optimization type (QP, QCQP, ...). Experiments show that they accurately generalize, preserve group geometry and incur only negligible additional transport cost compared to baseline methods. If model linearity cannot be exploited, our approach also significantly outperforms the baselines.
Authors: Hao Liang, Jiayu Cheng, Sean R. Sinclair, Yali Du
Abstract: Exogenous MDPs (Exo-MDPs) capture sequential decision-making where uncertainty comes solely from exogenous inputs that evolve independently of the learner's actions. This structure is especially common in operations research applications such as inventory control, energy storage, and resource allocation, where exogenous randomness (e.g., demand, arrivals, or prices) drives system behavior. Despite decades of empirical evidence that greedy, exploitation-only methods work remarkably well in these settings, theory has lagged behind: all existing regret guarantees for Exo-MDPs rely on explicit exploration or tabular assumptions. We show that exploration is unnecessary. We propose Pure Exploitation Learning (PEL) and prove the first general finite-sample regret bounds for exploitation-only algorithms in Exo-MDPs. In the tabular case, PEL achieves $\widetilde{O}(H^2|\Xi|\sqrt{K})$. For large, continuous endogenous state spaces, we introduce LSVI-PE, a simple linear-approximation method whose regret is polynomial in the feature dimension, exogenous state space, and horizon, independent of the endogenous state and action spaces. Our analysis introduces two new tools: counterfactual trajectories and Bellman-closed feature transport, which together allow greedy policies to have accurate value estimates without optimism. Experiments on synthetic and resource-management tasks show that PEL consistently outperforming baselines. Overall, our results overturn the conventional wisdom that exploration is required, demonstrating that in Exo-MDPs, pure exploitation is enough.
Authors: Melika Mobini, Vincent Holst, Floriano Tori, Andres Algaba, Vincent Ginis
Abstract: Large language models are increasingly used to curate bibliographies, raising the question: are their reference lists distinguishable from human ones? We build paired citation graphs, ground truth and GPT-4o-generated (from parametric knowledge), for 10,000 focal papers ($\approx$ 275k references) from SciSciNet, and added a field-matched random baseline that preserves out-degree and field distributions while breaking latent structure. We compare (i) structure-only node features (degree/closeness/eigenvector centrality, clustering, edge count) with (ii) 3072-D title/abstract embeddings, using an RF on graph-level aggregates and Graph Neural Networks with node features. Structure alone barely separates GPT from ground truth (RF accuracy $\approx$ 0.60) despite cleanly rejecting the random baseline ($\approx$ 0.89--0.92). By contrast, embeddings sharply increase separability: RF on aggregated embeddings reaches $\approx$ 0.83, and GNNs with embedding node features achieve 93\% test accuracy on GPT vs.\ ground truth. We show the robustness of our findings by replicating the pipeline with Claude Sonnet 4.5 and with multiple embedding models (OpenAI and SPECTER), with RF separability for ground truth vs.\ Claude $\approx 0.77$ and clean rejection of the random baseline. Thus, LLM bibliographies, generated purely from parametric knowledge, closely mimic human citation topology, but leave detectable semantic fingerprints; detection and debiasing should target content signals rather than global graph structure.
Authors: Raul de la Rosa, Ivana Dusparic, Nicolas Cardozo
Abstract: Reinforcement Learning (RL) agents often struggle in real-world applications where environmental conditions are non-stationary, particularly when reward functions shift or the available action space expands. This paper introduces MORPHIN, a self-adaptive Q-learning framework that enables on-the-fly adaptation without full retraining. By integrating concept drift detection with dynamic adjustments to learning and exploration hyperparameters, MORPHIN adapts agents to changes in both the reward function and on-the-fly expansions of the agent's action space, while preserving prior policy knowledge to prevent catastrophic forgetting. We validate our approach using a Gridworld benchmark and a traffic signal control simulation. The results demonstrate that MORPHIN achieves superior convergence speed and continuous adaptation compared to a standard Q-learning baseline, improving learning efficiency by up to 1.7x.
Authors: Anchen Sun, Zhibin Chen, Xiaodong Cai
Abstract: The Cox Proportional Hazards (PH) model is widely used in survival analysis. Recently, artificial neural network (ANN)-based Cox-PH models have been developed. However, training these Cox models with high-dimensional features typically requires a substantial number of labeled samples containing information about time-to-event. The limited availability of labeled data for training often constrains the performance of ANN-based Cox models. To address this issue, we employed a deep semi-supervised learning (DSSL) approach to develop single- and multi-modal ANN-based Cox models based on the Mean Teacher (MT) framework, which utilizes both labeled and unlabeled data for training. We applied our model, named Cox-MT, to predict the prognosis of several types of cancer using data from The Cancer Genome Atlas (TCGA). Our single-modal Cox-MT models, utilizing TCGA RNA-seq data or whole slide images, significantly outperformed the existing ANN-based Cox model, Cox-nnet, using the same data set across four types of cancer considered. As the number of unlabeled samples increased, the performance of Cox-MT significantly improved with a given set of labeled data. Furthermore, our multi-modal Cox-MT model demonstrated considerably better performance than the single-modal model. In summary, the Cox-MT model effectively leverages both labeled and unlabeled data to significantly enhance prediction accuracy compared to existing ANN-based Cox models trained solely on labeled data.
Authors: Ziwei Liu, Borui Kang, Hangjie Yuan, Zixiang Zhao, Wei Li, Yifan Zhu, Tao Feng
Abstract: As digital environments (data distribution) are in flux, with new GUI data arriving over time-introducing new domains or resolutions-agents trained on static environments deteriorate in performance. In this work, we introduce Continual GUI Agents, a new task that requires GUI agents to perform continual learning under shifted domains and resolutions. We find existing methods fail to maintain stable grounding as GUI distributions shift over time, due to the diversity of UI interaction points and regions in fluxing scenarios. To address this, we introduce GUI-Anchoring in Flux (GUI-AiF), a new reinforcement fine-tuning framework that stabilizes continual learning through two novel rewards: Anchoring Point Reward in Flux (APR-iF) and Anchoring Region Reward in Flux (ARR-iF). These rewards guide the agents to align with shifting interaction points and regions, mitigating the tendency of existing reward strategies to over-adapt to static grounding cues (e.g., fixed coordinates or element scales). Extensive experiments show GUI-AiF surpasses state-of-the-art baselines. Our work establishes the first continual learning framework for GUI agents, revealing the untapped potential of reinforcement fine-tuning for continual GUI Agents.
Authors: Dawit Kiros Redie, Reza Arablouei, Stefan Werner
Abstract: Biased gradient compression with error feedback (EF) reduces communication in federated learning (FL), but under non-IID data, the residual error can decay slowly, causing gradient mismatch and stalled progress in the early rounds. We propose step-ahead partial error feedback (SA-PEF), which integrates step-ahead (SA) correction with partial error feedback (PEF). SA-PEF recovers EF when the step-ahead coefficient $\alpha=0$ and step-ahead EF (SAEF) when $\alpha=1$. For non-convex objectives and $\delta$-contractive compressors, we establish a second-moment bound and a residual recursion that guarantee convergence to stationarity under heterogeneous data and partial client participation. The resulting rates match standard non-convex Fed-SGD guarantees up to constant factors, achieving $O((\eta,\eta_0TR)^{-1})$ convergence to a variance/heterogeneity floor with a fixed inner step size. Our analysis reveals a step-ahead-controlled residual contraction $\rho_r$ that explains the observed acceleration in the early training phase. To balance SAEF's rapid warm-up with EF's long-term stability, we select $\alpha$ near its theory-predicted optimum. Experiments across diverse architectures and datasets show that SA-PEF consistently reaches target accuracy faster than EF.
Authors: Guoan Wang, Feiyu Wang, Zongwei Lv, Yikun Zong, Tong Yang
Abstract: As large language models (LLMs) continue to scale, deployment is increasingly bottlenecked by the memory wall, motivating a shift toward extremely low-bit quantization. However, most quantization-aware training (QAT) methods apply hard rounding and the straight-through estimator (STE) from the beginning of the training, which prematurely discretizes the optimization landscape and induces persistent gradient mismatch between latent weights and quantized weights, hindering effective optimization of quantized models. To address this, we propose Hestia, a Hessian-guided differentiable QAT framework for extremely low-bit LLMs, which replaces the rigid step function with a temperature-controlled softmax relaxation to maintain gradient flow early in training while progressively hardening quantization. Furthermore, Hestia leverages a tensor-wise Hessian trace metric as a lightweight curvature signal to drive fine-grained temperature annealing, enabling sensitivity-aware discretization across the model. Evaluations on Llama-3.2 show that Hestia consistently outperforms existing ternary QAT baselines, yielding average zero-shot improvements of 5.39% and 4.34% for the 1B and 3B models. These results indicate that Hessian-guided relaxation effectively recovers representational capacity, establishing a more robust training path for 1.58-bit LLMs. The code is available at https://github.com/hestia2026/Hestia.
Authors: Zhiheng Jiang, Yunzhe Wang, Ryan Marr, Ellen Novoseller, Benjamin T. Files, Volkan Ustun
Abstract: Preference-Conditioned Policy Learning (PCPL) in Multi-Objective Reinforcement Learning (MORL) aims to approximate diverse Pareto-optimal solutions by conditioning policies on user-specified preferences over objectives. This enables a single model to flexibly adapt to arbitrary trade-offs at run-time by producing a policy on or near the Pareto front. However, existing benchmarks for PCPL are largely restricted to toy tasks and fixed environments, limiting their realism and scalability. To address this gap, we introduce GraphAllocBench, a flexible benchmark built on a novel graph-based resource allocation sandbox environment inspired by city management, which we call CityPlannerEnv. GraphAllocBench provides a rich suite of problems with diverse objective functions, varying preference conditions, and high-dimensional scalability. We also propose two new evaluation metrics -- Proportion of Non-Dominated Solutions (PNDS) and Ordering Score (OS) -- that directly capture preference consistency while complementing the widely used hypervolume metric. Through experiments with Multi-Layer Perceptrons (MLPs) and graph-aware models, we show that GraphAllocBench exposes the limitations of existing MORL approaches and paves the way for using graph-based methods such as Graph Neural Networks in complex, high-dimensional combinatorial allocation tasks. Beyond its predefined problem set, GraphAllocBench enables users to flexibly vary objectives, preferences, and allocation rules, establishing it as a versatile and extensible benchmark for advancing PCPL. Code: https://anonymous.4open.science/r/GraphAllocBench
Authors: Elizabeth L. Baker, Alexander Denker, Jes Frellsen
Abstract: Score-based diffusion models have recently been extended to infinite-dimensional function spaces, with uses such as inverse problems arising from partial differential equations. In the Bayesian formulation of inverse problems, the aim is to sample from a posterior distribution over functions obtained by conditioning a prior on noisy observations. While diffusion models provide expressive priors in function space, the theory of conditioning them to sample from the posterior remains open. We address this, assuming that either the prior lies in the Cameron-Martin space, or is absolutely continuous with respect to a Gaussian measure. We prove that the models can be conditioned using an infinite-dimensional extension of Doob's $h$-transform, and that the conditional score decomposes into an unconditional score and a guidance term. As the guidance term is intractable, we propose a simulation-free score matching objective (called Supervised Guidance Training) enabling efficient and stable posterior sampling. We illustrate the theory with numerical examples on Bayesian inverse problems in function spaces. In summary, our work offers the first function-space method for fine-tuning trained diffusion models to accurately sample from a posterior.
Authors: Nan Qiao, Sheng Yue, Shuning Wang, Yongheng Deng, Ju Ren
Abstract: A fundamental challenge in offline reinforcement learning is distributional shift. Scarce data or datasets dominated by out-of-distribution (OOD) areas exacerbate this issue. Our theoretical analysis and experiments show that the standard squared error objective induces a harmful TD cross covariance. This effect amplifies in OOD areas, biasing optimization and degrading policy learning. To counteract this mechanism, we develop two complementary strategies: partitioned buffer sampling that restricts updates to localized replay partitions, attenuates irregular covariance effects, and aligns update directions, yielding a scheme that is easy to integrate with existing implementations, namely Clustered Cross-Covariance Control for TD (C^4). We also introduce an explicit gradient-based corrective penalty that cancels the covariance induced bias within each update. We prove that buffer partitioning preserves the lower bound property of the maximization objective, and that these constraints mitigate excessive conservatism in extreme OOD areas without altering the core behavior of policy constrained offline reinforcement learning. Empirically, our method showcases higher stability and up to 30% improvement in returns over prior methods, especially with small datasets and splits that emphasize OOD areas.
Authors: Shakhyar Gogoi
Abstract: COMET-SG1 is a lightweight, stability-oriented autoregressive regression model designed for time-series prediction on edge and embedded AI systems. Unlike recurrent neural networks or transformer-based sequence models, COMET-SG1 operates through linear behavior-space encoding, memory-anchored transition estimation, and deterministic state updates. This structure prioritizes bounded long-horizon behavior under fully autoregressive inference, a critical requirement for edge deployment where prediction errors accumulate over time. Experiments on non-stationary synthetic time-series data demonstrate that COMET-SG1 achieves competitive short-horizon accuracy while exhibiting significantly reduced long-horizon drift compared to MLP, LSTM, and k-nearest neighbor baselines. With a compact parameter footprint and operations compatible with fixed-point arithmetic, COMET-SG1 provides a practical and interpretable approach for stable autoregressive prediction in edge and embedded AI applications.
Authors: Rub\'en Jim\'enez, Oriol Pujol
Abstract: Deployed machine learning systems must continuously evolve as data, architectures, and regulations change, often without access to original training data or model internals. In such settings, black-box copying provides a practical refactoring mechanism, i.e. upgrading legacy models by learning replicas from input-output queries alone. When restricted to hard-label outputs, copying turns into a discontinuous surface reconstruction problem from pointwise queries, severely limiting the ability to recover boundary geometry efficiently. We propose a distance-based copying (distillation) framework that replaces hard-label supervision with signed distances to the teacher's decision boundary, converting copying into a smooth regression problem that exploits local geometry. We develop an $\alpha$-governed smoothing and regularization scheme with H\"older/Lipschitz control over the induced target surface, and introduce two model-agnostic algorithms to estimate signed distances under label-only access. Experiments on synthetic problems and UCI benchmarks show consistent improvements in fidelity and generalization accuracy over hard-label baselines, while enabling distance outputs as uncertainty-related signals for black-box replicas.
Authors: Steve Hanneke, Mingyue Xu
Abstract: Multitask learning and related frameworks have achieved tremendous success in modern applications. In multitask learning problem, we are given a set of heterogeneous datasets collected from related source tasks and hope to enhance the performance above what we could hope to achieve by solving each of them individually. The recent work of arXiv:2006.15785 has showed that, without access to distributional information, no algorithm based on aggregating samples alone can guarantee optimal risk as long as the sample size per task is bounded. In this paper, we focus on understanding the statistical limits of multitask learning. We go beyond the no-free-lunch theorem in arXiv:2006.15785 by establishing a stronger impossibility result of adaptation that holds for arbitrarily large sample size per task. This improvement conveys an important message that the hardness of multitask learning cannot be overcame by having abundant data per task. We also discuss the notion of optimal adaptivity that may be of future interests.
Authors: Arshia Soltani Moakhar, Tanapoom Laoaron, Faraz Ghahremani, Kiarash Banihashem, MohammadTaghi Hajiaghayi
Abstract: This paper advances the theoretical understanding of active learning label complexity for decision trees as binary classifiers. We make two main contributions. First, we provide the first analysis of the disagreement coefficient for decision trees-a key parameter governing active learning label complexity. Our analysis holds under two natural assumptions required for achieving polylogarithmic label complexity, (i) each root-to-leaf path queries distinct feature dimensions, and (ii) the input data has a regular, grid-like structure. We show these assumptions are essential, as relaxing them leads to polynomial label complexity. Second, we present the first general active learning algorithm for binary classification that achieves a multiplicative error guarantee, producing a $(1+\epsilon)$-approximate classifier. By combining these results, we design an active learning algorithm for decision trees that uses only a polylogarithmic number of label queries in the dataset size, under the stated assumptions. Finally, we establish a label complexity lower bound, showing our algorithm's dependence on the error tolerance $\epsilon$ is close to optimal.
Authors: Kaito Baba, Yoshihiko Ozaki, Shuhei Watanabe
Abstract: We propose conditional PED-ANOVA (condPED-ANOVA), a principled framework for estimating hyperparameter importance (HPI) in conditional search spaces, where the presence or domain of a hyperparameter can depend on other hyperparameters. Although the original PED-ANOVA provides a fast and efficient way to estimate HPI within the top-performing regions of the search space, it assumes a fixed, unconditional search space and therefore cannot properly handle conditional hyperparameters. To address this, we introduce a conditional HPI for top-performing regions and derive a closed-form estimator that accurately reflects conditional activation and domain changes. Experiments show that naive adaptations of existing HPI estimators yield misleading or uninterpretable importance estimates in conditional settings, whereas condPED-ANOVA consistently provides meaningful importances that reflect the underlying conditional structure.
Authors: Jonas H\"ubotter, Frederike L\"ubeck, Lejs Behric, Anton Baumann, Marco Bagatella, Daniel Marta, Ido Hakimi, Idan Shenfeld, Thomas Kleine Buening, Carlos Guestrin, Andreas Krause
Abstract: Large language models are increasingly post-trained with reinforcement learning in verifiable domains such as code and math. Yet, current methods for reinforcement learning with verifiable rewards (RLVR) learn only from a scalar outcome reward per attempt, creating a severe credit-assignment bottleneck. Many verifiable environments actually provide rich textual feedback, such as runtime errors or judge evaluations, that explain why an attempt failed. We formalize this setting as reinforcement learning with rich feedback and introduce Self-Distillation Policy Optimization (SDPO), which converts tokenized feedback into a dense learning signal without any external teacher or explicit reward model. SDPO treats the current model conditioned on feedback as a self-teacher and distills its feedback-informed next-token predictions back into the policy. In this way, SDPO leverages the model's ability to retrospectively identify its own mistakes in-context. Across scientific reasoning, tool use, and competitive programming on LiveCodeBench v6, SDPO improves sample efficiency and final accuracy over strong RLVR baselines. Notably, SDPO also outperforms baselines in standard RLVR environments that only return scalar feedback by using successful rollouts as implicit feedback for failed attempts. Finally, applying SDPO to individual questions at test time accelerates discovery on difficult binary-reward tasks, achieving the same discovery probability as best-of-k sampling or multi-turn conversations with 3x fewer attempts.
Authors: Steve Azzolin, Stefano Teso, Bruno Lepri, Andrea Passerini, Sagar Malhotra
Abstract: Explanations provided by Self-explainable Graph Neural Networks (SE-GNNs) are fundamental for understanding the model's inner workings and for identifying potential misuse of sensitive attributes. Although recent works have highlighted that these explanations can be suboptimal and potentially misleading, a characterization of their failure cases is unavailable. In this work, we identify a critical failure of SE-GNN explanations: explanations can be unambiguously unrelated to how the SE-GNNs infer labels. We show that, on the one hand, many SE-GNNs can achieve optimal true risk while producing these degenerate explanations, and on the other, most faithfulness metrics can fail to identify these failure modes. Our empirical analysis reveals that degenerate explanations can be maliciously planted (allowing an attacker to hide the use of sensitive attributes) and can also emerge naturally, highlighting the need for reliable auditing. To address this, we introduce a novel faithfulness metric that reliably marks degenerate explanations as unfaithful, in both malicious and natural settings. Our code is available in the supplemental.
Authors: Minwu Kim, Safal Shrestha, Keith Ross
Abstract: Reinforcement Learning with Verifiable Rewards (RLVR) has substantially improved the reasoning abilities of large language models (LLMs), yet training often stalls as problems become saturated. We identify the core challenge as the poor accessibility of informative failures: learning signals exist but are rarely encountered during standard rollouts. To address this, we propose failure-prefix conditioning, a simple and effective method for learning from saturated problems. Rather than starting from the original question, our approach reallocates exploration by conditioning training on prefixes derived from rare incorrect reasoning trajectories, thereby exposing the model to failure-prone states. We observe that failure-prefix conditioning yields performance gains matching those of training on medium-difficulty problems, while preserving token efficiency. Furthermore, we analyze the model's robustness, finding that our method reduces performance degradation under misleading failure prefixes, albeit with a mild trade-off in adherence to correct early reasoning. Finally, we demonstrate that an iterative approach, which refreshes failure prefixes during training, unlocks additional gains after performance plateaus. Overall, our results suggest that failure-prefix conditioning offers an effective pathway to extend RLVR training on saturated problems.
Authors: Brian Christian, Jessica A. F. Thompson, Elle Michelle Yang, Vincent Adam, Hannah Rose Kirk, Christopher Summerfield, Tsvetomira Dumbalska
Abstract: Reward models (RMs) are central to aligning large language models (LLMs) with human values but have received less attention than pre-trained and post-trained LLMs themselves. Because RMs are initialized from LLMs, they inherit representations that shape their behavior, but the nature and extent of this influence remain understudied. In a comprehensive study of 10 leading open-weight RMs using validated psycholinguistic corpora, we show that RMs exhibit significant differences along multiple dimensions of human value as a function of their base model. Using the "Big Two" psychological axes, we show a robust preference of Llama RMs for "agency" and a corresponding robust preference of Gemma RMs for "communion." This phenomenon holds even when the preference data and finetuning process are identical, and we trace it back to the logits of the respective instruction-tuned and pre-trained models. These log-probability differences themselves can be formulated as an implicit RM; we derive usable implicit reward scores and show that they exhibit the very same agency/communion difference. We run experiments training RMs with ablations for preference data source and quantity, which demonstrate that this effect is not only repeatable but surprisingly durable. Despite RMs being designed to represent human preferences, our evidence shows that their outputs are influenced by the pretrained LLMs on which they are based. This work underscores the importance of safety and alignment efforts at the pretraining stage, and makes clear that open-source developers' choice of base model is as much a consideration of values as of performance.
Authors: Zihao Wang, Hang Yin, Lihui Liu, Hanghang Tong, Yangqiu Song, Ginny Wong, Simon See
Abstract: This paper studies the minimal dimension required to embed subset memberships ($m$ elements and ${m\choose k}$ subsets of at most $k$ elements) into vector spaces, denoted as Minimal Embeddable Dimension (MED). The tight bounds of MED are derived theoretically and supported empirically for various notions of "distances" or "similarities," including the $\ell_2$ metric, inner product, and cosine similarity. In addition, we conduct numerical simulation in a more achievable setting, where the ${m\choose k}$ subset embeddings are chosen as the centroid of the embeddings of the contained elements. Our simulation easily realizes a logarithmic dependency between the MED and the number of elements to embed. These findings imply that embedding-based retrieval limitations stem primarily from learnability challenges, not geometric constraints, guiding future algorithm design.
Authors: Olaf Yunus Laitinen Imanov, Derya Umut Kulali, Taner Yilmaz
Abstract: Time series forecasting is a fundamental problem with applications in climate, energy, healthcare, and finance. Many existing approaches require domain-specific feature engineering and substantial labeled data for each task. We introduce PatchFormer, a patch-based time series foundation model that uses hierarchical masked reconstruction for self-supervised pretraining and lightweight adapters for efficient transfer. PatchFormer segments time series into patches and learns multiscale temporal representations with learnable aggregation across temporal scales. Pretraining uses masked patch reconstruction with dynamic masking and objectives that encourage both local accuracy and global consistency, followed by cross-domain knowledge distillation. Experiments on 24 benchmark datasets spanning weather, energy, traffic, finance, and healthcare demonstrate state-of-the-art zero-shot multi-horizon forecasting, reducing mean squared error by 27.3 percent relative to strong baselines while requiring 94 percent less task-specific training data. The model exhibits near log-linear scaling with more pretraining data up to 100 billion points and processes length-512 sequences 3.8x faster than full-sequence transformers.
Authors: Weixin Chen, Li Chen, Yuhan Zhao
Abstract: Despite growing efforts to mitigate unfairness in recommender systems, existing fairness-aware methods typically fix the fairness requirement at training time and provide limited post-training flexibility. However, in real-world scenarios, diverse stakeholders may demand differing fairness requirements over time, so retraining for different fairness requirements becomes prohibitive. To address this limitation, we propose Cofair, a single-train framework that enables post-training fairness control in recommendation. Specifically, Cofair introduces a shared representation layer with fairness-conditioned adapter modules to produce user embeddings specialized for varied fairness levels, along with a user-level regularization term that guarantees user-wise monotonic fairness improvements across these levels. We theoretically establish that the adversarial objective of Cofair upper bounds demographic parity and the regularization term enforces progressive fairness at user level. Comprehensive experiments on multiple datasets and backbone models demonstrate that our framework provides dynamic fairness at different levels, delivering comparable or better fairness-accuracy curves than state-of-the-art baselines, without the need to retrain for each new fairness requirement. Our code is publicly available at https://github.com/weixinchen98/Cofair.
Authors: Hao Sun, Da-Wei Zhou
Abstract: Traditional machine learning systems are typically designed for static data distributions, which suffer from catastrophic forgetting when learning from evolving data streams. Class-Incremental Learning (CIL) addresses this challenge by enabling learning systems to continuously learn new classes while preserving prior knowledge. With the rise of pre-trained models (PTMs) such as CLIP, leveraging their strong generalization and semantic alignment capabilities has become a promising direction in CIL. However, existing CLIP-based CIL methods are often scattered across disparate codebases, rely on inconsistent configurations, hindering fair comparisons, reproducibility, and practical adoption. Therefore, we propose C3Box (CLIP-based Class-inCremental learning toolBOX), a modular and comprehensive Python toolbox. C3Box integrates representative traditional CIL methods, ViT-based CIL methods, and state-of-the-art CLIP-based CIL methods into a unified CLIP-based framework. By inheriting the streamlined design of PyCIL, C3Box provides a JSON-based configuration and standardized execution pipeline. This design enables reproducible experimentation with low engineering overhead and makes C3Box a reliable benchmark platform for continual learning research. Designed to be user-friendly, C3Box relies only on widely used open-source libraries and supports major operating systems. The code is available at https://github.com/LAMDA-CL/C3Box.
Authors: An\'ibal Silva, Mois\'es Santos, Andr\'e Restivo, Carlos Soares
Abstract: Tabular data remains a challenging domain for generative models. In particular, the standard Variational Autoencoder (VAE) architecture, typically composed of multilayer perceptrons, struggles to model relationships between features, especially when handling mixed data types. In contrast, Transformers, through their attention mechanism, are better suited for capturing complex feature interactions. In this paper, we empirically investigate the impact of integrating Transformers into different components of a VAE. We conduct experiments on 57 datasets from the OpenML CC18 suite and draw two main conclusions. First, results indicate that positioning Transformers to leverage latent and decoder representations leads to a trade-off between fidelity and diversity. Second, we observe a high similarity between consecutive blocks of a Transformer in all components. In particular, in the decoder, the relationship between the input and output of a Transformer is approximately linear.
Authors: Immanuel Abdi, Akshat Gupta, Micah Mok, Alexander Lu, Nicholas Lee, Gopala Anumanchipalli
Abstract: One of the biggest missing capabilities in current AI systems is the ability to learn continuously after deployment. Implementing such continually learning systems have several challenges, one of which is the large memory requirement of gradient-based algorithms that are used to train state-of-the-art LLMs. Evolutionary Strategies (ES) have recently re-emerged as a gradient-free alternative to traditional learning algorithms and have shown encouraging performance on specific tasks in LLMs. In this paper, we perform a comprehensive analysis of ES and specifically evaluate its forgetting curves when training for an increasing number of update steps. We first find that ES is able to reach performance numbers close to GRPO for math and reasoning tasks with a comparable compute budget. However, and most importantly for continual learning, the performance gains in ES is accompanied by significant forgetting of prior abilities, limiting its applicability for training models online. We also explore the reason behind this behavior and show that the updates made using ES are much less sparse and have orders of magnitude larger $\ell_2$ norm compared to corresponding GRPO updates, explaining the contrasting forgetting curves between the two algorithms. With this study, we aim to highlight the issue of forgetting in gradient-free algorithms like ES and hope to inspire future work to mitigate these issues.
Authors: Jingxin Wang, Shitong Guo, Ruicheng Dai, Wenhui Liang, Ruogu Ding, Xin Ning, Weikang Qian
Abstract: Targeting error-tolerant applications, approximate circuits introduce controlled errors to significantly improve performance, power, and area (PPA) of circuits. In this work, we introduce GTAC, a novel generative Transformer-based model for producing approximate circuits. By leveraging principles of approximate computing and AI-driven EDA, our model innovatively integrates error thresholds into the design process. Experimental results show that compared with a state-of-the-art method, GTAC further reduces 6.4% area under the error rate constraint, while being 4.3x faster.
Authors: Yanru Chen, Runyang Tian, Yue Pan, Zheyu Li, Weihong Xu, Tajana Rosing
Abstract: The proliferation of large language models (LLMs) is accelerating the integration of multimodal assistants into edge devices, where inference is executed under stringent latency and energy constraints, often exacerbated by intermittent connectivity. These challenges become particularly acute in the context of multimodal LLMs (MLLMs), as high-dimensional visual inputs are transformed into extensive token sequences, thereby inflating the key-value (KV) cache and imposing substantial data movement overheads to the LLM backbone. To address these issues, we present CHIME, a chiplet-based heterogeneous near-memory acceleration for edge MLLMs inference. CHIME leverages the complementary strengths of integrated monolithic 3D (M3D) DRAM and RRAM chiplets: DRAM supplies low-latency bandwidth for attention, while RRAM offers dense, non-volatile storage for weights. This heterogeneous hardware is orchestrated by a co-designed mapping framework that executes fused kernels near data, minimizing cross-chiplet traffic to maximize effective bandwidth. On FastVLM (0.6B/1.7B) and MobileVLM (1.7B/3B), CHIME achieves up to 54x speedup and up to 246x better energy efficiency per inference as compared to the edge GPU NVIDIA Jetson Orin NX. It sustains 116.5-266.5 token/J compared to Jetson's 0.7-1.1 token/J. Furthermore, it delivers up to 69.2x higher throughput than the state-of-the-art PIM accelerator FACIL. Compared to the M3D DRAM-only design, CHIME's heterogeneous memory further improves energy efficiency by 7% and performance by 2.4x.
Authors: Yuval Harary, Almog Sharoni, Esteban Garz\'on, Marco Lanuzza, Adam Teman, Leonid Yavits
Abstract: Binary Neural Networks (BNNs), where weights and activations are constrained to binary values (+1, -1), are a highly efficient alternative to traditional neural networks. Unfortunately, typical BNNs, while binarizing linear layers (matrix-vector multiplication), still implement other network layers (batch normalization, softmax, output layer, and sometimes the input layer of a convolutional neural network) in full precision. This limits the area and energy benefits and requires architectural support for full precision operations. We propose PiC-BNN, a true end-to-end binary in-approximate search (Hamming distance tolerant) Content Addressable Memory based BNN accelerator. PiC-BNN is designed and manufactured in a commercial 65nm process. PiC-BNN uses Hamming distance tolerance to apply the law of large numbers to enable accurate classification without implementing full precision operations. PiC-BNN achieves baseline software accuracy (95.2%) on the MNIST dataset and 93.5% on the Hand Gesture (HG) dataset, a throughput of 560K inferences/s, and presents a power efficiency of 703M inferences/s/W when implementing a binary MLP model for MNIST/HG dataset classification.
Authors: Yitian Chen, Cheng Cheng, Yinan Sun, Zi Ling, Dongdong Ge
Abstract: Large Language Models (LLMs) have demonstrated impressive progress in optimization modeling, fostering a rapid expansion of new methodologies and evaluation benchmarks. However, the boundaries of their capabilities in automated formulation and problem solving remain poorly understood, particularly when extending to complex, real-world tasks. To bridge this gap, we propose OPT-ENGINE, an extensible benchmark framework designed to evaluate LLMs on optimization modeling with controllable and scalable difficulty levels. OPT-ENGINE spans 10 canonical tasks across operations research, with five Linear Programming and five Mixed-Integer Programming. Utilizing OPT-ENGINE, we conduct an extensive study of LLMs' reasoning capabilities, addressing two critical questions: 1.) Do LLMs' performance remain robust when generalizing to out-of-distribution optimization tasks that scale in complexity beyond current benchmark levels? and 2.) At what stage, from problem interpretation to solution generation, do current LLMs encounter the most significant bottlenecks? Our empirical results yield two key insights: first, tool-integrated reasoning with external solvers exhibits significantly higher robustness as task complexity escalates, while pure-text reasoning reaches a ceiling; second, the automated formulation of constraints constitutes the primary performance bottleneck. These findings provide actionable guidance for developing next-generation LLMs for advanced optimization. Our code is publicly available at \textcolor{blue}{https://github.com/Cardinal-Operations/OPTEngine}.
Authors: Kei Saito
Abstract: Non-Resolution Reasoning (NRR) provides a formal framework for maintaining semantic ambiguity rather than forcing premature interpretation collapse. While the foundational architecture establishes state spaces and operators for ambiguity-preserving computation, the critical question of how natural language maps to these mathematical structures remains open. This paper introduces the text-to-state mapping function {\phi} that transforms linguistic input into superposition states within the NRR framework. We formalize the Contradiction-Preservation Principle, which requires that genuinely ambiguous expressions maintain non-zero entropy in their state representations, and develop extraction protocols using existing Large Language Models as interpretation generators. Empirical validation across 68 test sentences spanning lexical, structural, and pragmatic ambiguity demonstrates that our mapping achieves mean Shannon entropy H(S) = 1.087 bits for ambiguous inputs while baseline single-interpretation approaches yield H(S) = 0.000. The framework provides the missing algorithmic bridge between raw text and the formal state spaces on which NRR operators act, enabling architectural collapse deferment in language model inference.
Authors: Nikhil Raghav, Avisek Gupta, Swagatam Das, Md Sahidullah
Abstract: Speaker diarization aims to segment audio recordings into regions corresponding to individual speakers. Although unsupervised speaker diarization is inherently challenging, the prospect of identifying speaker regions without pretraining or weak supervision motivates research on clustering techniques. In this work, we share the notable observation that measuring multiple kernel similarities of speaker embeddings to thereafter craft a sparse graph for spectral clustering in a principled manner is sufficient to achieve state-of-the-art performances in a fully unsupervised setting. Specifically, we consider four polynomial kernels and a degree one arccosine kernel to measure similarities in speaker embeddings, using which sparse graphs are constructed in a principled manner to emphasize local similarities. Experiments show the proposed approach excels in unsupervised speaker diarization over a variety of challenging environments in the DIHARD-III, AMI, and VoxConverse corpora. To encourage further research, our implementations are available at https://github.com/nikhilraghav29/MK-SGC-SC.
Authors: Jonathan Vacher (MAP5 - UMR 8145)
Abstract: Deep neural networks (DNNs) achieve remarkable performance on a wide range of tasks, yet their mathematical analysis remains fragmented: stability and generalization are typically studied in disparate frameworks and on a case-by-case basis. Architecturally, DNNs rely on the recursive application of parametrized functions, a mechanism that can be unstable and difficult to train, making stability a primary concern. Even when training succeeds, there are few rigorous results on how well such models generalize beyond the observed data, especially in the generative setting. In this work, we leverage the theory of stochastic Iterated Function Systems (IFS) and show that two important deep architectures can be viewed as, or canonically associated with, place-dependent IFS. This connection allows us to import results from random dynamical systems to (i) establish the existence and uniqueness of invariant measures under suitable contractivity assumptions, and (ii) derive a Wasserstein generalization bound for generative modeling. The bound naturally leads to a new training objective that directly controls the collage-type approximation error between the data distribution and its image under the learned transfer operator. We illustrate the theory on a controlled 2D example and empirically evaluate the proposed objective on standard image datasets (MNIST, CelebA, CIFAR-10).
Authors: Jonas Elsborg, Felix {\AE}rtebjerg, Luca Thiede, Al\'an Aspuru-Guzik, Tejs Vegge, Arghya Bhowmik
Abstract: We introduce ELECTRAFI, a fast, end-to-end differentiable model for predicting periodic charge densities in crystalline materials. ELECTRAFI constructs anisotropic Gaussians in real space and exploits their closed-form Fourier transforms to analytically evaluate plane-wave coefficients via the Poisson summation formula. This formulation delegates non-local and periodic behavior to analytic transforms, enabling reconstruction of the full periodic charge density with a single inverse FFT. By avoiding explicit real-space grid probing, periodic image summation, and spherical harmonic expansions, ELECTRAFI matches or exceeds state-of-the-art accuracy across periodic benchmarks while being up to $633 \times$ faster than the strongest competing method, reconstructing crystal charge densities in a fraction of a second. When used to initialize DFT calculations, ELECTRAFI reduces total DFT compute cost by up to ~20%, whereas slower charge density models negate savings due to high inference times. Our results show that accuracy and inference cost jointly determine end-to-end DFT speedups, and motivate our focus on efficiency.
Authors: Haoyuan Deng, Yuanjiang Xue, Haoyang Du, Boyang Zhou, Zhenyu Wu, Ziwei Wang
Abstract: Human-in-the-loop guidance has emerged as an effective approach for enabling faster convergence in online reinforcement learning (RL) of complex real-world manipulation tasks. However, existing human-in-the-loop RL (HiL-RL) frameworks often suffer from low sample efficiency, requiring substantial human interventions to achieve convergence and thereby leading to high labor costs. To address this, we propose a sample-efficient real-world human-in-the-loop RL framework named \method, which requires fewer human intervention by actively selecting informative samples. Specifically, stable reduction of policy entropy enables improved trade-off between exploration and exploitation with higher sample efficiency. We first build influence functions of different samples on the policy entropy, which is efficiently estimated by the covariance of action probabilities and soft advantages of policies. Then we select samples with moderate values of influence functions, where shortcut samples that induce sharp entropy drops and noisy samples with negligible effect are pruned. Extensive experiments on four real-world manipulation tasks demonstrate that \method achieves a 42.1\% higher success rate while requiring 10.1\% fewer human interventions compared to the state-of-the-art HiL-RL method, validating its effectiveness. The project page providing code, videos, and mathematical formulations can be found at https://e2hil.github.io/.
Authors: Nourin Shahin, Izzat Alsmadi
Abstract: As large language models (LLMs) move from research prototypes to enterprise systems, their security vulnerabilities pose serious risks to data privacy and system integrity. This study benchmarks various Llama model variants against the OWASP Top 10 for LLM Applications framework, evaluating threat detection accuracy, response safety, and computational overhead. Using the FABRIC testbed with NVIDIA A30 GPUs, we tested five standard Llama models and five Llama Guard variants on 100 adversarial prompts covering ten vulnerability categories. Our results reveal significant differences in security performance: the compact Llama-Guard-3-1B model achieved the highest detection rate of 76% with minimal latency (0.165s per test), whereas base models such as Llama-3.1-8B failed to detect threats (0% accuracy) despite longer inference times (0.754s). We observe an inverse relationship between model size and security effectiveness, suggesting that smaller, specialized models often outperform larger general-purpose ones in security tasks. Additionally, we provide an open-source benchmark dataset including adversarial prompts, threat labels, and attack metadata to support reproducible research in AI security, [1].
Authors: Temple He, Jaeha Lee, Hirosi Ooguri
Abstract: We develop a reinforcement learning algorithm to study the holographic entropy cone. Given a target entropy vector, our algorithm searches for a graph realization whose min-cut entropies match the target vector. If the target vector does not admit such a graph realization, it must lie outside the cone, in which case the algorithm finds a graph whose corresponding entropy vector most nearly approximates the target and allows us to probe the location of the facets. For the $\sf N=3$ cone, we confirm that our algorithm successfully rediscovers monogamy of mutual information beginning with a target vector outside the holographic entropy cone. We then apply the algorithm to the $\sf N=6$ cone, analyzing the 6 "mystery" extreme rays of the subadditivity cone from arXiv:2412.15364 that satisfy all known holographic entropy inequalities yet lacked graph realizations. We found realizations for 3 of them, proving they are genuine extreme rays of the holographic entropy cone, while providing evidence that the remaining 3 are not realizable, implying unknown holographic inequalities exist for $\sf N=6$.
Authors: J. Ben Tamo, Daniel Carlander-Reuterfelt, Jonathan Rubin, Dezhi Hong, Mingxian Wang, Oleg Poliannikov
Abstract: Despite multilingual pretraining, large language models often struggle with non-English tasks, particularly in language control, the ability to respond in the intended language. We identify and characterize two key failure modes: the multilingual transfer bottleneck (correct language, incorrect task response) and the language consistency bottleneck (correct task response, wrong language). To systematically surface these issues, we design a four-scenario evaluation protocol spanning MMLU, MGSM, and XQuAD benchmarks. To probe these issues with interpretability, we extend logit lens analysis to track language probabilities layer by layer and compute cross-lingual semantic similarity of hidden states. The results reveal a three-phase internal structure: early layers align inputs into a shared semantic space, middle layers perform task reasoning, and late layers drive language-specific generation. Guided by these insights, we introduce selective fine-tuning of only the final layers responsible for language control. On Qwen-3-32B and Bloom-7.1B, this method achieves over 98 percent language consistency across six languages while fine-tuning only 3-5 percent of parameters, without sacrificing task accuracy. Importantly, this result is nearly identical to that of full-scope fine-tuning (for example, above 98 percent language consistency for both methods across all prompt scenarios) but uses a fraction of the computational resources. To the best of our knowledge, this is the first approach to leverage layer-localization of language control for efficient multilingual adaptation.
Authors: Suyash Deshmukh, Chayan Chatterjee, Abigail Petulante, Tabata Aira Ferreira, Karan Jani
Abstract: Transient noise artifacts, or glitches, fundamentally limit the sensitivity of gravitational-wave (GW) interferometers and can mimic true astrophysical signals, particularly the short-duration intermediate-mass black hole (IMBH) mergers. Current glitch classification methods, such as Gravity Spy, rely on supervised models trained from scratch using labeled datasets. These approaches suffer from a significant ``label bottleneck," requiring massive, expertly annotated datasets to achieve high accuracy and often struggling to generalize to new glitch morphologies or exotic GW signals encountered in observing runs. In this work, we present a novel cross-domain framework that treats GW strain data through the lens of audio processing. We utilize the Audio Spectrogram Transformer (AST), a model pre-trained on large-scale audio datasets, and adapt it to the GW domain. Instead of learning time-frequency features from scratch, our method exploits the strong inductive bias inherent in pre-trained audio models, transferring learned representations of natural sound to the characterization of detector noise and GW signals, including IMBHs. We validate this approach by analyzing strain data from the third (O3) and fourth (O4) observing runs of the LIGO detectors. We used t-Distributed Stochastic Neighbor Embedding (t-SNE), an unsupervised clustering technique, to visualize the AST-derived embeddings of signals and glitches, revealing well-separated groups that align closely with independently validated Gravity Spy glitch classes. Our results indicate that the inductive bias from audio pre-training allows superior feature extraction compared to traditional supervised techniques, offering a robust, data-efficient pathway for discovering new, anomalous transients, and classifying complex noise artifacts in the era of next-generation detectors.
Authors: Divit Rawal, Sriram Vishwanath
Abstract: We prove an exponential separation in sample complexity between Euclidean and hyperbolic representations for learning on hierarchical data under standard Lipschitz regularization. For depth-$R$ hierarchies with branching factor $m$, we first establish a geometric obstruction for Euclidean space: any bounded-radius embedding forces volumetric collapse, mapping exponentially many tree-distant points to nearby locations. This necessitates Lipschitz constants scaling as $\exp(\Omega(R))$ to realize even simple hierarchical targets, yielding exponential sample complexity under capacity control. We then show this obstruction vanishes in hyperbolic space: constant-distortion hyperbolic embeddings admit $O(1)$-Lipschitz realizability, enabling learning with $n = O(mR \log m)$ samples. A matching $\Omega(mR \log m)$ lower bound via Fano's inequality establishes that hyperbolic representations achieve the information-theoretic optimum. We also show a geometry-independent bottleneck: any rank-$k$ prediction space captures only $O(k)$ canonical hierarchical contrasts.
Authors: Gautham Vinod, Bruce Coburn, Siddeshwar Raghavan, Jiangpeng He, Fengqing Zhu
Abstract: The rise of chronic diseases related to diet, such as obesity and diabetes, emphasizes the need for accurate monitoring of food intake. While AI-driven dietary assessment has made strides in recent years, the ill-posed nature of recovering size (portion) information from monocular images for accurate estimation of ``how much did you eat?'' is a pressing challenge. Some 3D reconstruction methods have achieved impressive geometric reconstruction but fail to recover the crucial real-world scale of the reconstructed object, limiting its usage in precision nutrition. In this paper, we bridge the gap between 3D computer vision and digital health by proposing a method that recovers a true-to-scale 3D reconstructed object from a monocular image. Our approach leverages rich visual features extracted from models trained on large-scale datasets to estimate the scale of the reconstructed object. This learned scale enables us to convert single-view 3D reconstructions into true-to-life, physically meaningful models. Extensive experiments and ablation studies on two publicly available datasets show that our method consistently outperforms existing techniques, achieving nearly a 30% reduction in mean absolute volume-estimation error, showcasing its potential to enhance the domain of precision nutrition. Code: https://gitlab.com/viper-purdue/size-matters
Authors: Francisco Alc\'antara-\'Avila, Andr\'es Cremades, Sergio Hoyas, Ricardo Vinuesa
Abstract: In this work, we investigate the physical mechanisms governing turbulent kinetic energy transport using explainable deep learning (XDL). An XDL model based on SHapley Additive exPlanations (SHAP) is used to identify and percolate high-importance structures for the evolution of the turbulent kinetic energy budget terms of a turbulent channel flow at a friction Reynolds number of $Re_\tau = 125$. The results show that the important structures are predominantly located in the near-wall region and are more frequently associated with sweep-type events. In the viscous layer, the SHAP structures relevant for production and viscous diffusion are almost entirely contained within those relevant for dissipation, revealing a clear hierarchical organization of near-wall turbulence. In the outer layer, this hierarchical organization breaks down and only velocity-pressure-gradient correlation and turbulent transport SHAP structures remain, with a moderate spatial coincidence of approximately $60\%$. Finally, we show that none of the coherent structures classically studied in turbulence are capable of representing the mechanisms behind the various terms of the turbulent kinetic energy budget throughout the channel. These results reveal dissipation as the dominant organizing mechanism of near-wall turbulence, constraining production and viscous diffusion within a single structural hierarchy that breaks down in the outer layer.
Authors: Atik Faysal, Mohammad Rostami, Reihaneh Gh. Roshan, Nikhil Muralidhar, Huaxia Wang
Abstract: We address the challenge of training Vision Transformers (ViTs) when labeled data is scarce but unlabeled data is abundant. We propose Semi-Supervised Masked Autoencoder (SSMAE), a framework that jointly optimizes masked image reconstruction and classification using both unlabeled and labeled samples with dynamically selected pseudo-labels. SSMAE introduces a validation-driven gating mechanism that activates pseudo-labeling only after the model achieves reliable, high-confidence predictions that are consistent across both weakly and strongly augmented views of the same image, reducing confirmation bias. On CIFAR-10 and CIFAR-100, SSMAE consistently outperforms supervised ViT and fine-tuned MAE, with the largest gains in low-label regimes (+9.24% over ViT on CIFAR-10 with 10% labels). Our results demonstrate that when pseudo-labels are introduced is as important as how they are generated for data-efficient transformer training. Codes are available at https://github.com/atik666/ssmae.
Authors: Abhishek Chakraborty, Angelia Nedi\'c
Abstract: We consider minimizing an objective function subject to constraints defined by the intersection of lower-level sets of convex functions. We study two cases: (i) strongly convex and Lipschitz-smooth objective function and (ii) convex but possibly nonsmooth objective function. To deal with the constraints that are not easy to project on, we use a randomized feasibility algorithm with Polyak steps and a random number of sampled constraints per iteration, while taking (sub)gradient steps to minimize the objective function. For case (i), we prove linear convergence in expectation of the objective function values to any prescribed tolerance using an adaptive stepsize. For case (ii), we develop a fully problem parameter-free and adaptive stepsize scheme that yields an $O(1/\sqrt{T})$ worst-case rate in expectation. The infeasibility of the iterates decreases geometrically with the number of feasibility updates almost surely, while for the averaged iterates, we establish an expected lower bound on the function values relative to the optimal value that depends on the distribution for the random number of sampled constraints. For certain choices of sample-size growth, optimal rates are achieved. Finally, simulations on a Quadratically Constrained Quadratic Programming (QCQP) problem and Support Vector Machines (SVM) demonstrate the computational efficiency of our algorithm compared to other state-of-the-art methods.
Authors: Lee Xiong, Zhirong Chen, Rahul Mayuranath, Shangran Qiu, Arda Ozdemir, Lu Li, Yang Hu, Dave Li, Jingtao Ren, Howard Cheng, Fabian Souto Herrera, Ahmed Agiza, Baruch Epshtein, Anuj Aggarwal, Julia Ulziisaikhan, Chao Wang, Dinesh Ramasamy, Parshva Doshi, Sri Reddy, Arnold Overwijk
Abstract: We present LLaTTE (LLM-Style Latent Transformers for Temporal Events), a scalable transformer architecture for production ads recommendation. Through systematic experiments, we demonstrate that sequence modeling in recommendation systems follows predictable power-law scaling similar to LLMs. Crucially, we find that semantic features bend the scaling curve: they are a prerequisite for scaling, enabling the model to effectively utilize the capacity of deeper and longer architectures. To realize the benefits of continued scaling under strict latency constraints, we introduce a two-stage architecture that offloads the heavy computation of large, long-context models to an asynchronous upstream user model. We demonstrate that upstream improvements transfer predictably to downstream ranking tasks. Deployed as the largest user model at Meta, this multi-stage framework drives a 4.3\% conversion uplift on Facebook Feed and Reels with minimal serving overhead, establishing a practical blueprint for harnessing scaling laws in industrial recommender systems.
Authors: Darshan Deshpande, Anand Kannappan, Rebecca Qian
Abstract: Recent advances in reinforcement learning for code generation have made robust environments essential to prevent reward hacking. As LLMs increasingly serve as evaluators in code-based RL, their ability to detect reward hacking remains understudied. In this paper, we propose a novel taxonomy of reward exploits spanning across 54 categories and introduce TRACE (Testing Reward Anomalies in Code Environments), a synthetically curated and human-verified benchmark containing 517 testing trajectories. Unlike prior work that evaluates reward hack detection in isolated classification scenarios, we contrast these evaluations with a more realistic, contrastive anomaly detection setup on TRACE. Our experiments reveal that models capture reward hacks more effectively in contrastive settings than in isolated classification settings, with GPT-5.2 with highest reasoning mode achieving the best detection rate at 63%, up from 45% in isolated settings on TRACE. Building on this insight, we demonstrate that state-of-the-art models struggle significantly more with semantically contextualized reward hacks compared to syntactically contextualized ones. We further conduct qualitative analyses of model behaviors, as well as ablation studies showing that the ratio of benign to hacked trajectories and analysis cluster sizes substantially impact detection performance. We release the benchmark and evaluation harness to enable the community to expand TRACE and evaluate their models.
Authors: Abha Jha, Akanksha Mahajan, Ashwath Vaithinathan Aravindan, Praveen Saravanan, Sai Sailaja Policharla, Sonal Chaturbhuj Gehlot
Abstract: Large Language Models (LLMs) often produce hallucinated or unverifiable content, undermining their reliability in factual domains. This work investigates Reinforcement Learning with Verifiable Rewards (RLVR) as a training paradigm that explicitly rewards abstention ("I don't know") alongside correctness to promote intellectual humility. We fine-tune and evaluate Granite-3.3-2B-Instruct and Qwen-3-4B-Instruct on the MedMCQA and Hendrycks Math benchmarks using a ternary reward structure ($-1$, r_abs, 1) under varying abstention reward structures. We further study the effect of combining RLVR with supervised fine-tuning strategies that teach abstention prior to reinforcement learning. Our results show that moderate abstention rewards (r_abs $\approx -0.25$ to 0.3) consistently reduce incorrect responses without severe accuracy degradation on multiple-choice tasks, with larger models exhibiting greater robustness to abstention incentives. On open-ended question answering, we observe limitations due to insufficient exploration, which can be partially mitigated through supervised abstention training. Overall, these findings demonstrate the feasibility and flexibility of verifiable reward design as a practical approach for hallucination mitigation in language models. Reproducible code for our abstention training framework is available here https://github.com/Mystic-Slice/rl-abstention.
Authors: Marcus Emmanuel Barnes, Taher A. Ghaleb, Safwat Hassan
Abstract: Logs are essential for understanding Continuous Integration (CI) behavior, particularly for diagnosing build failures and performance regressions. Yet their growing volume and verbosity make both manual inspection and automated analysis increasingly costly, time-consuming, and environmentally costly. While prior work has explored log compression, anomaly detection, and LLM-based log analysis, most efforts target structured system logs rather than the unstructured, noisy, and verbose logs typical of CI workflows. We present LogSieve, a lightweight, RCA-aware and semantics-preserving log reduction technique that filters low-information lines while retaining content relevant to downstream reasoning. Evaluated on CI logs from 20 open-source Android projects using GitHub Actions, LogSieve achieves an average 42% reduction in lines and 40% reduction in tokens with minimal semantic loss. This pre-inference reduction lowers computational cost and can proportionally reduce energy use (and associated emissions) by decreasing the volume of data processed during LLM inference. Compared with structure-first baselines (LogZip and random-line removal), LogSieve preserves much higher semantic and categorical fidelity (Cosine = 0.93, GPTScore = 0.93, 80% exact-match accuracy). Embedding-based classifiers automate relevance detection with near-human accuracy (97%), enabling scalable and sustainable integration of semantics-aware filtering into CI workflows. LogSieve thus bridges log management and LLM reasoning, offering a practical path toward greener and more interpretable CI automation.
Authors: Rapha\"el Langevin
Abstract: Finite mixture models are widely used in econometric analyses to capture unobserved heterogeneity. This paper shows that maximum likelihood estimation of finite mixtures of parametric densities can suffer from substantial finite-sample bias in all parameters under mild regularity conditions. The bias arises from the influence of outliers in component densities with unbounded or large support and increases with the degree of overlap among mixture components. I show that maximizing the classification-mixture likelihood function, equipped with a consistent classifier, yields parameter estimates that are less biased than those obtained by standard maximum likelihood estimation (MLE). I then derive the asymptotic distribution of the resulting estimator and provide conditions under which oracle efficiency is achieved. Monte Carlo simulations show that conventional mixture MLE exhibits pronounced finite-sample bias, which diminishes as the sample size or the statistical distance between component densities tends to infinity. The simulations further show that the proposed estimation strategy generally outperforms standard MLE in finite samples in terms of both bias and mean squared errors under relatively weak assumptions. An empirical application to latent group panel structures using health administrative data shows that the proposed approach reduces out-of-sample prediction error by approximately 17.6% relative to the best results obtained from standard MLE procedures.
Authors: Weizhou Wang, Xuanxi Zhang, Jonathan Weare, Aaron R. Dinner
Abstract: Accurate simulation of nuclear quantum effects is essential for molecular modeling but expensive using path integral molecular dynamics (PIMD). We present GG-PI, a ring-polymer-based framework that combines generative modeling of the single-bead conditional density with Gibbs sampling to recover quantum statistics from classical simulation data. GG-PI uses inexpensive standard classical simulations or existing data for training and allows transfer across temperatures without retraining. On standard test systems, GG-PI significantly reduces wall clock time compared to PIMD. Our approach extends easily to a wide range of problems with similar Markov structure.
Authors: Parisa Fard Moshiri, Poonam Lohan, Burak Kantarci, Emil Janulewicz
Abstract: Service Function Chaining (SFC) requires efficient placement of Virtual Network Functions (VNFs) to satisfy diverse service requirements while maintaining high resource utilization in Data Centers (DCs). Conventional static resource allocation often leads to overprovisioning or underprovisioning due to the dynamic nature of traffic loads and application demands. To address this challenge, we propose a hybrid forecast-driven Deep reinforcement learning (DRL) framework that combines predictive intelligence with SFC provisioning. Specifically, we leverage DRL to generate datasets capturing DC resource utilization and service demands, which are then used to train deep learning forecasting models. Using Optuna-based hyperparameter optimization, the best-performing models, Spatio-Temporal Graph Neural Network, Temporal Graph Neural Network, and Long Short-Term Memory, are combined into an ensemble to enhance stability and accuracy. The ensemble predictions are integrated into the DC selection process, enabling proactive placement decisions that consider both current and future resource availability. Experimental results demonstrate that the proposed method not only sustains high acceptance ratios for resource-intensive services such as Cloud Gaming and VoIP but also significantly improves acceptance ratios for latency-critical categories such as Augmented Reality increases from 30% to 50%, while Industry 4.0 improves from 30% to 45%. Consequently, the prediction-based model achieves significantly lower E2E latencies of 20.5%, 23.8%, and 34.8% reductions for VoIP, Video Streaming, and Cloud Gaming, respectively. This strategy ensures more balanced resource allocation, and reduces contention.
Authors: Skyler Wu, Yash Nair, Emmanuel J. Cand\'es
Abstract: Exhaustively evaluating many large language models (LLMs) on a large suite of benchmarks is expensive. We cast benchmarking as finite-population inference and, under a fixed query budget, seek tight confidence intervals (CIs) for model accuracy with valid frequentist coverage. We propose Factorized Active Querying (FAQ), which (a) leverages historical information through a Bayesian factor model; (b) adaptively selects questions using a hybrid variance-reduction/active-learning sampling policy; and (c) maintains validity through Proactive Active Inference -- a finite-population extension of active inference (Zrnic & Candes, 2024) that enables direct question selection while preserving coverage. With negligible overhead cost, FAQ delivers up to $5\times$ effective sample size gains over strong baselines on two benchmark suites, across varying historical-data missingness levels: this means that it matches the CI width of uniform sampling while using up to $5\times$ fewer queries. We release our source code and our curated datasets to support reproducible evaluation and future research.
Authors: Jie Tang, Chuanlong Xie, Xianli Zeng, Lixing Zhu
Abstract: Machine learning models in high-stakes applications, such as recidivism prediction and automated personnel selection, often exhibit systematic performance disparities across sensitive subpopulations, raising critical concerns regarding algorithmic bias. Fairness auditing addresses these risks through two primary functions: certification, which verifies adherence to fairness constraints; and flagging, which isolates specific demographic groups experiencing disparate treatment. However, existing auditing techniques are frequently limited by restrictive distributional assumptions or prohibitive computational overhead. We propose a novel empirical likelihood-based (EL) framework that constructs robust statistical measures for model performance disparities. Unlike traditional methods, our approach is non-parametric; the proposed disparity statistics follow asymptotically chi-square or mixed chi-square distributions, ensuring valid inference without assuming underlying data distributions. This framework uses a constrained optimization profile that admits stable numerical solutions, facilitating both large-scale certification and efficient subpopulation discovery. Empirically, the EL methods outperform bootstrap-based approaches, yielding coverage rates closer to nominal levels while reducing computational latency by several orders of magnitude. We demonstrate the practical utility of this framework on the COMPAS dataset, where it successfully flags intersectional biases, specifically identifying a significantly higher positive prediction rate for African-American males under 25 and a systemic under-prediction for Caucasian females relative to the population mean.
Authors: Tanmay Karmakar, Sourav Saha, Debapriyo Majumdar, Surjyanee Halder
Abstract: Neural ranking models (NRMs) achieve strong retrieval effectiveness, yet prior work has shown they are vulnerable to adversarial perturbations. We revisit this robustness question with a minimal, query-aware attack that promotes a target document by inserting or substituting a single, semantically aligned word - the query center. We study heuristic and gradient-guided variants, including a white-box method that identifies influential insertion points. On TREC-DL 2019/2020 with BERT and monoT5 re-rankers, our single-word attacks achieve up to 91% success while modifying fewer than two tokens per document on average, achieving competitive rank and score boosts with far fewer edits under a comparable white-box setup to ensure fair evaluation against PRADA. We also introduce new diagnostic metrics to analyze attack sensitivity beyond aggregate success rates. Our analysis reveals a Goldilocks zone in which mid-ranked documents are most vulnerable. These findings demonstrate practical risks and motivate future defenses for robust neural ranking.
Authors: Jiahuan Yu, Mingtao Hu, Zichao Lin, Minjia Zhang
Abstract: Large Language Model (LLM) serving faces a fundamental tension between stringent latency Service Level Objectives (SLOs) and limited GPU memory capacity. When high request rates exhaust the KV cache budget, existing LLM inference systems often suffer severe head-of-line (HOL) blocking. While prior work explored PCIe-based offloading, these approaches cannot sustain responsiveness under high request rates, often failing to meet tight Time-To-First-Token (TTFT) and Time-Between-Tokens (TBT) SLOs. We present SuperInfer, a high-performance LLM inference system designed for emerging Superchips (e.g., NVIDIA GH200) with tightly coupled GPU-CPU architecture via NVLink-C2C. SuperInfer introduces RotaSched, the first proactive, SLO-aware rotary scheduler that rotates requests to maintain responsiveness on Superchips, and DuplexKV, an optimized rotation engine that enables full-duplex transfer over NVLink-C2C. Evaluations on GH200 using various models and datasets show that SuperInfer improves TTFT SLO attainment rates by up to 74.7% while maintaining comparable TBT and throughput compared to state-of-the-art systems, demonstrating that SLO-aware scheduling and memory co-design unlocks the full potential of Superchips for responsive LLM serving.
Authors: Xin Zhang, Zijin Yang, Kejiang Chen, Linfeng Ma, Weiming Zhang, Nenghai Yu
Abstract: Latent-based watermarks, integrated into the generation process of latent diffusion models (LDMs), simplify detection and attribution of generated images. However, recent black-box forgery attacks, where an attacker needs at least one watermarked image and black-box access to the provider's model, can embed the provider's watermark into images not produced by the provider, posing outsized risk to provenance and trust. We propose SemBind, the first defense framework for latent-based watermarks that resists black-box forgery by binding latent signals to image semantics via a learned semantic masker. Trained with contrastive learning, the masker yields near-invariant codes for the same prompt and near-orthogonal codes across prompts; these codes are reshaped and permuted to modulate the target latent before any standard latent-based watermark. SemBind is generally compatible with existing latent-based watermarking schemes and keeps image quality essentially unchanged, while a simple mask-ratio parameter offers a tunable trade-off between anti-forgery strength and robustness. Across four mainstream latent-based watermark methods, our SemBind-enabled anti-forgery variants markedly reduce false acceptance under black-box forgery while providing a controllable robustness-security balance.
Authors: Kargi Chauhan, Mahalakshmi Venkateswarlu
Abstract: Large Language Models (LLMs) are increasingly deployed for personalized product recommendations, with practitioners commonly assuming that longer user purchase histories lead to better predictions. We challenge this assumption through a systematic benchmark of four state of the art LLMs GPT-4o-mini, DeepSeek-V3, Qwen2.5-72B, and Gemini 2.5 Flash across context lengths ranging from 5 to 50 items using the REGEN dataset. Surprisingly, our experiments with 50 users in a within subject design reveal no significant quality improvement with increased context length. Quality scores remain flat across all conditions (0.17--0.23). Our findings have significant practical implications: practitioners can reduce inference costs by approximately 88\% by using context (5--10 items) instead of longer histories (50 items), without sacrificing recommendation quality. We also analyze latency patterns across providers and find model specific behaviors that inform deployment decisions. This work challenges the existing ``more context is better'' paradigm and provides actionable guidelines for cost effective LLM based recommendation systems.
Authors: Zeyu Xing, Xing Li, Hui-Ling Zhen, Mingxuan Yuan, Sinno Jialin Pan
Abstract: KV caches, typically used only to speed up autoregressive decoding, encode contextual information that can be reused for downstream tasks at no extra cost. We propose treating the KV cache as a lightweight representation, eliminating the need to recompute or store full hidden states. Despite being weaker than dedicated embeddings, KV-derived representations are shown to be sufficient for two key applications: \textbf{(i) Chain-of-Embedding}, where they achieve competitive or superior performance on Llama-3.1-8B-Instruct and Qwen2-7B-Instruct; and \textbf{(ii) Fast/Slow Thinking Switching}, where they enable adaptive reasoning on Qwen3-8B and DeepSeek-R1-Distil-Qwen-14B, reducing token generation by up to $5.7\times$ with minimal accuracy loss. Our findings establish KV caches as a free, effective substrate for sampling and reasoning, opening new directions for representation reuse in LLM inference. Code: https://github.com/cmd2001/ICLR2026_KV-Embedding.
Authors: Zhuang Chen, Dazhen Wan, Zhangkai Zheng, Guanqun Bi, Xiyao Xiao, Binghang Li, Minlie Huang
Abstract: While large language models show promise in mental healthcare, evaluating their therapeutic competence remains challenging due to the unstructured and longitudinal nature of counseling. We argue that current evaluation paradigms suffer from an unanchored defect, leading to two forms of instability: process drift, where unsteered client simulation wanders away from specific counseling goals, and standard drift, where static pointwise scoring lacks the stability for reliable judgment. To address this, we introduce Ps, a unified framework that calibrates the therapeutic competence of LLMs via trajectory-anchored tournaments. We first anchor the interaction trajectory in simulation, where clients precisely control the fluid consultation process to probe multifaceted capabilities. We then anchor the battle trajectory in judgments through an efficient Swiss-system tournament, utilizing dynamic pairwise battles to yield robust Elo ratings. Beyond ranking, we demonstrate that tournament trajectories can be transformed into credible reward signals, enabling on-policy reinforcement learning to enhance LLMs' performance. Extensive experiments validate the effectiveness of PsychePass and its strong consistency with human expert judgments.
Authors: Ali Zia, Usman Ali, Umer Ramzan, Abdul Rehman, Abdelwahed Khamis, Wei Xiang
Abstract: Deep topological data analysis (TDA) offers a principled framework for capturing structural invariants such as connectivity and cycles that persist across scales, making it a natural fit for anomaly segmentation (AS). Unlike thresholdbased binarisation, which produces brittle masks under distribution shift, TDA allows anomalies to be characterised as disruptions to global structure rather than local fluctuations. We introduce TopoOT, a topology-aware optimal transport (OT) framework that integrates multi-filtration persistence diagrams (PDs) with test-time adaptation (TTA). Our key innovation is Optimal Transport Chaining, which sequentially aligns PDs across thresholds and filtrations, yielding geodesic stability scores that identify features consistently preserved across scales. These stabilityaware pseudo-labels supervise a lightweight head trained online with OT-consistency and contrastive objectives, ensuring robust adaptation under domain shift. Across standard 2D and 3D anomaly detection benchmarks, TopoOT achieves state-of-the-art performance, outperforming the most competitive methods by up to +24.1% mean F1 on 2D datasets and +10.2% on 3D AS benchmarks.
Authors: Brian Y. Tsui, Alan Y. Fang, Tiffany J. Hwu
Abstract: Robotic manipulation has increasingly adopted vision-language-action (VLA) models, which achieve strong performance but typically require task-specific demonstrations and fine-tuning, and often generalize poorly under domain shift. We investigate whether general-purpose large language model (LLM) agent frameworks, originally developed for software engineering, can serve as an alternative control paradigm for embodied manipulation. We introduce FAEA (Frontier Agent as Embodied Agent), which applies an LLM agent framework directly to embodied manipulation without modification. Using the same iterative reasoning that enables software agents to debug code, FAEA enables embodied agents to reason through manipulation strategies. We evaluate an unmodified frontier agent, Claude Agent SDK, across the LIBERO, ManiSkill3, and MetaWorld benchmarks. With privileged environment state access, FAEA achieves success rates of 84.9%, 85.7%, and 96%, respectively. This level of task success approaches that of VLA models trained with less than 100 demonstrations per task, without requiring demonstrations or fine-tuning. With one round of human feedback as an optional optimization, performance increases to 88.2% on LIBERO. This demonstration-free capability has immediate practical value: FAEA can autonomously explore novel scenarios in simulation and generate successful trajectories for training data augmentation in embodied learning. Our results indicate that general-purpose agents are sufficient for a class of manipulation tasks dominated by deliberative, task-level planning. This opens a path for robotics systems to leverage actively maintained agent infrastructure and benefit directly from ongoing advances in frontier models. Code is available at https://github.com/robiemusketeer/faea-sim
Authors: Murad Farzulla
Abstract: Cryptocurrency projects articulate value propositions through whitepapers, making claims about functionality and technical capabilities. This study investigates whether these narratives align with observed market behavior. We construct a pipeline combining zero-shot NLP classification (BART-MNLI) with CP tensor decomposition to compare three spaces: (1) a claims matrix from 24 whitepapers across 10 semantic categories, (2) market statistics for 49 assets over two years of hourly data, and (3) latent factors from tensor decomposition (rank 2, 92.45% variance explained). Using Procrustes rotation and Tucker's congruence coefficient, we test alignment across 23 common entities. Results show weak alignment: claims-statistics (phi=0.341, p=0.332), claims-factors (phi=0.077, p=0.747), and statistics-factors (phi=0.197, p<0.001). The statistics-factors significance validates our methodology, confirming the pipeline detects relationships when present. Inter-model validation with DeBERTa-v3 yields 32% exact agreement but 67% top-3 agreement. Cross-sectional analysis reveals heterogeneous contributions: NEAR, MKR, ATOM show positive alignment while ENS, UNI, Bitcoin diverge most. Excluding Bitcoin confirms results are not driven by market dominance. We interpret findings as weak alignment between whitepaper narratives and market factor structure. Limited power (n=23) precludes distinguishing weak from no alignment, but strong alignment (phi>=0.70) can be confidently rejected. Implications for narrative economics and investment analysis are discussed.
Authors: Yangyi Shen, Tianjian Feng, Jiaqi Han, Wen Wang, Tianlang Chen, Chunhua Shen, Jure Leskovec, Stefano Ermon
Abstract: Diffusion Language Models (DLMs) offer order-agnostic generation that can explore many possible decoding trajectories. However, current decoding methods commit to a single trajectory, limiting exploration in trajectory space. We introduce Order-Token Search to explore this space through jointly searching over generation order and token values. Its core is a likelihood estimator that scores denoising actions, enabling stable pruning and efficient exploration of diverse trajectories. Across mathematical reasoning and coding benchmarks, Order-Token Search consistently outperforms baselines on GSM8K, MATH500, Countdown, and HumanEval (3.1%, 3.8%, 7.9%, and 6.8% absolute over backbone), matching or surpassing diffu-GRPO post-trained d1-LLaDA. Our work establishes joint search as a key component for advancing decoding in DLMs.
Authors: Asifullah Khan, Aimen Wadood, Mubashar Iqbal, Umme Zahoora
Abstract: Ransomware has become one of the most serious cybersecurity threats causing major financial losses and operational disruptions worldwide.Traditional detection methods such as static analysis, heuristic scanning and behavioral analysis often fall short when used alone. To address these limitations, this paper presents multimodal multi agent ransomware analysis framework designed for ransomware classification. Proposed multimodal multiagent architecture combines information from static, dynamic and network sources. Each data type is handled by specialized agent that uses auto encoder based feature extraction. These representations are then integrated through a fusion agent. After that fused representation are used by transformer based classifier. It identifies the specific ransomware family. The agents interact through an interagent feedback mechanism that iteratively refines feature representations by suppressing low confidence information. The framework was evaluated on large scale datasets containing thousands of ransomware and benign samples. Multiple experiments were conducted on ransomware dataset. It outperforms single modality and nonadaptive fusion baseline achieving improvement of up to 0.936 in Macro-F1 for family classification and reducing calibration error. Over 100 epochs, the agentic feedback loop displays a stable monotonic convergence leading to over +0.75 absolute improvement in terms of agent quality and a final composite score of around 0.88 without fine tuning of the language models. Zeroday ransomware detection remains family dependent on polymorphism and modality disruptions. Confidence aware abstention enables reliable real world deployment by favoring conservativeand trustworthy decisions over forced classification. The findings indicate that proposed approach provides a practical andeffective path toward improving real world ransomware defense systems.
Authors: Xiao Yang, Yinan Ni, Yuqi Tang, Zhimin Qiu, Chen Wang, Tingzhou Yuan
Abstract: This study addresses the challenge of accurately identifying multi-task contention types in high-dimensional system environments and proposes a unified contention classification framework that integrates representation transformation, structural modeling, and a task decoupling mechanism. The method first constructs system state representations from high-dimensional metric sequences, applies nonlinear transformations to extract cross-dimensional dynamic features, and integrates multiple source information such as resource utilization, scheduling behavior, and task load variations within a shared representation space. It then introduces a graph-based modeling mechanism to capture latent dependencies among metrics, allowing the model to learn competitive propagation patterns and structural interference across resource links. On this basis, task-specific mapping structures are designed to model the differences among contention types and enhance the classifier's ability to distinguish multiple contention patterns. To achieve stable performance, the method employs an adaptive multi-task loss weighting strategy that balances shared feature learning with task-specific feature extraction and generates final contention predictions through a standardized inference process. Experiments conducted on a public system trace dataset demonstrate advantages in accuracy, recall, precision, and F1, and sensitivity analyses on batch size, training sample scale, and metric dimensionality further confirm the model's stability and applicability. The study shows that structured representations and multi-task classification based on high-dimensional metrics can significantly improve contention pattern recognition and offer a reliable technical approach for performance management in complex computing environments.
Authors: Gaku Omiya, Pierre-Louis Poirion, Akiko Takeda
Abstract: Randomized subspace methods reduce per-iteration cost; however, in nonconvex optimization, most analyses are expectation-based, and high-probability bounds remain scarce even under sub-Gaussian noise. We first prove that randomized subspace SGD (RS-SGD) admits a high-probability convergence bound under sub-Gaussian noise, achieving the same order of oracle complexity as prior in-expectation results. Motivated by the prevalence of heavy-tailed gradients in modern machine learning, we then propose randomized subspace normalized SGD (RS-NSGD), which integrates direction normalization into subspace updates. Assuming the noise has bounded $p$-th moments, we establish both in-expectation and high-probability convergence guarantees, and show that RS-NSGD can achieve better oracle complexity than full-dimensional normalized SGD.
Authors: Jiahe Li, Junru Chen, Fanqi Shen, Jialan Yang, Jada Li, Zhizhang Yuan, Baowen Cheng, Meng Li, Yang Yang
Abstract: Enabling natural communication through brain-computer interfaces (BCIs) remains one of the most profound challenges in neuroscience and neurotechnology. While existing frameworks offer partial solutions, they are constrained by oversimplified semantic representations and a lack of interpretability. To overcome these limitations, we introduce Semantic Intent Decoding (SID), a novel framework that translates neural activity into natural language by modeling meaning as a flexible set of compositional semantic units. SID is built on three core principles: semantic compositionality, continuity and expandability of semantic space, and fidelity in reconstruction. We present BrainMosaic, a deep learning architecture implementing SID. BrainMosaic decodes multiple semantic units from EEG/SEEG signals using set matching and then reconstructs coherent sentences through semantic-guided reconstruction. This approach moves beyond traditional pipelines that rely on fixed-class classification or unconstrained generation, enabling a more interpretable and expressive communication paradigm. Extensive experiments on multilingual EEG and clinical SEEG datasets demonstrate that SID and BrainMosaic offer substantial advantages over existing frameworks, paving the way for natural and effective BCI-mediated communication.
Authors: Ofek Aloni, Barak Fishbain
Abstract: Generating dense physical fields from sparse measurements is a fundamental question in sampling, signal processing, and many other applications. State-of-the-art methods either use spatial statistics or rely on examples of dense fields in the training phase, which often are not available, and thus rely on synthetic data. Here, we present a reconstruction method that generates dense fields from sparse measurements, without assuming availability of the spatial statistics, nor of examples of the dense fields. This is made possible through the introduction of an automatically differentiable numerical simulator into the training phase of the method. The method is shown to have superior results over statistical and neural network based methods on a set of three standard problems from fluid mechanics.
Authors: Jianwei Peng (Humboldt-Universit\"at zu Berlin), Stefan Lessmann (Humboldt-Universit\"at zu Berlin, Bucharest University of Economic Studies)
Abstract: Survival analysis has become a standard approach for modelling time to default by time-varying covariates in credit risk. Unlike most existing methods that implicitly assume a stationary data-generating process, in practise, mortgage portfolios are exposed to various forms of data drift caused by changing borrower behaviour, macroeconomic conditions, policy regimes and so on. This study investigates the impact of data drift on survival-based credit risk models and proposes a dynamic joint modelling framework to improve robustness under non-stationary environments. The proposed model integrates a longitudinal behavioural marker derived from balance dynamics with a discrete-time hazard formulation, combined with landmark one-hot encoding and isotonic calibration. Three types of data drift (sudden, incremental and recurring) are simulated and analysed on mortgage loan datasets from Freddie Mac. Experiments and corresponding evidence show that the proposed landmark-based joint model consistently outperforms classical survival models, tree-based drift-adaptive learners and gradient boosting methods in terms of discrimination and calibration across all drift scenarios, which confirms the superiority of our model design.
Authors: Lucas Kook, S{\o}ren Wengel Mogensen
Abstract: Learning the dependence structure among variables in complex systems is a central problem across medical, natural, and social sciences. These structures can be naturally represented by graphs, and the task of inferring such graphs from data is known as graph learning or as causal discovery if the graphs are given a causal interpretation. Existing approaches typically rely on restrictive assumptions about the data-generating process, employ greedy oracle algorithms, or solve approximate formulations of the graph learning problem. As a result, they are either sensitive to violations of central assumptions or fail to guarantee globally optimal solutions. We address these limitations by introducing a nonparametric graph learning framework based on nonparametric conditional independence testing and integer programming. We reformulate the graph learning problem as an integer-programming problem and prove that solving the integer-programming problem provides a globally optimal solution to the original graph learning problem. Our method leverages efficient encodings of graphical separation criteria, enabling the exact recovery of larger graphs than was previously feasible. We provide an implementation in the openly available R package 'glip' which supports learning (acyclic) directed (mixed) graphs and chain graphs. From the resulting output one can compute representations of the corresponding Markov equivalence classes or weak equivalence classes. Empirically, we demonstrate that our approach is faster than other existing exact graph learning procedures for a large fraction of instances and graphs of various sizes. GLIP also achieves state-of-the-art performance on simulated data and benchmark datasets across all aforementioned classes of graphs.
Authors: F. D. Amaro, R. Antonietti, E. Baracchini, L. Benussi, C. Capoccia, M. Caponero, L. G. M. de Carvalho, G. Cavoto, I. A. Costa, A. Croce, M. D'Astolfo, G. D'Imperio, G. Dho, E. Di Marco, J. M. F. dos Santos, D. Fiorina, F. Iacoangeli, Z. Islam, E. Kemp, H. P. Lima Jr, G. Maccarrone, R. D. P. Mano, D. J. G. Marques, G. Mazzitelli, P. Meloni, A. Messina, C. M. B. Monteiro, R. A. Nobrega, G. M. Oppedisano, I. F. Pains, E. Paoletti, F. Petrucci, S. Piacentini, D. Pierluigi, D. Pinci, F. Renga, A. Russo, G. Saviano, P. A. O. C. Silva, N. J. Spooner, R. Tesauro, S. Tomassini, D. Tozzi
Abstract: The CYGNO experiment employs an optical-readout Time Projection Chamber (TPC) to search for rare low-energy interactions using finely resolved scintillation images. While the optical readout provides rich topological information, it produces large, sparse megapixel images that challenge real-time triggering, data reduction, and background discrimination. We summarize two complementary machine-learning approaches developed within CYGNO. First, we present a fast and fully unsupervised strategy for online data reduction based on reconstruction-based anomaly detection. A convolutional autoencoder trained exclusively on pedestal images (i.e. frames acquired with GEM amplification disabled) learns the detector noise morphology and highlights particle-induced structures through localized reconstruction residuals, from which compact Regions of Interest (ROIs) are extracted. On real prototype data, the selected configuration retains (93.0 +/- 0.2)% of reconstructed signal intensity while discarding (97.8 +/- 0.1)% of the image area, with ~25 ms per-frame inference time on a consumer GPU. Second, we report a weakly supervised application of the Classification Without Labels (CWoLa) framework to data acquired with an Americium--Beryllium neutron source. Using only mixed AmBe and standard datasets (no event-level labels), a convolutional classifier learns to identify nuclear-recoil-like topologies. The achieved performance approaches the theoretical limit imposed by the mixture composition and isolates a high-score population with compact, approximately circular morphologies consistent with nuclear recoils.
Authors: Efthymios Costa, Ioanna Papatsouma, Angelos Markos
Abstract: Cluster analysis relates to the task of assigning objects into groups which ideally present some desirable characteristics. When a cluster structure is confined to a subset of the feature space, traditional clustering techniques face unprecedented challenges. We present an information-theoretic framework that overcomes the problems associated with sparse data, allowing for joint feature weighting and clustering. Our proposal constitutes a competitive alternative to existing clustering algorithms for sparse data, as demonstrated through simulations on synthetic data. The effectiveness of our method is established by an application on a real-world genomics data set.
Authors: Samira Yazdanpourmoghadam, Mahan Balal Pour, Vahid Partovi Nia
Abstract: Combinatorial optimization problems such as the Job-Shop Scheduling Problem (JSP) and Knapsack Problem (KP) are fundamental challenges in operations research, logistics, and eterprise resource planning (ERP). These problems often require sophisticated algorithms to achieve near-optimal solutions within practical time constraints. Recent advances in deep learning have introduced transformer-based architectures as promising alternatives to traditional heuristics and metaheuristics. We leverage the Multi-Type Transformer (MTT) architecture to address these benchmarks in a unified framework. We present an extensive experimental evaluation across standard benchmark datasets for JSP and KP, demonstrating that MTT achieves competitive performance on different size of these benchmark problems. We showcase the potential of multi-type attention on a real application in Ferro-Titanium industry. To the best of our knowledge, we are the first to apply multi-type transformers in real manufacturing.
Authors: Claudio Bonanno, Andrea Bulgarelli, Elia Cellini, Alessandro Nada, Dario Panfalone, Davide Vadacchino, Lorenzo Verzichelli
Abstract: As lattice gauge theories with non-trivial topological features are driven towards the continuum limit, standard Markov Chain Monte Carlo simulations suffer for topological freezing, i.e., a dramatic growth of autocorrelations in topological observables. A widely used strategy is the adoption of Open Boundary Conditions (OBC), which restores ergodic sampling of topology but at the price of breaking translation invariance and introducing unphysical boundary artifacts. In this contribution we summarize a scalable, exact flow-based strategy to remove them by transporting configurations from a prior with a OBC defect to a fully periodic ensemble, and apply it to 4d SU(3) Yang--Mills theory. The method is based on a Stochastic Normalizing Flow (SNF) that alternates non-equilibrium Monte Carlo updates with localized, gauge-equivariant defect coupling layers implemented via masked parametric stout smearing. Training is performed by minimizing the average dissipated work, equivalent to a Kullback--Leibler divergence between forward and reverse non-equilibrium path measures, to achieve more reversible trajectories and improved efficiency. We discuss the scaling with the number of degrees of freedom affected by the defect and show that defect SNFs achieve better performances than purely stochastic non-equilibrium methods at comparable cost. Finally, we validate the approach by reproducing reference results for the topological susceptibility.
Authors: Yichi Zhang, Fengqing Zhu
Abstract: Training learned image compression (LIC) models entails navigating a challenging optimization landscape defined by the fundamental trade-off between rate and distortion. Standard first-order optimizers, such as SGD and Adam, struggle with \emph{gradient conflicts} arising from competing objectives, leading to slow convergence and suboptimal rate-distortion performance. In this work, we demonstrate that a simple utilization of a second-order quasi-Newton optimizer, \textbf{SOAP}, dramatically improves both training efficiency and final performance across diverse LICs. Our theoretical and empirical analyses reveal that Newton preconditioning inherently resolves the intra-step and inter-step update conflicts intrinsic to the R-D objective, facilitating faster, more stable convergence. Beyond acceleration, we uncover a critical deployability benefit: second-order trained models exhibit significantly fewer activation and latent outliers. This substantially enhances robustness to post-training quantization. Together, these results establish second-order optimization, achievable as a seamless drop-in replacement of the imported optimizer, as a powerful, practical tool for advancing the efficiency and real-world readiness of LICs.
Authors: Zacharias Komodromos, Kleanthis Malialis, Artemis Kontou, Panayiotis Kolios
Abstract: Accurate forecasting of infectious disease incidence is critical for public health planning and timely intervention. While most data-driven forecasting approaches rely primarily on historical data from a single country, such data are often limited in length and variability, restricting the performance of machine learning (ML) models. In this work, we investigate a cross-country learning approach for infectious disease forecasting, in which a single model is trained on time series data from multiple countries and evaluated on a country of interest. This setting enables the model to exploit shared epidemic dynamics across countries and to benefit from an enlarged training set. We examine this approach through a case study on COVID-19 case forecasting in Cyprus, using surveillance data from European countries. We evaluate multiple ML models and analyse the impact of the lookback window length and cross-country `data augmentation' on multi-step forecasting performance. Our results show that incorporating data from other countries can lead to consistent improvements over models trained solely on national data. Although the empirical focus is on Cyprus and COVID-19, the proposed framework and findings are applicable to infectious disease forecasting more broadly, particularly in settings with limited national historical data.
Authors: Massimo Solinas, Agnes Valenti, Nawaf Bou-Rabee, Roeland Wiersema
Abstract: Scientific computing has long relied on double precision (64-bit floating point) arithmetic to guarantee accuracy in simulations of real-world phenomena. However, the growing availability of hardware accelerators such as Graphics Processing Units (GPUs) has made low-precision formats attractive due to their superior performance, reduced memory footprint, and improved energy efficiency. In this work, we investigate the role of mixed-precision arithmetic in neural-network based Variational Monte Carlo (VMC), a widely used method for solving computationally otherwise intractable quantum many-body systems. We first derive general analytical bounds on the error introduced by reduced precision on Metropolis-Hastings MCMC, and then empirically validate these bounds on the use-case of VMC. We demonstrate that significant portions of the algorithm, in particular, sampling the quantum state, can be executed in half precision without loss of accuracy. More broadly, this work provides a theoretical framework to assess the applicability of mixed-precision arithmetic in machine-learning approaches that rely on MCMC sampling. In the context of VMC, we additionally demonstrate the practical effectiveness of mixed-precision strategies, enabling more scalable and energy-efficient simulations of quantum many-body systems.
Authors: Ethan Shen, Danny Tormoen, Saurabh Shah, Ali Farhadi, Tim Dettmers
Abstract: Open-weight coding agents should hold a fundamental advantage over closed-source systems: they can be specialized to private codebases, encoding repository-specific information directly in their weights. Yet the cost and complexity of training has kept this advantage theoretical. We show it is now practical. We present Soft-Verified Efficient Repository Agents (SERA), an efficient method for training coding agents that enables the rapid and cheap creation of agents specialized to private codebases. Using only supervised finetuning (SFT), SERA achieves state-of-the-art results among fully open-source (open data, method, code) models while matching the performance of frontier open-weight models like Devstral-Small-2. Creating SERA models is 26x cheaper than reinforcement learning and 57x cheaper than previous synthetic data methods to reach equivalent performance. Our method, Soft Verified Generation (SVG), generates thousands of trajectories from a single code repository. Combined with cost-efficiency, this enables specialization to private codebases. Beyond repository specialization, we apply SVG to a larger corpus of codebases, generating over 200,000 synthetic trajectories. We use this dataset to provide detailed analysis of scaling laws, ablations, and confounding factors for training coding agents. Overall, we believe our work will greatly accelerate research on open coding agents and showcase the advantage of open-source models that can specialize to private codebases. We release SERA as the first model in Ai2's Open Coding Agents series, along with all our code, data, and Claude Code integration to support the research community.
Authors: Yiran Huang, Karsten Roth, Quentin Bouniot, Wenjia Xu, Zeynep Akata
Abstract: Transformer-based multimodal large language models often exhibit in-context learning (ICL) abilities. Motivated by this phenomenon, we ask: how do transformers learn to associate information across modalities from in-context examples? We investigate this question through controlled experiments on small transformers trained on synthetic classification tasks, enabling precise manipulation of data statistics and model architecture. We begin by revisiting core principles of unimodal ICL in modern transformers. While several prior findings replicate, we find that Rotary Position Embeddings (RoPE) increases the data complexity threshold for ICL. Extending to the multimodal setting reveals a fundamental learning asymmetry: when pretrained on high-diversity data from a primary modality, surprisingly low data complexity in the secondary modality suffices for multimodal ICL to emerge. Mechanistic analysis shows that both settings rely on an induction-style mechanism that copies labels from matching in-context exemplars; multimodal training refines and extends these circuits across modalities. Our findings provide a mechanistic foundation for understanding multimodal ICL in modern transformers and introduce a controlled testbed for future investigation.
Authors: Shahd Seddik, Fahd Seddik, Iman Saberi, Fatemeh Fard, Minh Hieu Huynh, Patanamon Thongtanunam
Abstract: Large Language Models (LLMs) excel at code generation but struggle with complex problems. Retrieval-Augmented Generation (RAG) mitigates this issue by integrating external knowledge, yet retrieval models often miss relevant context, and generation models hallucinate with irrelevant data. We propose Programming Knowledge Graph (PKG) for semantic representation and fine-grained retrieval of code and text. Our approach enhances retrieval precision through tree pruning and mitigates hallucinations via a re-ranking mechanism that integrates non-RAG solutions. Structuring external data into finer-grained nodes improves retrieval granularity. Evaluations on HumanEval and MBPP show up to 20% pass@1 accuracy gains and a 34% improvement over baselines on MBPP. Our findings demonstrate that our proposed PKG approach along with re-ranker effectively address complex problems while maintaining minimal negative impact on solutions that are already correct without RAG. The replication package is published at https://github.com/iamshahd/ProgrammingKnowledgeGraph
Authors: Yilin Song, Dan M. Kluger, Harsh Parikh, Tian Gu
Abstract: Machine learning predictions are increasingly used to supplement incomplete or costly-to-measure outcomes in fields such as biomedical research, environmental science, and social science. However, treating predictions as ground truth introduces bias while ignoring them wastes valuable information. Prediction-Powered Inference (PPI) offers a principled framework that leverages predictions from large unlabeled datasets to improve statistical efficiency while maintaining valid inference through explicit bias correction using a smaller labeled subset. Despite its potential, the growing PPI variants and the subtle distinctions between them have made it challenging for practitioners to determine when and how to apply these methods responsibly. This paper demystifies PPI by synthesizing its theoretical foundations, methodological extensions, connections to existing statistics literature, and diagnostic tools into a unified practical workflow. Using the Mosaiks housing price data, we show that PPI variants produce tighter confidence intervals than complete-case analysis, but that double-dipping, i.e. reusing training data for inference, leads to anti-conservative confidence intervals and coverages. Under missing-not-at-random mechanisms, all methods, including classical inference using only labeled data, yield biased estimates. We provide a decision flowchart linking assumption violations to appropriate PPI variants, a summary table of selective methods, and practical diagnostic strategies for evaluating core assumptions. By framing PPI as a general recipe rather than a single estimator, this work bridges methodological innovation and applied practice, helping researchers responsibly integrate predictions into valid inference.
Authors: Waldyn G. Martinez
Abstract: Modern industrial and service processes generate high-dimensional, non-Gaussian, and contamination-prone data that challenge the foundational assumptions of classical Statistical Process Control (SPC). Heavy tails, multimodality, nonlinear dependencies, and sparse special-cause observations can distort baseline estimation, mask true anomalies, and prevent reliable identification of an in-control (IC) reference set. To address these challenges, we introduce VSCOUT, a distribution-free framework designed specifically for retrospective (Phase I) monitoring in high-dimensional settings. VSCOUT combines an Automatic Relevance Determination Variational Autoencoder (ARD-VAE) architecture with ensemble-based latent outlier filtering and changepoint detection. The ARD prior isolates the most informative latent dimensions, while the ensemble and changepoint filters identify pointwise and structural contamination within the determined latent space. A second-stage retraining step removes flagged observations and re-estimates the latent structure using only the retained inliers, mitigating masking and stabilizing the IC latent manifold. This two-stage refinement produces a clean and reliable IC baseline suitable for subsequent Phase II deployment. Extensive experiments across benchmark datasets demonstrate that VSCOUT achieves superior sensitivity to special-cause structure while maintaining controlled false alarms, outperforming classical SPC procedures, robust estimators, and modern machine-learning baselines. Its scalability, distributional flexibility, and resilience to complex contamination patterns position VSCOUT as a practical and effective method for retrospective modeling and anomaly detection in AI-enabled environments.
Authors: Andrew Kyle Lampinen, Yuxuan Li, Eghbal Hosseini, Sangnie Bhardwaj, Murray Shanahan
Abstract: Language model representations often contain linear directions that correspond to high-level concepts. Here, we study the dynamics of these representations: how representations evolve along these dimensions within the context of (simulated) conversations. We find that linear representations can change dramatically over a conversation; for example, information that is represented as factual at the beginning of a conversation can be represented as non-factual at the end and vice versa. These changes are content-dependent; while representations of conversation-relevant information may change, generic information is generally preserved. These changes are robust even for dimensions that disentangle factuality from more superficial response patterns, and occur across different model families and layers of the model. These representation changes do not require on-policy conversations; even replaying a conversation script written by an entirely different model can produce similar changes. However, adaptation is much weaker from simply having a sci-fi story in context that is framed more explicitly as such. We also show that steering along a representational direction can have dramatically different effects at different points in a conversation. These results are consistent with the idea that representations may evolve in response to the model playing a particular role that is cued by a conversation. Our findings may pose challenges for interpretability and steering -- in particular, they imply that it may be misleading to use static interpretations of features or directions, or probes that assume a particular range of features consistently corresponds to a particular ground-truth value. However, these types of representational dynamics also point to exciting new research directions for understanding how models adapt to context.
Authors: Adarsh Ganeshram, Haydn Maust, Valentin Duruisseaux, Zongyi Li, Yixuan Wang, Daniel Leibovici, Oscar Bruno, Thomas Hou, Anima Anandkumar
Abstract: The physics-informed neural operator (PINO) is a machine learning paradigm that has demonstrated promising results for learning solutions to partial differential equations (PDEs). It leverages the Fourier Neural Operator to learn solution operators in function spaces and leverages physics losses during training to penalize deviations from known physics laws. Spectral differentiation provides an efficient way to compute derivatives for the physics losses, but it inherently assumes periodicity. When applied to non-periodic functions, this assumption can lead to significant errors, including Gibbs phenomena near domain boundaries which degrade the accuracy of both function representations and derivative computations. To overcome this limitation, we introduce the FC-PINO (Fourier-Continuation-based Physics-Informed Neural Operator) architecture which extends the accuracy and efficiency of PINO and spectral differentiation to non-periodic and non-smooth PDEs. In FC-PINO, we propose integrating Fourier continuation into the PINO framework, and test two different continuation approaches: FC-Legendre and FC-Gram. By transforming non-periodic signals into periodic functions on extended domains in a well-conditioned manner, Fourier continuation enables fast and accurate derivative computations. This approach avoids the discretization sensitivity of finite differences and the memory overhead of automatic differentiation. We demonstrate that standard PINO fails (without padding) or struggles (even with padding) to solve non-periodic and non-smooth PDEs with high precision, across challenging benchmarks. In contrast, the proposed FC-PINO provides accurate, robust, and scalable solutions, substantially outperforming PINO alternatives, and demonstrating that Fourier continuation is critical for extending PINO to a wider range of PDE problems when high-precision solutions are needed.
Authors: Boyu Liu, Lianke Qin, Zhao Song, Yitan Wang, Jiale Zhao
Abstract: Submodular function maximization is a critical building block for diverse tasks, such as document summarization, sensor placement, and image segmentation. Yet its practical utility is often limit by the $O(knd^2)$ computational bottleneck. In this paper, we propose an integrated framework that addresses efficiency and privacy simultaneously. First, we introduce a novel matrix-based computation paradigm that accelerates function evaluations. Second, we develop approximate data structures that further streamline the optimization process, achieving a theoretical complexity of $O(\epsilon^{-2}(nd+kn+kd^2)\log(k/\delta))$. Third, we integrate ($\epsilon, \delta$)-DP guaranties to address the privacy concerns inherent in sensitive optimization tasks.
Authors: Young In Kim, Andrea Agiollo, Pratiksha Agrawal, Johannes O. Royset, Rajiv Khanna
Abstract: Optimization algorithms that seek flatter minima, such as Sharpness-Aware Minimization (SAM), are credited with improved generalization and robustness to noise. We ask whether such gains impact membership privacy. Surprisingly, we find that SAM is more prone to Membership Inference Attacks (MIA) than classical SGD across multiple datasets and attack methods, despite achieving lower test error. This suggests that the geometric mechanism of SAM that improves generalization simultaneously exacerbates membership leakage. We investigate this phenomenon through extensive analysis of memorization and influence scores. Our results reveal that SAM is more capable of capturing atypical subpatterns, leading to higher memorization scores of samples. Conversely, SGD depends more heavily on majority features, exhibiting worse generalization on atypical subgroups and lower memorization. Crucially, this characteristic of SAM can be linked to lower variance in the prediction confidence of unseen samples, thereby amplifying membership signals. Finally, we model SAM under a perfectly interpolating linear regime and theoretically show that sharpness regularization inherently reduces variance, guaranteeing a higher MIA advantage for confidence and likelihood ratio attacks.
Authors: Ali Ebadi, Ebrahim Sahafizadeh
Abstract: This literature review aimed to compare various time-series analysis approaches utilized in forecasting COVID-19 cases in Africa. The study involved a methodical search for English-language research papers published between January 2020 and July 2023, focusing specifically on papers that utilized time-series analysis approaches on COVID-19 datasets in Africa. A variety of databases including PubMed, Google Scholar, Scopus, and Web of Science were utilized for this process. The research papers underwent an evaluation process to extract relevant information regarding the implementation and performance of the time-series analysis models. The study highlighted the different methodologies employed, evaluating their effectiveness and limitations in forecasting the spread of the virus. The result of this review could contribute deeper insights into the field, and future research should consider these insights to improve time series analysis models and explore the integration of different approaches for enhanced public health decision-making.
Authors: Yiming Hu, Xiangxiang Chu, Yong Wang
Abstract: Neural Architecture Search (NAS) relies heavily on labeled data, which is labor-intensive and time-consuming to obtain. In this paper, we propose a novel NAS method based on an unsupervised paradigm, specifically Masked Autoencoders (MAE), thereby eliminating the need for labeled data. By replacing the supervised learning objective with an image reconstruction task, our approach enables the efficient discovery of network architectures without compromising performance and generalization ability. Additionally, we address the problem of performance collapse encountered in the widely-used Differentiable Architecture Search (DARTS) in the unsupervised setting by designing a hierarchical decoder. Extensive experiments across various datasets demonstrate the effectiveness and robustness of our method, offering empirical evidence of its superiority over the counterparts.
Authors: Richard Csaky, Mats W. J. van Es, Oiwi Parker Jones, Mark Woolrich
Abstract: Foundation models trained with self-supervised objectives are increasingly applied to brain recordings, but autoregressive generation of realistic multichannel neural time series remains comparatively underexplored, particularly for Magnetoencephalography (MEG). We study (i) modified multichannel WaveNet variants and (ii) a GPT-2-style Transformer, autoregressively trained by next-step prediction on unlabelled MEG. For the Transformer, we propose a simple quantization/tokenization and embedding scheme (channel, subject, and task-condition embeddings) that repurposes a language-model architecture for continuous, high-rate multichannel time series and enables conditional simulation of task-evoked activity. Across forecasting, long-horizon generation, and downstream decoding, GPT2MEG more faithfully reproduces temporal, spectral, and task-evoked statistics of real MEG than WaveNet variants and linear autoregressive baselines, and scales to multiple subjects via subject embeddings. Code available at https://github.com/ricsinaruto/MEG-transfer-decoding.
Authors: Xulin Chen, Ruipeng Liu, Zhenyu Gan, Garrett E. Katz
Abstract: Uncertainties in transition dynamics pose a critical challenge in reinforcement learning (RL), often resulting in performance degradation of trained policies when deployed on hardware. Many robust RL approaches follow two strategies: enforcing smoothness in actor or actor-critic modules with Lipschitz regularization, or learning robust Bellman operators. However, the first strategy does not investigate the impact of critic-only Lipschitz regularization on policy robustness, while the second lacks comprehensive validation in real-world scenarios. Building on this gap and prior work, we propose PPO-PGDLC, an algorithm based on Proximal Policy Optimization (PPO) that integrates Projected Gradient Descent (PGD) with a Lipschitz-regularized critic (LC). The PGD component calculates the adversarial state within an uncertainty set to approximate the robust Bellman operator, and the Lipschitz-regularized critic further improves the smoothness of learned policies. Experimental results on two classic control tasks and one real-world robotic locomotion task demonstrates that, compared to several baseline algorithms, PPO-PGDLC achieves better performance and predicts smoother actions under environmental perturbations.
Authors: Madison Cooley, Robert M. Kirby, Shandian Zhe, Varun Shankar
Abstract: Physics-informed neural networks (PINNs) have emerged as a powerful approach for solving partial differential equations (PDEs) by training neural networks with loss functions that incorporate physical constraints. In this work, we introduce HyResPINNs, a two-level convex-gated architecture designed to maximize approximation expressiveness for a fixed number of degrees of freedom (DoF). The first level involves a trainable, per-block combination of smooth basis functions with trainable sparsity, and deep neural networks; the second involves the ability to gate entire blocks (much like in ResNets or Highway Nets), allowing for expressivity along the depth dimension of the architecture. Our empirical evaluation on a diverse set of challenging PDE problems demonstrates that HyResPINNs consistently achieve superior accuracy to baseline methods while remaining competitive relative to training times. These results highlight the potential of HyResPINNs to combine desirable features from traditional scientific computing methods and modern machine learning, paving the way for more robust and expressive approaches to physics-informed modeling.
Authors: Piyush Jha, Arnav Arora, Vijay Ganesh
Abstract: We introduce LLMStinger, a novel approach that leverages Large Language Models (LLMs) to automatically generate adversarial suffixes for jailbreak attacks. Unlike traditional methods, which require complex prompt engineering or white-box access, LLMStinger uses a reinforcement learning (RL) loop to fine-tune an attacker LLM, generating new suffixes based on existing attacks for harmful questions from the HarmBench benchmark. Our method significantly outperforms existing red-teaming approaches (we compared against 15 of the latest methods), achieving a +57.2% improvement in Attack Success Rate (ASR) on LLaMA2-7B-chat and a +50.3% ASR increase on Claude 2, both models known for their extensive safety measures. Additionally, we achieved a 94.97% ASR on GPT-3.5 and 99.4% on Gemma-2B-it, demonstrating the robustness and adaptability of LLMStinger across open and closed-source models.
Authors: Martijn Bartelds, Ananjan Nandi, Moussa Koulako Bala Doumbouya, Dan Jurafsky, Tatsunori Hashimoto, Karen Livescu
Abstract: Modern deep learning models often achieve high overall performance, but consistently fail on specific subgroups. Group distributionally robust optimization (group DRO) addresses this problem by minimizing the worst-group loss, but it fails when group losses misrepresent performance differences between groups. This is common in domains like speech, where the widely used connectionist temporal classification (CTC) loss not only scales with input length but also varies with linguistic and acoustic properties, leading to spurious differences between group losses. We present CTC-DRO, which addresses the shortcomings of the group DRO objective by smoothing the group weight update to prevent overemphasis on consistently high-loss groups, while using input length-matched batching to mitigate CTC's scaling issues. We evaluate CTC-DRO on the task of multilingual automatic speech recognition (ASR) across five language sets from the diverse ML-SUPERB 2.0 benchmark. CTC-DRO consistently outperforms group DRO and CTC-based baseline models, reducing the worst-language error by up to 47.1% and the average error by up to 32.9%. CTC-DRO can be applied to ASR with minimal computational costs, and, while motivated by multilingual ASR, offers the potential for reducing group disparities in other domains with similar challenges.
Authors: Annie Marsden, Elad Hazan
Abstract: We study the problem of preconditioning in sequential prediction. From the theoretical lens of linear dynamical systems, we show that convolving the target sequence corresponds to applying a polynomial to the hidden transition matrix. Building on this insight, we propose a universal preconditioning method that convolves the target with coefficients from orthogonal polynomials such as Chebyshev or Legendre. We prove that this approach reduces regret for two distinct prediction algorithms and yields the first ever sublinear and hidden-dimension-independent regret bounds (up to logarithmic factors) that hold for systems with marginally table and asymmetric transition matrices. Finally, extensive synthetic and real-world experiments show that this simple preconditioning strategy improves the performance of a diverse range of algorithms, including recurrent neural networks, and generalizes to signals beyond linear dynamical systems.
Authors: Wenlin Chen, Mingtian Zhang, Jiajun He, Zijing Ou, Jos\'e Miguel Hern\'andez-Lobato, Bernhard Sch\"olkopf, David Barber
Abstract: Score-based distillation methods (e.g., variational score distillation) train one-step diffusion models by first pre-training a teacher score model and then distilling it into a one-step student model. However, the gradient estimator in the distillation stage usually suffers from two sources of bias: (1) biased teacher supervision due to score estimation error incurred during pre-training, and (2) the student model's score estimation error during distillation. These biases can degrade the quality of the resulting one-step diffusion model. To address this, we propose DiffRatio, a new framework for training one-step diffusion models: instead of estimating the teacher and student scores independently and then taking their difference, we directly estimate the score difference as the gradient of a learned log density ratio between the student and data distributions across diffusion time steps. This approach greatly simplifies the training pipeline, significantly reduces gradient estimation bias, and improves one-step generation quality. Additionally, it also reduces auxiliary network size by using a lightweight density-ratio network instead of two full score networks, which improves computational and memory efficiency. DiffRatio achieves competitive one-step generation results on CIFAR-10 and ImageNet (64x64 and 512x512), outperforming most teacher-supervised distillation methods. Moreover, the learned density ratio naturally serves as a verifier, enabling a principled inference-time parallel scaling scheme that further improves the generation quality without external rewards or additional sequential computation.
Authors: Toshinori Kitamura, Arnob Ghosh, Tadashi Kozuno, Wataru Kumagai, Kazumi Kasaura, Kenta Hoshino, Yohei Hosoe, Yutaka Matsuo
Abstract: We study the reinforcement learning (RL) problem in a constrained Markov decision process (CMDP), where an agent explores the environment to maximize the expected cumulative reward while satisfying a single constraint on the expected total utility value in every episode. While this problem is well understood in the tabular setting, theoretical results for function approximation remain scarce. This paper closes the gap by proposing an RL algorithm for linear CMDPs that achieves $\tilde{\mathcal{O}}(\sqrt{K})$ regret with an episode-wise zero-violation guarantee. Furthermore, our method is computationally efficient, scaling polynomially with problem-dependent parameters while remaining independent of the state space size. Our results significantly improve upon recent linear CMDP algorithms, which either violate the constraint or incur exponential computational costs.
Authors: Haonan Yu, Junhao Liu, Xin Zhang
Abstract: Anchors is a popular local model-agnostic explanation technique whose applicability is limited by its computational inefficiency. To address this limitation, we propose a memorization-based framework that accelerates Anchors while preserving explanation fidelity and interpretability. Our approach leverages the iterative nature of Anchors' algorithm which gradually refines an explanation until it is precise enough for a given input by storing and reusing intermediate results obtained during prior explanations. Specifically, we maintain a memory of low-precision, high-coverage rules and introduce a rule transformation framework to adapt them to new inputs: the horizontal transformation adapts a pre-trained explanation to the current input by replacing features, and the vertical transformation refines the general explanation until it is precise enough for the input. We evaluate our method across tabular, text, and image datasets, demonstrating that it significantly reduces explanation generation time while maintaining fidelity and interpretability, thereby enabling the practical adoption of Anchors in time-sensitive applications.
Authors: Yiyang Lu, Mohammad Pedramfar, Vaneet Aggarwal
Abstract: Projection-based algorithms for Constrained Online Convex Optimization (COCO) achieve optimal $\mathcal{O}(T^{1/2})$ regret guarantees but face scalability challenges due to the computational complexity of projections. To circumvent this, projection-free methods utilizing Linear Optimization Oracles (LOO) have been proposed, albeit typically achieving slower $\mathcal{O}(T^{3/4})$ regret rates. In this work, we examine whether the $\mathcal{O}(T^{1/2})$ rate can be recovered in the projection-free setting by strengthening the oracle assumption. We introduce BAGEL, an algorithm utilizing a Separation Oracle (SO) that achieves $\mathcal{O}(T^{1/2})$ regret and $\tilde{\mathcal{O}}(T^{1/2})$ cumulative constraint violation (CCV) for convex cost functions. Our analysis shows that by leveraging an infeasible projection via SO, we can match the time-horizon dependence of projection-based methods with $\tilde{\mathcal{O}}(T)$ oracle calls, provided dependence on the geometry of the action set. This establishes a specific regime where projection-free methods can attain the same convergence rates as projection-based counterparts.
Authors: Gilad Yehudai, Noah Amsel, Joan Bruna
Abstract: It is well understood that different neural network architectures are suited to different tasks, but is there always a single best architecture for a given task? We compare the expressive power of transformers, RNNs, and transformers with chain of thought tokens on a simple and natural class of tasks we term Compositional Reasoning Questions (CRQ). This family captures multi-step problems with tree-like compositional structure, such as evaluating Boolean formulas. We prove that under standard hardness assumptions, \emph{none} of these three architectures is capable of solving CRQs unless some hyperparameter (depth, embedding dimension, and number of chain of thought tokens, respectively) grows with the size of the input. We then provide constructions for solving CRQs with each architecture. For transformers, our construction uses depth that is logarithmic in the problem size. For RNNs, logarithmic embedding dimension is necessary and sufficient, so long as the inputs are provided in a certain order. For transformers with chain of thought, our construction uses $n$ CoT tokens for input size $n$. These results show that, while CRQs are inherently hard, there are several different ways for language models to overcome this hardness. Even for a single class of problems, each architecture has strengths and weaknesses, and none is strictly better than the others.
Authors: Gilad Yehudai, Clayton Sanford, Maya Bechler-Speicher, Orr Fischer, Ran Gilad-Bachrach, Amir Globerson
Abstract: Transformers have revolutionized the field of machine learning. In particular, they can be used to solve complex algorithmic problems, including graph-based tasks. In such algorithmic tasks a key question is what is the minimal size of a transformer that can implement the task. Recent work has begun to explore this problem for graph-based tasks, showing that for sub-linear embedding dimension (i.e., model width) logarithmic depth suffices. However, an open question, which we address here, is what happens if width is allowed to grow linearly, while depth is kept fixed. Here we analyze this setting, and provide the surprising result that with linear width, constant depth suffices for solving a host of graph-based problems. This suggests that a moderate increase in width can allow much shallower models, which are advantageous in terms of inference and train time. For other problems, we show that quadratic width is required. Our results demonstrate the complex and intriguing landscape of transformer implementations of graph-based algorithms. We empirically investigate these trade-offs between the relative powers of depth and width and find tasks where wider models have the same accuracy as deep models, while having much faster train and inference time due to parallelizable hardware.
Authors: Zhaoxiang Shen, Ra\'ul I. Sosa, Jakub Lengiewicz, Alexandre Tkatchenko, St\'ephane P. A. Bordas
Abstract: Accurate prediction of many-body dispersion (MBD) interactions is essential for understanding the van der Waals forces that govern the behavior of many complex molecular systems. However, the high computational cost of MBD calculations limits their direct application in large-scale simulations. In this work, we introduce a machine learning surrogate model specifically designed to predict MBD forces in polymer melts, a system that demands accurate MBD description and offers structural advantages for machine learning approaches. Our model is based on a trimmed SchNet architecture that selectively retains the most relevant atomic connections and incorporates trainable radial basis functions for geometric encoding. We validate our surrogate model on datasets from polyethylene, polypropylene, and polyvinyl chloride melts, demonstrating high predictive accuracy and robust generalization across diverse polymer systems. In addition, the model captures key physical features, such as the characteristic decay behavior of MBD interactions, providing valuable insights for optimizing cutoff strategies. Characterized by high computational efficiency, our surrogate model enables practical incorporation of MBD effects into large-scale molecular simulations.
Authors: J. S. van Hulst, W. P. M. H. Heemels, D. J. Antunes
Abstract: Reinforcement learning (RL) is a powerful framework for decision-making in uncertain environments, but it often requires large amounts of data to learn an optimal policy. We address this challenge by incorporating prior model knowledge to guide exploration and accelerate the learning process. Specifically, we assume access to a model set that contains the true transition kernel and reward function. We optimize over this model set to obtain upper and lower bounds on the Q-function, which are then used to guide the exploration of the agent. We provide theoretical guarantees on the convergence of the Q-function to the optimal Q-function under the proposed class of exploring policies. Furthermore, we also introduce a data-driven regularized version of the model set optimization problem that ensures the convergence of the class of exploring policies to the optimal policy. Lastly, we show that when the model set has a specific structure, namely the bounded-parameter MDP (BMDP) framework, the regularized model set optimization problem becomes convex and simple to implement. In this setting, we also prove finite-time convergence to the optimal policy under mild assumptions. We demonstrate the effectiveness of the proposed exploration strategy, which we call BUMEX (Bounded Uncertainty Model-based Exploration), in a simulation study. The results indicate that the proposed method can significantly accelerate learning in benchmark examples. A toolbox is available at https://github.com/JvHulst/BUMEX.
Authors: Jing Han, Hanting Chen, Kai Han, Xiaomeng Huang, Wenjun Xu, Dacheng Tao, Ping Zhang
Abstract: This position paper argues that the next generation of artificial intelligence in meteorological and climate sciences must transition from fragmented hybrid heuristics toward a unified paradigm of physics-guided multimodal transformers. While purely data-driven models have achieved significant gains in predictive accuracy, they often treat atmospheric processes as mere visual patterns, frequently producing results that lack scientific consistency or violate fundamental physical laws. We contend that current ``hybrid'' attempts to bridge this gap remain ad-hoc and struggle to scale across the heterogeneous nature of meteorological data ranging from satellite imagery to sparse sensor measurements. We argue that the transformer architecture, through its inherent capacity for cross-modal alignment, provides the only viable foundation for a systematic integration of domain knowledge via physical constraint embedding and physics-informed loss functions. By advocating for this unified architectural shift, we aim to steer the community away from ``black-box'' curve fitting and toward AI systems that are inherently falsifiable, scientifically grounded, and robust enough to address the existential challenges of extreme weather and climate change.
Authors: Lawrence Liu, Inesh Chakrabarti, Yixiao Li, Mengdi Wang, Tuo Zhao, Lin F. Yang
Abstract: Large language models (LLMs) exhibit remarkable performance across various natural language processing tasks but suffer from immense computational and memory demands, limiting their deployment in resource-constrained environments. To address this challenge, we propose NoWag (Normalized Weight and Activation Guided Compression), a unified framework for one-shot shape preserving compression algorithms. We apply NoWag to compress Llama-2 (7B, 13B, 70B) and Llama-3 (8B, 70B) models using two popular shape-preserving techniques: vector quantization (NoWag-VQ) and unstructured/semi-structured pruning (NoWag-P). Our results show that NoWag-VQ significantly outperforms state-of-the-art one-shot vector quantization methods, while NoWag-P performs competitively against leading pruning techniques. These findings highlight underlying commonalities between these compression paradigms and suggest promising directions for future research. Our code is available at https://github.com/LawrenceRLiu/NoWag
Authors: Noah Schutte, Grigorii Veviurko, Krzysztof Postek, Neil Yorke-Smith
Abstract: When solving optimization problems under uncertainty with contextual data, utilizing machine learning to predict the uncertain parameters' values is a popular and effective approach. Decision-focused learning (DFL) aims at learning a predictive model such that decision quality, instead of prediction accuracy, is maximized. Common practice is to predict a single scenario representing the uncertain parameters, implicitly assuming that there exists a deterministic problem approximation (proxy) that allows for optimal decision-making. The opposite has also been considered, where the underlying distribution is estimated with a parameterized distribution. However, little is known about when either choice is valid. This paper investigates for the first time problem properties that justify using a certain decision proxy. Using this, we present alternative decision proxies for DFL, with little or no compromise on the complexity of the learning task. We show the effectiveness of presented approaches in experiments on continuous and discrete problems, as well as problems with uncertainty in the objective function and in the constraints.
Authors: Xi He, Yi Miao, Max A. Little
Abstract: This paper introduces the first globally optimal algorithm for the empirical risk minimization problem of two-layer maxout and ReLU networks, i.e., minimizing the number of misclassifications. The algorithm has a worst-case time complexity of $O\left(N^{DK+1}\right)$, where $K$ denotes the number of hidden neurons and $D$ represents the number of features. It can be can be generalized to accommodate arbitrary computable loss functions without affecting its computational complexity. Our experiments demonstrate that the proposed algorithm provides provably exact solutions for small-scale datasets. To handle larger datasets, we introduce a novel coreset selection method that reduces the data size to a manageable scale, making it feasible for our algorithm. This extension enables efficient processing of large-scale datasets and achieves significantly improved performance, with a 20-30\% reduction in misclassifications for both training and prediction, compared to state-of-the-art approaches (neural networks trained using gradient descent and support vector machines), when applied to the same models (two-layer networks with fixed hidden nodes and linear models).
Authors: Leon G\"otz, Marcel Kollovieh, Stephan G\"unnemann, Leo Schwinn
Abstract: Existing time series tokenization methods predominantly encode a constant number of samples into individual tokens. This inflexible approach can generate excessive tokens for even simple patterns like extended constant values, resulting in substantial computational overhead. Inspired by the success of byte pair encoding, we propose the first pattern-centric tokenization scheme for time series analysis. Based on a discrete vocabulary of frequent motifs, our method merges samples with underlying patterns into tokens, compressing time series adaptively. Exploiting our finite set of motifs and the continuous properties of time series, we further introduce conditional decoding as a lightweight yet powerful post-hoc optimization method, which requires no gradient computation and adds no computational overhead. On recent time series foundation models, our motif-based tokenization improves forecasting performance by 36% and boosts efficiency by 1990% on average. Conditional decoding further reduces MSE by up to 44%. In an extensive analysis, we demonstrate the adaptiveness of our tokenization to diverse temporal patterns, its generalization to unseen data, and its meaningful token representations capturing distinct time series properties, including statistical moments and trends.
Authors: Lianfeng Shi, Ao Li, Benjamin Ward-Cherrier
Abstract: Spiking Neural Networks (SNNs) have emerged as a new generation of energy-efficient neural networks suitable for implementation on neuromorphic hardware. As neuromorphic hardware has limited memory and computational resources, parameter pruning and quantization have recently been explored to improve the efficiency of SNNs. State-of-the-art SNN pruning/quantization methods employ multiple compression and training iterations, increasing the cost for pre-trained or very large SNNs. In this paper, we propose a novel one-shot post-training compression framework, Spiking Brain Compression (SBC), that extends the classical Optimal Brain Surgeon method to SNNs. SBC replaces the current-based objective found in the common layer-wise compression method with a spike-train-based objective whose Hessian is cheaply computable, allowing a single backward pass to compress parameters and analytically rescale the rest. Applying SBC to SNN pruning and quantization across event-based and static datasets (up to ImageNet), including SEW-ResNet152 and spike-driven Transformers, we achieve state-of-the-art one-shot post-training compression for SNNs, with single- to double-digit accuracy gains over ANN compression baselines ported to SNNs. We further report a synaptic-operation-based energy proxy and a calibration-size ablation, demonstrating robust performance under sub-one-sample-per-class calibration.
Authors: Shuang Chen, Yue Guo, Zhaochen Su, Yafu Li, Yulun Wu, Jiacheng Chen, Jiayu Chen, Weijie Wang, Xiaoye Qu, Yu Cheng
Abstract: Inspired by the remarkable reasoning capabilities of Deepseek-R1 in complex textual tasks, many works attempt to incentivize similar capabilities in Multimodal Large Language Models (MLLMs) by directly applying reinforcement learning (RL). However, they still struggle to activate complex reasoning. In this paper, rather than examining multimodal RL in isolation, we delve into current training pipelines and identify three crucial phenomena: 1) Effective cold start initialization is critical for enhancing MLLM reasoning. Intriguingly, we find that initializing with carefully selected text data alone can lead to performance surpassing many recent multimodal reasoning models, even before multimodal RL. 2) Standard GRPO applied to multimodal RL suffers from gradient stagnation, which degrades training stability and performance. 3) Subsequent text-only RL training, following the multimodal RL phase, further enhances multimodal reasoning. This staged training approach effectively balances perceptual grounding and cognitive reasoning development. By incorporating the above insights and addressing multimodal RL issues, we introduce ReVisual-R1, achieving a new state-of-the-art among open-source 7B MLLMs on challenging benchmarks including MathVerse, MathVision, WeMath, LogicVista, DynaMath, and challenging AIME2024 and AIME2025.
Authors: Hongzheng Chen, Yingheng Wang, Yaohui Cai, Hins Hu, Jiajie Li, Shirley Huang, Chenhui Deng, Rongjian Liang, Shufeng Kong, Haoxing Ren, Samitha Samaranayake, Carla P. Gomes, Zhiru Zhang
Abstract: While Large Language Models (LLMs) have demonstrated significant advancements in reasoning and agent-based problem-solving, current evaluation methodologies fail to adequately assess their capabilities: existing benchmarks either rely on closed-ended questions prone to saturation and memorization, or subjective comparisons that lack consistency and rigor. In this work, we introduce HeuriGym, an agentic framework designed for evaluating heuristic algorithms generated by LLMs for combinatorial optimization problems, characterized by clearly defined objectives and expansive solution spaces. HeuriGym empowers LLMs to propose heuristics, receive evaluative feedback via code execution, and iteratively refine their solutions. We evaluate nine state-of-the-art models on nine problems across domains such as computer systems, logistics, and biology, exposing persistent limitations in tool use, planning, and adaptive reasoning. To quantify performance, we propose the Quality-Yield Index (QYI), a metric that captures both solution pass rate and quality. Even top models like GPT-o4-mini-high and Gemini-2.5-Pro attain QYI scores of only 0.6, well below the expert baseline of 1. Our open-source benchmark aims to guide the development of LLMs toward more effective and realistic problem-solving in scientific and engineering domains.
Authors: Pol G. Recasens, Alberto Gutierrez, Jordi Torres, Josep. Ll Berral, Javier Carnerero-Cano, Anisa Halimi, Kieran Fraser
Abstract: Large Language Models (LLMs) are increasingly used for synthetic tabular data generation through in-context learning (ICL), offering a practical solution for data augmentation in data scarce scenarios. While prior work has shown the potential of LLMs to improve downstream task performance through augmenting underrepresented groups, these benefits often assume access to a subset of unbiased in-context examples, representative of the real dataset. In real-world settings, however, data is frequently noisy and demographically skewed. In this paper, we systematically study how statistical biases within in-context examples propagate to the distribution of synthetic tabular data, showing that even mild in-context biases lead to global statistical distortions. We further introduce an adversarial scenario where a malicious contributor can inject bias into the synthetic dataset via a subset of in-context examples, ultimately compromising the fairness of downstream classifiers for a targeted and protected subgroup. Finally, we evaluate mitigation strategies based on preprocessing in-context examples, demonstrating that while such interventions can attenuate disparity, the inherent sensitivity of LLMs to adversarial prompts remains a persistent challenge. Our findings highlight a critical new vulnerability in LLM-based data generation pipelines within sensitive domains.
Authors: Antonin Joly, Nicolas Keriven, Aline Roumy
Abstract: Graph coarsening aims to diminish the size of a graph to lighten its memory footprint, and has numerous applications in graph signal processing and machine learning. It is usually defined using a reduction matrix and a lifting matrix, which, respectively, allows to project a graph signal from the original graph to the coarsened one and back. This results in a loss of information measured by the so-called Restricted Spectral Approximation (RSA). Most coarsening frameworks impose a fixed relationship between the reduction and lifting matrices, generally as pseudo-inverses of each other, and seek to define a coarsening that minimizes the RSA. In this paper, we remark that the roles of these two matrices are not entirely symmetric: indeed, putting constraints on the lifting matrix alone ensures the existence of important objects such as the coarsened graph's adjacency matrix or Laplacian. In light of this, in this paper, we introduce a more general notion of reduction matrix, that is not necessarily the pseudo-inverse of the lifting matrix. We establish a taxonomy of ``admissible'' families of reduction matrices, discuss the different properties that they must satisfy and whether they admit a closed-form description or not. We show that, for a fixed coarsening represented by a fixed lifting matrix, the RSA can be further reduced simply by modifying the reduction matrix. We explore different examples, including some based on a constrained optimization process of the RSA. Since this criterion has also been linked to the performance of Graph Neural Networks, we also illustrate the impact of this choices on different node classification tasks on coarsened graphs.
Authors: Tianxiang Zhan, Ming Jin, Yuanpeng He, Yuxuan Liang, Yong Deng, Shirui Pan
Abstract: Recurring concept drift poses a dual challenge in online time series forecasting: mitigating catastrophic forgetting while adhering to strict privacy constraints that prevent retaining historical data. Existing approaches predominantly rely on parameter updates or experience replay, which inevitably suffer from knowledge overwriting or privacy risks. To address this, we propose the Continuous Evolution Pool (CEP), a privacy-preserving framework that maintains a dynamic pool of specialized forecasters. Instead of storing raw samples, CEP utilizes lightweight statistical genes to decouple concept identification from forecasting. Specifically, it employs a Retrieval mechanism to identify the nearest concept based on gene similarity, an Evolution strategy to spawn new forecasters upon detecting distribution shifts, and an Elimination policy to prune obsolete models under memory constraints. Experiments on real-world datasets demonstrate that CEP significantly outperforms state-of-the-art baselines, reducing forecasting error by over 20% without accessing historical ground truth.
Authors: Burak Demirel, Yu Wang, Cristian Tatino, Pablo Soldati
Abstract: Modern RAN operate in highly dynamic and heterogeneous environments, where hand-tuned, rule-based RRM algorithms often underperform. While RL can surpass such heuristics in constrained settings, the diversity of deployments and unpredictable radio conditions introduce major generalization challenges. Data-driven policies frequently overfit to training conditions, degrading performance in unseen scenarios. To address this, we propose a generalization-centered RL framework for RAN control that: (i) robustly reconstructs dynamically varying states from partial and noisy observations, while encoding static and semi-static information, such as radio nodes, cell attributes, and their topology, through graph representations; (ii) applies domain randomization to broaden the training distribution; and (iii) distributes data generation across multiple actors while centralizing training in a cloud-compatible architecture aligned with O-RAN principles. Although generalization increases computational and data-management complexity, our distributed design mitigates this by scaling data collection and training across diverse network conditions. Applied to downlink link adaptation in five 5G benchmarks, our policy improves average throughput and spectral efficiency by ~10% over an OLLA baseline (10% BLER target) in full-buffer MIMO/mMIMO and by >20% under high mobility. It matches specialized RL in full-buffer traffic and achieves up to 4- and 2-fold gains in eMBB and mixed-traffic benchmarks, respectively. In nine-cell deployments, GAT models offer 30% higher throughput over MLP baselines. These results, combined with our scalable architecture, offer a path toward AI-native 6G RAN using a single, generalizable RL agent.
Authors: Jiaqiang Li, Jianbin Tan, Xueqin Wang
Abstract: Equation discovery is a fundamental learning task for uncovering the underlying dynamics of complex systems, with wide-ranging applications in areas such as brain connectivity analysis, climate modeling, gene regulation, and physical simulation. However, many existing approaches rely on accurate derivative estimation and are limited to first-order dynamical systems, restricting their applicability in real-world scenarios. In this work, we propose Sparse Equation Matching (SEM), a unified framework that encompasses several existing equation discovery methods under a common formulation. SEM introduces an integral-based sparse regression approach using Green's functions, enabling derivative-free estimation of differential operators and their associated driving functions in general-order dynamical systems. The effectiveness of SEM is demonstrated through extensive simulations, benchmarking its performance against derivative-based approaches. We then apply SEM to electroencephalographic (EEG) data recorded during multiple oculomotor tasks, collected from 52 participants in a brain-computer interface experiment. Our method identifies active brain regions across participants and reveals task-specific connectivity patterns. These findings offer valuable insights into brain connectivity and the underlying neural mechanisms.
Authors: Hongyu Wang, Jingquan Wang, Dan Negrut
Abstract: Data-driven modeling of constrained multibody dynamics remains challenged by (i) the training cost of Neural ODEs, which typically require backpropagation through an ODE solver, and (ii) error accumulation in rollout predictions. We introduce a Flow-Matching Neural ODE (FNODE) framework that learns the acceleration mapping directly from trajectory data by supervising accelerations rather than integrated states, turning training into a supervised regression problem and eliminating the ODE-adjoint/solver backpropagation bottleneck. Acceleration targets are obtained efficiently via numerical differentiation using a hybrid fast Fourier transform (FFT) and finite-difference (FD) scheme. Kinematic constraints are enforced through coordinate partitioning: FNODE learns accelerations only for the independent generalized coordinates, while the dependent coordinates are recovered by solving the position-level constraint equations. We evaluate FNODE on single and triple mass-spring-damper systems, a double pendulum, a slider crank with and without friction, a vehicle model, and a cart-pole, and compare against MBD-NODE, LSTM, and fully connected baselines. Across these benchmarks, FNODE achieves improved prediction accuracy and training/runtime efficiency, while maintaining constraint satisfaction through the partitioning procedure. Our code and scripts are released as open source to support reproducibility and follow-on research.
Authors: Jian Chen, Jiabao Dou
Abstract: The automatic classification of occupational accident reports is pivotal for workplace safety analysis but is persistently hindered by severe class imbalance and data scarcity. In this paper, we propose ABEX-RAT, a resource-efficient framework that synergizes generative data augmentation with robust adversarial learning. Unlike computationally expensive large language models (LLMs) fine-tuning, our approach employs a two-stage abstractive-expansive (ABEX) pipeline: it first utilizes a prompt-guided LLM to distill label-critical semantics into concise abstracts, which are then expanded into diverse synthetic samples to balance the data distribution. Subsequently, we train a lightweight classifier using a random adversarial training (RAT) protocol, which stochastically injects perturbations to enhance generalization without significant computational overhead. Experimental results on the OSHA dataset demonstrate that ABEXRAT establishes a new state-of-the-art, achieving a Macro-F1 score of 90.32% and significantly outperforming both traditional baselines and fine-tuned large models. This confirms that targeted augmentation combined with robust training offers a superior, data-efficient alternative for specialized domain classification. The source code will be made publicly available upon acceptance.
Authors: Ilyass Moummad, Kawtar Zaher, Lukas Rauch, Alexis Joly
Abstract: Information retrieval with compact binary embeddings, also referred to as hashing, is crucial for scalable fast search applications, yet state-of-the-art hashing methods require expensive, scenario-specific training. In this work, we introduce Hashing-Baseline, a strong training-free hashing method leveraging powerful pretrained encoders that produce rich pretrained embeddings. We revisit classical, training-free hashing techniques: principal component analysis, random orthogonal projection, and threshold binarization, to produce a strong baseline for hashing. Our approach combines these techniques with frozen embeddings from state-of-the-art vision and audio encoders to yield competitive retrieval performance without any additional learning or fine-tuning. To demonstrate the generality and effectiveness of this approach, we evaluate it on standard image retrieval benchmarks as well as a newly introduced benchmark for audio hashing.
Authors: Xinyu He, Xianghui Cao
Abstract: Post-training fundamentally alters the behavior of large language models (LLMs), yet its impact on the internal parameter space remains poorly understood. In this work, we conduct a systematic singular value decomposition (SVD) analysis of principal linear layers in pretrained LLMs, focusing on two widely adopted post-training methods: instruction tuning and long-chain-of-thought (Long-CoT) distillation. Our analysis reveals two unexpected and robust structural changes: (1) a near-uniform geometric scaling of singular values across layers; and (2) highly consistent orthogonal transformations are applied to the left and right singular vectors of each matrix. Based on these findings, We propose a simple yet effective framework to describe the coordinated dynamics of parameters in LLMs, which elucidates why post-training inherently relies on the foundational capabilities developed during pre-training. Further experiments demonstrate that singular value scaling underpins the temperature-controlled regulatory mechanisms of post-training, while the coordinated rotation of singular vectors encodes the essential semantic alignment. These results challenge the prevailing view of the parameter space in large models as a black box, uncovering the first clear regularities in how parameters evolve during training, and providing a new perspective for deeper investigation into model parameter changes.
Authors: Rapha\"el Berthier
Abstract: Diagonal linear networks are neural networks with linear activation and diagonal weight matrices. Their theoretical interest is that their implicit regularization can be rigorously analyzed: from a small initialization, the training of diagonal linear networks converges to the linear predictor with minimal 1-norm among minimizers of the training loss. In this paper, we deepen this analysis showing that the full training trajectory of diagonal linear networks is closely related to the lasso regularization path. In this connection, the training time plays the role of an inverse regularization parameter. Both rigorous results and simulations are provided to illustrate this conclusion. Under a monotonicity assumption on the lasso regularization path, the connection is exact while in the general case, we show an approximate connection.
Authors: Hakaze Cho, Haolin Yang, Gouki Minegishi, Naoya Inoue
Abstract: In-context Learning (ICL) is an emerging few-shot learning paradigm based on modern Language Models (LMs), yet its inner mechanism remains unclear. In this paper, we investigate the mechanism through a novel perspective of information removal. Specifically, we demonstrate that in the zero-shot scenario, LMs encode queries into non-selective representations in hidden states containing information for all possible tasks, leading to arbitrary outputs without focusing on the intended task, resulting in near-zero accuracy. Meanwhile, we find that selectively removing specific information from hidden states by a low-rank filter effectively steers LMs toward the intended task. Building on these findings, by measuring the hidden states on carefully designed metrics, we observe that few-shot ICL effectively simulates such task-oriented information removal processes, selectively removing the redundant information from entangled non-selective representations, and improving the output based on the demonstrations, which constitutes a key mechanism underlying ICL. Moreover, we identify essential attention heads inducing the removal operation, termed Denoising Heads, which enables the ablation experiments blocking the information removal operation from the inference, where the ICL accuracy significantly degrades, especially when the correct label is absent from the few-shot demonstrations, confirming both the critical role of the information removal mechanism and denoising heads.
Authors: Michael Drolet, Firas Al-Hafez, Aditya Bhatt, Jan Peters, Oleg Arenz
Abstract: Discrete latent bottlenecks in variational autoencoders (VAEs) offer high bit efficiency and can be modeled with autoregressive discrete distributions, enabling parameter-efficient multimodal search with transformers. However, discrete random variables do not allow for exact differentiable parameterization; therefore, discrete VAEs typically rely on approximations, such as Gumbel-Softmax reparameterization or straight-through gradient estimates, or employ high-variance gradient-free methods such as REINFORCE that have had limited success on high-dimensional tasks such as image reconstruction. Inspired by popular techniques in policy search, we propose a training framework for discrete VAEs that leverages the natural gradient of a non-parametric encoder to update the parametric encoder without requiring reparameterization. Our method, combined with automatic step size adaptation and a transformer-based encoder, scales to challenging datasets such as ImageNet and outperforms both approximate reparameterization methods and quantization-based discrete autoencoders in reconstructing high-dimensional data from compact latent spaces.
Authors: Yixuan Wang, Dan Guralnik, Saiedeh Akbari, Warren Dixon
Abstract: This article initiates the study of a basic question about model pruning. Given a vector $s$ of importance scores assigned to model components, how many of the scored components could be discarded without sacrificing performance? We propose Effective Model Pruning (EMP), which derives the desired sparsity directly from the score distribution using the notion of effective sample size from particle filtering, also known as the inverse Simpson index. Rather than prescribe a pruning criterion, EMP supplies a universal adaptive threshold derived from the distribution of the score $s$ over the model components: EMP maps $s$ to a number $N_{eff}=N_{eff}(s)$, called the effective sample size. The $N-N_{eff}$ lowest scoring components are discarded. A tight lower bound on the preserved mass fraction $s_{eff}$ (the sum of retained normalized scores) in terms of $N_{eff}$ is derived. This process yields models with a provable upper bound on the loss change relative to the original dense model. Numerical experiments are performed demonstrating this phenomenon across a variety of network architectures including MLPs, CNNs, Transformers, LLMs, and KAN. It is also shown that EMP addresses a rich set of pruning criteria such as weight magnitude, attention score, KAN importance score, and even feature-level signals such as image pixels.
Authors: Silvia Sapora, Devon Hjelm, Alexander Toshev, Omar Attia, Bogdan Mazoure
Abstract: Inverse Reinforcement Learning aims to recover reward models from expert demonstrations, but traditional methods yield black-box models that are difficult to interpret and debug. In this work, we introduce GRACE (Generating Rewards As CodE), a method for using Large Language Models within an evolutionary search to reverse-engineer an interpretable, code-based reward function directly from expert trajectories. The resulting reward function is executable code that can be inspected and verified. We empirically validate GRACE on the MuJoCo, BabyAI and AndroidWorld benchmarks, where it efficiently learns highly accurate rewards, even in complex, multi-task settings. Further, we demonstrate that the resulting reward leads to strong policies, compared to both competitive Imitation Learning and online RL approaches with ground-truth rewards. Finally, we show that GRACE is able to build complex reward APIs in multi-task setups.
Authors: Kurt Butler, Guanchao Feng, Petar Djuric
Abstract: Feature attributions are post-training analysis methods that assess how various input features of a machine learning model contribute to an output prediction. Their interpretation is straightforward when features act independently, but it becomes less clear when the predictive model involves interactions, such as multiplicative relationships or joint feature contributions. In this work, we propose a general theory of higher-order feature attribution, which we develop on the foundation of Integrated Gradients (IG). This work extends existing frameworks in the literature on explainable AI. When using IG as the method of feature attribution, we discover natural connections to statistics and topological signal processing. We provide several theoretical results that establish the theory, and we validate our theory on a few examples.
Authors: Murtaza Rangwala, Farag Azzedin, Richard O. Sinnott, Rajkumar Buyya
Abstract: Decentralized Federated Learning enables privacy-preserving collaborative training without centralized servers but remains vulnerable to Byzantine attacks. Existing defenses require exchanging high-dimensional model vectors with all neighbors each round, creating prohibitive costs at scale. We propose SketchGuard, which decouples Byzantine filtering from aggregation via sketch-based screening. SketchGuard compresses $d$-dimensional models to $k$-dimensional sketches ($k \ll d$) using Count Sketch, then fetches full models only from accepted neighbors, reducing communication complexity from $O(d|N_i|)$ to $O(k|N_i| + d|S_i|)$, where $|N_i|$ is the neighbor count and $|S_i| \le |N_i|$ is the accepted count. We prove convergence in strongly convex and non-convex settings, showing that approximation errors introduce only a $(1+O(\epsilon))$ factor in the effective threshold. Experiments demonstrate SketchGuard maintains state-of-the-art robustness (mean TER deviation $\leq$0.5 percentage points) while reducing computation by up to 82% and communication by 50-70%.
Authors: Xu Yang, Salvatore Rastelli, Alexander Jung
Abstract: We study federated clustering, where interconnected devices collaboratively cluster the data points of private local datasets. Focusing on hard clustering via the k-means principle, we formulate federated k-means as an instance of generalized total variation minimization (GTVMin). This leads to a federated k-means algorithm in which each device updates its local cluster centroids by solving a regularized k-means problem with a regularizer that enforces consistency between neighbouring devices. The resulting algorithm is privacy-friendly, as only aggregated information is exchanged.
Authors: Florent Delgrange, Raphael Avalos, Willem R\"opke
Abstract: Safe policy improvement (SPI) offers theoretical control over policy updates, yet existing guarantees largely concern offline, tabular reinforcement learning (RL). We study SPI in general online settings, when combined with world model and representation learning. We develop a theoretical framework showing that restricting policy updates to a well-defined neighborhood of the current policy ensures monotonic improvement and convergence. This analysis links transition and reward prediction losses to representation quality, yielding online, "deep" analogues of classical SPI theorems from the offline RL literature. Building on these results, we introduce DeepSPI, a principled on-policy algorithm that couples local transition and reward losses with regularised policy updates. On the ALE-57 benchmark, DeepSPI matches or exceeds strong baselines, including PPO and DeepMDPs, while retaining theoretical guarantees.
Authors: Divyesh Savaliya, Marius E. Yamakou
Abstract: Self-induced stochastic resonance (SISR) is the emergence of coherent oscillations in slow-fast excitable systems driven solely by noise, without external periodic forcing or proximity to a bifurcation. This work presents a physics-informed machine learning framework for modeling and predicting SISR in the stochastic FitzHugh-Nagumo neuron. We embed the governing stochastic differential equations and SISR-asymptotic timescale-matching constraints directly into a Physics-Informed Neural Network (PINN) based on a Noise-Augmented State Predictor architecture. The composite loss integrates data fidelity, dynamical residuals, and barrier-based physical constraints derived from Kramers' escape theory. The trained PINN accurately predicts the dependence of spike-train coherence on noise intensity, excitability, and timescale separation, matching results from direct stochastic simulations with substantial improvements in accuracy and generalization compared with purely data-driven methods, while requiring significantly less computation. The framework provides a data-efficient and interpretable surrogate model for simulating and analyzing noise-induced coherence in multiscale stochastic systems.
Authors: Xinlong Zhao, Tong Jia, Minghua He, Xixuan Yang, Ying Li
Abstract: Log-based anomaly detection is critical for ensuring the stability and reliability of web systems. One of the key problems in this task is the lack of sufficient labeled logs, which limits the rapid deployment in new systems. Existing works usually leverage large-scale labeled logs from a mature web system and a small amount of labeled logs from a new system, using transfer learning to extract and generalize general knowledge across both domains. However, these methods focus solely on the transfer of general knowledge and neglect the disparity and potential mismatch between such knowledge and the proprietary knowledge of target system, thus constraining performance. To address this limitation, we propose FusionLog, a novel zero-label cross-system log-based anomaly detection method that effectively achieves the fusion of general and proprietary knowledge, enabling cross-system generalization without any labeled target logs. Specifically, we first design a training-free router based on semantic similarity that dynamically partitions unlabeled target logs into 'general logs' and 'proprietary logs.' For general logs, FusionLog employs a small model based on system-agnostic representation meta-learning for direct training and inference, inheriting the general anomaly patterns shared between the source and target systems. For proprietary logs, we iteratively generate pseudo-labels and fine-tune the small model using multi-round collaborative knowledge distillation and fusion based on large language model (LLM) and small model (SM) to enhance its capability to recognize anomaly patterns specific to the target system. Experimental results on three public log datasets from different systems show that FusionLog achieves over 90% F1-score under a fully zero-label setting, significantly outperforming state-of-the-art cross-system log-based anomaly detection methods.
Authors: Sara Khosravi, Burak Demirel, Linghui Zhou, Javier Rasines, Pablo Soldati
Abstract: Adopting artificial intelligence (AI) in radio access networks (RANs) presents several challenges, including limited availability of link-level measurements (e.g., CQI reports), stringent real-time processing constraints (e.g., sub-1 ms per TTI), and network heterogeneity (different spectrum bands, cell types, and vendor equipment). A critical yet often overlooked barrier lies in the computational and memory limitations of RAN baseband hardware, particularly in legacy 4th Generation (4G) systems, which typically lack on-chip neural accelerators. As a result, only lightweight AI models (under 1 Mb and sub-100~\mu s inference time) can be effectively deployed, limiting both their performance and applicability. However, achieving strong generalization across diverse network conditions often requires large-scale models with substantial resource demands. To address this trade-off, this paper investigates policy distillation in the context of a reinforcement learning-based link adaptation task. We explore two strategies: single-policy distillation, where a scenario-agnostic teacher model is compressed into one generalized student model; and multi-policy distillation, where multiple scenario-specific teachers are consolidated into a single generalist student. Experimental evaluations in a high-fidelity, 5th Generation (5G)-compliant simulator demonstrate that both strategies produce compact student models that preserve the teachers' generalization capabilities while complying with the computational and memory limitations of existing RAN hardware.
Authors: Daojun Liang, Jing Chen, Xiao Wang, Yinglong Wang, Shuo Li
Abstract: Time-Series (TS) exhibits pronounced non-stationarity. Consequently, most forecasting methods display compromised robustness to concept drift, despite the prevalent application of instance normalization. We tackle this challenge by first analysing concept drift through a bias-variance lens and proving that weighted ensemble reduces variance without increasing bias. These insights motivate DeepBooTS, a novel end-to-end dual-stream residual-decreasing boosting method that progressively reconstructs the intrinsic signal. In our design, each block of a deep model becomes an ensemble of learners with an auxiliary output branch forming a highway to the final prediction. The block-wise outputs correct the residuals of previous blocks, leading to a learning-driven decomposition of both inputs and targets. This method enhances versatility and interpretability while substantially improving robustness to concept drift. Extensive experiments, including those on large-scale datasets, show that the proposed method outperforms existing methods by a large margin, yielding an average performance improvement of 15.8% across various datasets, establishing a new benchmark for TS forecasting.
Authors: Yating Ren, Yikun Ban, Huobin Tan
Abstract: Recently, structure-text contrastive learning has shown promising performance on text-attributed graphs by leveraging the complementary strengths of graph neural networks and language models. However, existing methods typically rely on homophily assumptions in similarity estimation and hard optimization objectives, which limit their applicability to heterophilic graphs. Although existing methods can mitigate heterophily through structural adjustments or neighbor aggregation, they usually treat textual embeddings as static targets, leading to suboptimal alignment. In this work, we identify multi-granular heterophily in text-attributed graphs, including complete heterophily, partial heterophily, and latent homophily, which makes structure-text alignment particularly challenging due to mixed, noisy, and missing semantic correlations. To achieve flexible and bidirectional alignment, we propose GCL-OT, a novel graph contrastive learning framework with optimal transport, equipped with tailored mechanisms for each type of heterophily. Specifically, for partial heterophily, we design a RealSoftMax-based similarity estimator to emphasize key neighbor-word interactions while easing background noise. For complete heterophily, we introduce a prompt-based filter that adaptively excludes irrelevant noise during optimal transport alignment. Furthermore, we incorporate OT-guided soft supervision to uncover potential neighbors with similar semantics, enhancing the learning of latent homophily. Theoretical analysis shows that GCL-OT can improve the mutual information bound and Bayes error guarantees. Extensive experiments on nine benchmarks show that GCL-OT outperforms state-of-the-art methods, demonstrating its effectiveness and robustness.
Authors: Jungyang Park, Suho Kang, Jaewoo Park, Jaehong Kim, Jaewoo Shin, Seonjoon Park, Youngjae Yu
Abstract: Knowledge Tracing (KT) aims to model student's knowledge state and predict future performance to enable personalized learning in Intelligent Tutoring Systems. However, traditional KT methods face fundamental limitations in explainability, as they rely solely on the response correctness, neglecting the rich information embedded in students' problem-solving processes. To address this gap, we propose Knowledge Tracing Leveraging Problem-Solving Process (KT-PSP), which incorporates students' problem-solving processes to capture the multidimensional aspects of mathematical proficiency. We also introduce KT-PSP-25, a new dataset specifically designed for the KT-PSP. Building on this, we present StatusKT, a KT framework that employs a teacher-student-teacher three-stage LLM pipeline to extract students' MP as intermediate signals. In this pipeline, the teacher LLM first extracts problem-specific proficiency indicators, then a student LLM generates responses based on the student's solution process, and a teacher LLM evaluates these responses to determine mastery of each indicator. The experimental results on KT-PSP-25 demonstrate that StatusKT improves the prediction performance of existing KT methods. Moreover, StatusKT provides interpretable explanations for its predictions by explicitly modeling students' mathematical proficiency.
Authors: Jucheng Shen, Gaurav Sarkar, Yeonju Ro, Sharath Nittur Sridhar, Zhangyang Wang, Aditya Akella, Souvik Kundu
Abstract: We present CadLLM, a training-free method to accelerate the inference throughput of diffusion-based LLMs (dLLMs). We first investigate the dynamic nature of token unmasking confidence across blocks and steps. Based on this observation, we present a lightweight adaptive approach that controls the generation block size, step size, and threshold based on the average confidence of unmasked tokens. We further reduce softmax overhead by dynamically leveraging a subset of the vocabulary to regulate sampling breadth. CadLLM is a plug-and-play, model-agnostic method compatible with KV-cache-based dLLMs. Extensive experiments on four popular tasks demonstrate that CadLLM yields up to 2.28x throughput improvement over the state-of-the-art baseline with competitive accuracy.
Authors: Chenxiao Gao, Haotian Sun, Na Li, Dale Schuurmans, Bo Dai
Abstract: In real-world applications with large state and action spaces, reinforcement learning (RL) typically employs function approximations to represent core components like the policies, value functions, and dynamics models. Although powerful approximations such as neural networks offer great expressiveness, they often present theoretical ambiguities, suffer from optimization instability and exploration difficulty, and incur substantial computational costs in practice. In this paper, we introduce the perspective of spectral representations as a solution to address these difficulties in RL. Stemming from the spectral decomposition of the transition operator, this framework yields an effective abstraction of the system dynamics for subsequent policy optimization while also providing a clear theoretical characterization. We reveal how to construct spectral representations for transition operators that possess latent variable structures or energy-based structures, which implies different learning methods to extract spectral representations from data. Notably, each of these learning methods realizes an effective RL algorithm under this framework. We also provably extend this spectral view to partially observable MDPs. Finally, we validate these algorithms on over 20 challenging tasks from the DeepMind Control Suite, where they achieve performances comparable or superior to current state-of-the-art model-free and model-based baselines.
Authors: Natnael Tilahun Sinshaw, Mengmei He, Tadesse K. Bahiru, Sudhir Kumar Mohapatra
Abstract: Text classification problems, such as gender classification from a blog, have been a well-matured research area that has been well studied using machine learning algorithms. It has several application domains in market analysis, customer recommendation, and recommendation systems. This study presents a comparative analysis of the widely used machine learning algorithms, namely Support Vector Machines (SVM), Naive Bayes (NB), Logistic Regression (LR), AdaBoost, XGBoost, and an SVM variant (SVM_R) with neuro-symbolic AI (NeSy). The paper also explores the effect of text representations such as TF-IDF, the Universal Sentence Encoder (USE), and RoBERTa. Additionally, various feature extraction techniques, including Chi-Square, Mutual Information, and Principal Component Analysis, are explored. Building on these, we introduce a comparative analysis of the machine learning and deep learning approaches in comparison to the NeSy. The experimental results show that the use of the NeSy approach matched strong MLP results despite a limited dataset. Future work on this research will expand the knowledge base, the scope of embedding types, and the hyperparameter configuration to further study the effectiveness of the NeSy approach.
Authors: Yen-Chieh Huang, Pi-Cheng Hsiu, Rui Fang, Ming-Syan Chen
Abstract: Long-context LLM inference is bottlenecked by the quadratic attention complexity and linear KV cache growth. Prior approaches mitigate this via post-hoc selection or eviction but overlook the root inefficiency: indiscriminate writing to memory. In this paper, we formalize KV cache management as a causal system of three primitives: KV Admission, Selection, and Eviction. We instantiate KV Admission via Write-Gated KV (WG-KV), a lightweight mechanism that learns to predict token utility before cache entry. By filtering out low-utility states early to maintain a compact global cache alongside a sliding local cache, WG-KV reduces memory usage by 46-68% and delivers 3.03-3.70x prefill and 1.85-2.56x decode speedups on Llama and Qwen models, while maintaining compatibility with FlashAttention and Paged-KV systems. These results demonstrate that learning what to write is a principled and practical recipe for efficient long-context inference. Code is available at https://github.com/EMCLab-Sinica/WG-KV.
Authors: Hui Sun
Abstract: This work puts forward a novel nonlinear optimal filter namely the Ensemble Schr{\"o}dinger Bridge nonlinear filter. The proposed filter finds marriage of the standard prediction procedure and the diffusion generative modeling for the analysis procedure to realize one filtering step. The designed approach finds no structural model error, and it is derivative free, training free and highly parallizable. Experimental results show that the designed algorithm performs well given highly nonlinear dynamics in (mildly) high dimension up to 40 or above under a chaotic environment. It also shows better performance than classical methods such as the ensemble Kalman filter and the Particle filter in numerous tests given different level of nonlinearity. Future work will focus on extending the proposed approach to practical meteorological applications and establishing a rigorous convergence analysis.
Authors: Jiayun Wu, Jiashuo Liu, Zhiyuan Zeng, Tianyang Zhan, Tianle Cai, Wenhao Huang
Abstract: LLM deployment in critical domains is currently impeded by persistent hallucinations--generating plausible but factually incorrect assertions. While scaling laws drove significant improvements in general capabilities, theoretical frameworks suggest hallucination is not merely stochastic error but a predictable statistical consequence of training objectives prioritizing mimicking data distribution over epistemic honesty. Standard RLVR paradigms, utilizing binary reward signals, inadvertently incentivize models as good test-takers rather than honest communicators, encouraging guessing whenever correctness probability exceeds zero. This paper presents an exhaustive investigation into behavioral calibration, which incentivizes models to stochastically admit uncertainty by abstaining when not confident, aligning model behavior with accuracy. Synthesizing recent advances, we propose and evaluate training interventions optimizing strictly proper scoring rules for models to output a calibrated probability of correctness. Our methods enable models to either abstain from producing a complete response or flag individual claims where uncertainty remains. Utilizing Qwen3-4B-Instruct, empirical analysis reveals behavior-calibrated reinforcement learning allows smaller models to surpass frontier models in uncertainty quantification--a transferable meta-skill decouplable from raw predictive accuracy. Trained on math reasoning tasks, our model's log-scale Accuracy-to-Hallucination Ratio gain (0.806) exceeds GPT-5's (0.207) in a challenging in-domain evaluation (BeyondAIME). Moreover, in cross-domain factual QA (SimpleQA), our 4B LLM achieves zero-shot calibration error on par with frontier models including Grok-4 and Gemini-2.5-Pro, even though its factual accuracy is much lower.
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 1.7$\times$ over EAGLE-3 across diverse models and workloads. We open-source FailFast at https://github.com/ruipeterpan/failfast.
Authors: Naman Agarwal, Siddhartha R. Dalal, Vishal Misra
Abstract: Transformers often appear to perform Bayesian reasoning in context, but verifying this rigorously has been impossible: natural data lack analytic posteriors, and large models conflate reasoning with memorization. We address this by constructing \emph{Bayesian wind tunnels} -- controlled environments where the true posterior is known in closed form and memorization is provably impossible. In these settings, small transformers reproduce Bayesian posteriors with $10^{-3}$-$10^{-4}$ bit accuracy, while capacity-matched MLPs fail by orders of magnitude, establishing a clear architectural separation. Across two tasks -- bijection elimination and Hidden Markov Model (HMM) state tracking -- we find that transformers implement Bayesian inference through a consistent geometric mechanism: residual streams serve as the belief substrate, feed-forward networks perform the posterior update, and attention provides content-addressable routing. Geometric diagnostics reveal orthogonal key bases, progressive query-key alignment, and a low-dimensional value manifold parameterized by posterior entropy. During training this manifold unfurls while attention patterns remain stable, a \emph{frame-precision dissociation} predicted by recent gradient analyses. Taken together, these results demonstrate that hierarchical attention realizes Bayesian inference by geometric design, explaining both the necessity of attention and the failure of flat architectures. Bayesian wind tunnels provide a foundation for mechanistically connecting small, verifiable systems to reasoning phenomena observed in large language models.
Authors: Dakuan Lu, Jiaqi Zhang, Cheng Yuan, Jiawei Shao, Xuelong Li
Abstract: Recent advances in large language models (LLMs) have been largely driven by scaling laws for individual models, which predict performance improvements as model parameters and data volume increase. However, the capabilities of any single LLM are inherently bounded. One solution originates from intricate interactions among multiple LLMs, rendering their collective performance surpasses that of any constituent model. Despite the rapid proliferation of multi-model integration techniques such as model routing and post-hoc ensembling, a unifying theoretical framework of performance scaling for multi-model collaboration remains absent. In this work, we propose the Law of Multi-model Collaboration, a scaling law that predicts the performance limits of LLM ensembles based on their aggregated parameter budget. To quantify the intrinsic upper bound of multi-model collaboration, we adopt a method-agnostic formulation and assume an idealized integration oracle where the total cross-entropy loss of each sample is determined by the minimum loss of any model in the model pool. Experimental results reveal that multi-model systems follow a power-law scaling with respect to the total parameter count, exhibiting a more significant improvement trend and a lower theoretical loss floor compared to single model scaling. Moreover, ensembles of heterogeneous model families achieve better performance scaling than those formed within a single model family, indicating that model diversity is a primary driver of collaboration gains. These findings suggest that model collaboration represents a critical axis for extending the intelligence frontier of LLMs.
Authors: Naman Agarwal, Siddhartha R. Dalal, Vishal Misra
Abstract: Recent work has shown that small transformers trained in controlled "wind-tunnel'' settings can implement exact Bayesian inference, and that their training dynamics produce a geometric substrate -- low-dimensional value manifolds and progressively orthogonal keys -- that encodes posterior structure. We investigate whether this geometric signature persists in production-grade language models. Across Pythia, Phi-2, Llama-3, and Mistral families, we find that last-layer value representations organize along a single dominant axis whose position strongly correlates with predictive entropy, and that domain-restricted prompts collapse this structure into the same low-dimensional manifolds observed in synthetic settings. To probe the role of this geometry, we perform targeted interventions on the entropy-aligned axis of Pythia-410M during in-context learning. Removing or perturbing this axis selectively disrupts the local uncertainty geometry, whereas matched random-axis interventions leave it intact. However, these single-layer manipulations do not produce proportionally specific degradation in Bayesian-like behavior, indicating that the geometry is a privileged readout of uncertainty rather than a singular computational bottleneck. Taken together, our results show that modern language models preserve the geometric substrate that enables Bayesian inference in wind tunnels, and organize their approximate Bayesian updates along this substrate.
Authors: Yang Cao, Sikun Yang, Xuyun Zhang, Yujiu Yang
Abstract: Anomaly detection aims to identify data instances that deviate significantly from majority of data, which has been widely used in fraud detection, network security, and industrial quality control. Existing methods struggle with datasets exhibiting varying local densities: distance-based methods miss local anomalies, while density-based approaches require careful parameter selection and incur quadratic time complexity. We observe that local anomalies, though indistinguishable under global analysis, become conspicuous when the data space is decomposed into restricted regions and each region is examined independently. Leveraging this geometric insight, we propose SVEAD (Stochastic Voronoi Ensembles Anomaly Detector), which constructs ensemble random Voronoi diagrams and scores points by normalized cell-relative distances weighted by local scale. The proposed method achieves linear time complexity and constant space complexity. Experiments on 45 datasets demonstrate that SVEAD outperforms 12 state-of-the-art approaches.
Authors: Matthew Landers, Taylor W. Killian, Thomas Hartvigsen, Afsaneh Doryab
Abstract: Reinforcement learning in discrete combinatorial action spaces requires searching over exponentially many joint actions to simultaneously select multiple sub-actions that form coherent combinations. Existing approaches either simplify policy learning by assuming independence across sub-actions, which often yields incoherent or invalid actions, or attempt to learn action structure and control jointly, which is slow and unstable. We introduce Structured Policy Initialization (SPIN), a two-stage framework that first pre-trains an Action Structure Model (ASM) to capture the manifold of valid actions, then freezes this representation and trains lightweight policy heads for control. On challenging discrete DM Control benchmarks, SPIN improves average return by up to 39% over the state of the art while reducing time to convergence by up to 12.8$\times$.
Authors: Fardin Ganjkhanloo, Emmett Springer, Erik H. Hoyer, Daniel L. Young, Holley Farley, Kimia Ghobadi
Abstract: In this study, we aim to better align fall risk prediction from the Johns Hopkins Fall Risk Assessment Tool (JHFRAT) with additional clinically meaningful measures via a data-driven modelling approach. We conducted a retrospective cohort analysis of 54,209 inpatient admissions from three Johns Hopkins Health System hospitals between March 2022 and October 2023. A total of 20,208 admissions were included as high fall risk encounters, and 13,941 were included as low fall risk encounters. To incorporate clinical knowledge and maintain interpretability, we employed constrained score optimization (CSO) models to reweight the JHFRAT scoring weights, while preserving its additive structure and clinical thresholds. Recalibration refers to adjusting item weights so that the resulting score can order encounters more consistently by the study's risk labels, and without changing the tool's form factor or deployment workflow. The model demonstrated significant improvements in predictive performance over the current JHFRAT (CSO AUC-ROC=0.91, JHFRAT AUC-ROC=0.86). This performance improvement translates to protecting an additional 35 high-risk patients per week across the Johns Hopkins Health System. The constrained score optimization models performed similarly with and without the EHR variables. Although the benchmark black-box model (XGBoost), improves upon the performance metrics of the knowledge-based constrained logistic regression (AUC-ROC=0.94), the CSO demonstrates more robustness to variations in risk labeling. This evidence-based approach provides a robust foundation for health systems to systematically enhance inpatient fall prevention protocols and patient safety using data-driven optimization techniques, contributing to improved risk assessment and resource allocation in healthcare settings.
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, yet standard architectures like CNNs and DeepSets struggle to capture the irregular geometries and relational physics inherent to domains like seismology. To address this, we propose SourceNet, a Transformer-based framework that bridges the profound Sim-to-Real gap via Physics-Structured Domain Randomization (PSDR), a protocol that randomizes governing physical dynamics to enforce invariance to unmodeled environmental heterogeneity. By pre-training on 100,000 synthetic events and fine-tuning on ~2,500 real-world events, SourceNet achieves state-of-the-art precision on held-out real data, demonstrating exceptional data efficiency and real-time capability compared to classical solvers. Beyond prediction, 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: Yuan Cheng, Fengzhuo Zhang, Yunlong Hou, Cunxiao Du, Chao Du, Tianyu Pang, Aixin Sun, Zhuoran Yang
Abstract: Large Language Models (LLMs) often exhibit slash attention patterns, where attention scores concentrate along the $\Delta$-th sub-diagonal for some offset $\Delta$. These patterns play a key role in passing information across tokens. But why do they emerge? In this paper, we demystify the emergence of these Slash-Dominant Heads (SDHs) from both empirical and theoretical perspectives. First, by analyzing open-source LLMs, we find that SDHs are intrinsic to models and generalize to out-of-distribution prompts. To explain the intrinsic emergence, we analyze the queries, keys, and Rotary Position Embedding (RoPE), which jointly determine attention scores. Our empirical analysis reveals two characteristic conditions of SDHs: (1) Queries and keys are almost rank-one, and (2) RoPE is dominated by medium- and high-frequency components. Under these conditions, queries and keys are nearly identical across tokens, and interactions between medium- and high-frequency components of RoPE give rise to SDHs. Beyond empirical evidence, we theoretically show that these conditions are sufficient to ensure the emergence of SDHs by formalizing them as our modeling assumptions. Particularly, we analyze the training dynamics of a shallow Transformer equipped with RoPE under these conditions, and prove that models trained via gradient descent exhibit SDHs. The SDHs generalize to out-of-distribution prompts.
Authors: Guixian Xu, Jinglai Li, Junqi Tang
Abstract: In this work, we provide a new convergence theory for plug-and-play proximal gradient descent (PnP-PGD) under prior mismatch where the denoiser is trained on a different data distribution to the inference task at hand. To the best of our knowledge, this is the first convergence proof of PnP-PGD under prior mismatch. Compared with the existing theoretical results for PnP algorithms, our new results removed the need for several restrictive and unverifiable assumptions. Moreover, we derive the convergence theory for equivariant PnP (EPnP) under the prior mismatch setting, proving that EPnP reduces error variance and explicitly tightens the convergence bound.
Authors: Kisung You
Abstract: Constant rescaling of a Riemannian metric appears in many computational settings, often through a global scale parameter that is introduced either explicitly or implicitly. Although this operation is elementary, its consequences are not always made clear in practice and may be confused with changes in curvature, manifold structure, or coordinate representation. In this note we provide a short, self-contained account of constant metric scaling on arbitrary Riemannian manifolds. We distinguish between quantities that change under such a scaling, including norms, distances, volume elements, and gradient magnitudes, and geometric objects that remain invariant, such as the Levi--Civita connection, geodesics, exponential and logarithmic maps, and parallel transport. We also discuss implications for Riemannian optimization, where constant metric scaling can often be interpreted as a global rescaling of step sizes rather than a modification of the underlying geometry. The goal of this note is purely expository and is intended to clarify how a global metric scale parameter can be introduced in Riemannian computation without altering the geometric structures on which these methods rely.
Authors: Sina Kazemdehbashi
Abstract: Time series data play a critical role in various fields, including finance, healthcare, marketing, and engineering. A wide range of techniques (from classical statistical models to neural network-based approaches such as Long Short-Term Memory (LSTM)) have been employed to address time series forecasting challenges. In this paper, we reframe time series forecasting as a two-part task: (1) predicting the trend (directional movement) of the time series at the next time step, and (2) forecasting the quantitative value at the next time step. The trend can be predicted using a binary classifier, while quantitative values can be forecasted using models such as LSTM and Bidirectional Long Short-Term Memory (Bi-LSTM). Building on this reframing, we propose the Trend-Adjusted Time Series (TATS) model, which adjusts the forecasted values based on the predicted trend provided by the binary classifier. We validate the proposed approach through both theoretical analysis and empirical evaluation. The TATS model is applied to a volatile financial time series (the daily gold price) with the objective of forecasting the next days price. Experimental results demonstrate that TATS consistently outperforms standard LSTM and Bi-LSTM models by achieving significantly lower forecasting error. In addition, our results indicate that commonly used metrics such as MSE and MAE are insufficient for fully assessing time series model performance. Therefore, we also incorporate trend detection accuracy, which measures how effectively a model captures trends in a time series.
Authors: Bostan Khan, Masoud Daneshtalab
Abstract: Federated Neural Architecture Search (FedNAS) aims to automate model design for privacy-preserving Federated Learning (FL) but currently faces two critical bottlenecks: unguided supernet training that yields suboptimal models, and costly multi-hour pipelines for post-training subnet discovery. We introduce DeepFedNAS, a novel, two-phase framework underpinned by a multi-objective fitness function that synthesizes mathematical network design with architectural heuristics. Enabled by a re-engineered supernet, DeepFedNAS introduces Federated Pareto Optimal Supernet Training, which leverages a pre-computed Pareto-optimal cache of high-fitness architectures as an intelligent curriculum to optimize shared supernet weights. Subsequently, its Predictor-Free Search Method eliminates the need for costly accuracy surrogates by utilizing this fitness function as a direct, zero-cost proxy for accuracy, enabling on-demand subnet discovery in mere seconds. DeepFedNAS achieves state-of-the-art accuracy (e.g., up to 1.21% absolute improvement on CIFAR-100), superior parameter and communication efficiency, and a substantial ~61x speedup in total post-training search pipeline time. By reducing the pipeline from over 20 hours to approximately 20 minutes (including initial cache generation) and enabling 20-second individual subnet searches, DeepFedNAS makes hardware-aware FL deployments instantaneous and practical. The complete source code and experimental scripts are available at: https://github.com/bostankhan6/DeepFedNAS
Authors: Song Xia, Meiwen Ding, Chenqi Kong, Wenhan Yang, Xudong Jiang
Abstract: Multimodal large language models (MLLMs) exhibit strong capabilities across diverse applications, yet remain vulnerable to adversarial perturbations that distort their feature representations and induce erroneous predictions. To address this vulnerability, we propose Feature-space Smoothing (FS), a general framework that provides certified robustness guarantees at the feature representation level of MLLMs. We theoretically prove that FS converts a given feature extractor into a smoothed variant that is guaranteed a certified lower bound on the cosine similarity between clean and adversarial features under $\ell_2$-bounded perturbations. Moreover, we establish that the value of this Feature Cosine Similarity Bound (FCSB) is determined by the intrinsic Gaussian robustness score of the given encoder. Building on this insight, we introduce the Gaussian Smoothness Booster (GSB), a plug-and-play module that enhances the Gaussian robustness score of pretrained MLLMs, thereby strengthening the robustness guaranteed by FS, without requiring additional MLLM retraining. Extensive experiments demonstrate that applying the FS to various MLLMs yields strong certified feature-space robustness and consistently leads to robust task-oriented performance across diverse applications.
Authors: Longteng Zhang, Sen Wu, Shuai Hou, Zhengyu Qing, Zhuo Zheng, Danning Ke, Qihong Lin, Qiang Wang, Shaohuai Shi, Xiaowen Chu
Abstract: Adapting large pre-trained language models to downstream tasks often entails fine-tuning millions of parameters or deploying costly dense weight updates, which hinders their use in resource-constrained environments. Low-rank Adaptation (LoRA) reduces trainable parameters by factorizing weight updates, yet the underlying dense weights still impose high storage and computation costs. Magnitude-based pruning can yield sparse models but typically degrades LoRA's performance when applied naively. In this paper, we introduce SALR (Sparsity-Aware Low-Rank Representation), a novel fine-tuning paradigm that unifies low-rank adaptation with sparse pruning under a rigorous mean-squared-error framework. We prove that statically pruning only the frozen base weights minimizes the pruning error bound, and we recover the discarded residual information via a truncated-SVD low-rank adapter, which provably reduces per-entry MSE by a factor of $(1 - r/\min(d,k))$. To maximize hardware efficiency, we fuse multiple low-rank adapters into a single concatenated GEMM, and we adopt a bitmap-based encoding with a two-stage pipelined decoding + GEMM design to achieve true model compression and speedup. Empirically, SALR attains 50\% sparsity on various LLMs while matching the performance of LoRA on GSM8K and MMLU, reduces model size by $2\times$, and delivers up to a $1.7\times$ inference speedup.
Authors: Nghia Thu Truong, Qui Phu Pham, Quang Nguyen, Dung Luong, Mai Tran
Abstract: Partial Optimal Transport (POT) addresses the problem of transporting only a fraction of the total mass between two distributions, making it suitable when marginals have unequal size or contain outliers. While Sinkhorn-based methods are widely used, their complexity bounds for POT remain suboptimal and can limit scalability. We introduce Accelerated Sinkhorn for POT (ASPOT), which integrates alternating minimization with Nesterov-style acceleration in the POT setting, yielding a complexity of $\mathcal{O}(n^{7/3}\varepsilon^{-5/3})$. We also show that an informed choice of the entropic parameter $\gamma$ improves rates for the classical Sinkhorn method. Experiments on real-world applications validate our theories and demonstrate the favorable performance of our proposed methods.
Authors: Wei Lin, Yining Jiang, Qingyu Song, Qiao Xiang, Hong Xu
Abstract: Zeroth-Order (ZO) optimization has emerged as a promising solution for fine-tuning LLMs under strict memory constraints, as it avoids the prohibitive memory cost of storing activations for backpropagation. However, existing ZO methods typically employ isotropic perturbations, neglecting the rich structural information available during the forward pass. In this paper, we identify a crucial link between gradient formation and activation structure: the gradient of a linear layer is confined to the subspace spanned by its input activations. Leveraging this insight, we propose Activation-Guided Zeroth-Order optimization (AGZO). Unlike prior methods, AGZO extracts a compact, activation-informed subspace on the fly during the forward pass and restricts perturbations to this low-rank subspace. We provide a theoretical framework showing that AGZO optimizes a subspace-smoothed objective and provably yields update directions with higher cosine similarity to the true gradient than isotropic baselines. Empirically, we evaluate AGZO on Qwen3 and Pangu models across various benchmarks. AGZO consistently outperforms state-of-the-art ZO baselines and significantly narrows the performance gap with first-order fine-tuning, while maintaining almost the same peak memory footprint as other ZO methods.
Authors: Daisuke Okanohara
Abstract: Learning systems acquire structured internal representations from data, yet classical information-theoretic results state that deterministic transformations do not increase information. This raises a fundamental question: how can learning produce abstraction and insight without violating information-theoretic limits? We argue that learning is inherently an irreversible process when performed over finite time, and that the realization of epistemic structure necessarily incurs entropy production. To formalize this perspective, we model learning as a transport process in the space of probability distributions over model configurations and introduce an epistemic free-energy framework. Within this framework, we define the free-energy reduction as a bookkeeping quantity that records the total reduction of epistemic free energy along a learning trajectory. This formulation highlights that realizing such a reduction over finite time necessarily incurs irreversible entropy production. We then derive the Epistemic Speed Limit (ESL), a finite-time inequality that lower-bounds the minimal entropy production required by any learning process to realize a given distributional transformation. This bound depends only on the Wasserstein distance between initial and final ensemble distributions and is independent of the specific learning algorithm.
Authors: Yoontae Hwang, Dongwoo Lee, Minseok Choi, Heechan Park, Yong Sup Ihn, Daham Kim, Deok-Young Lee
Abstract: Triad magnetometer components change with sensor attitude even when the IGRF total intensity target stays invariant. NavFormer forecasts this invariant target with rotation invariant scalar features and a Canonical SPD module that stabilizes the spectrum of window level second moments of the triads without sign discontinuities. The module builds a canonical frame from a Gram matrix per window and applies state dependent spectral scaling in the original coordinates. Experiments across five flights show lower error than strong baselines in standard training, few shot training, and zero shot transfer. The code is available at: https://anonymous.4open.science/r/NavFormer-Robust-IGRF-Forecasting-for-Autonomous-Navigators-0765
URLs: https://anonymous.4open.science/r/NavFormer-Robust-IGRF-Forecasting-for-Autonomous-Navigators-0765
Authors: Vincent Gurgul, Ying Chen, Stefan Lessmann
Abstract: This paper presents a Quantum Reinforcement Learning (QRL) solution to the dynamic portfolio optimization problem based on Variational Quantum Circuits. The implemented QRL approaches are quantum analogues of the classical neural-network-based Deep Deterministic Policy Gradient and Deep Q-Network algorithms. Through an empirical evaluation on real-world financial data, we show that our quantum agents achieve risk-adjusted performance comparable to, and in some cases exceeding, that of classical Deep RL models with several orders of magnitude more parameters. However, while quantum circuit execution is inherently fast at the hardware level, practical deployment on cloud-based quantum systems introduces substantial latency, making end-to-end runtime currently dominated by infrastructural overhead and limiting practical applicability. Taken together, our results suggest that QRL is theoretically competitive with state-of-the-art classical reinforcement learning and may become practically advantageous as deployment overheads diminish. This positions QRL as a promising paradigm for dynamic decision-making in complex, high-dimensional, and non-stationary environments such as financial markets. The complete codebase is released as open source at: https://github.com/VincentGurgul/qrl-dpo-public
Authors: Ren Zhuang, Ben Wang, Shuifa Sun
Abstract: Scaling test-time compute enhances long chain-of-thought (CoT) reasoning, yet existing approaches face a fundamental trade-off between computational cost and coverage quality: either incurring high training expense or yielding redundant trajectories. We introduce The Geometric Reasoner (TGR), a training-free framework that performs manifold-informed latent foresight search under strict memory bounds. At each chunk boundary, TGR scores candidate latent anchors via a lightweight look-ahead estimate combined with soft geometric regularizers that encourage smooth trajectories and diverse exploration. Chunk-wise KV cache resets keep memory linear in chunk length. On challenging math and code benchmarks, TGR improves robust trajectory coverage, measured by the area under the Pass@$k$ curve (AUC), by up to 13 points on Qwen3-8B, with negligible overhead of about 1.1--1.3 times.
Authors: Antanas Zilinskas, Robert N. Shorten, Jakub Marecek
Abstract: Forecasting rare events in multivariate time-series data is challenging due to severe class imbalance, long-range dependencies, and distributional uncertainty. We introduce EVEREST, a transformer-based architecture for probabilistic rare-event forecasting that delivers calibrated predictions and tail-aware risk estimation, with auxiliary interpretability via attention-based signal attribution. EVEREST integrates four components: (i) a learnable attention bottleneck for soft aggregation of temporal dynamics; (ii) an evidential head for estimating aleatoric and epistemic uncertainty via a Normal--Inverse--Gamma distribution; (iii) an extreme-value head that models tail risk using a Generalized Pareto Distribution; and (iv) a lightweight precursor head for early-event detection. These modules are jointly optimized with a composite loss (focal loss, evidential NLL, and a tail-sensitive EVT penalty) and act only at training time; deployment uses a single classification head with no inference overhead (approximately 0.81M parameters). On a decade of space-weather data, EVEREST achieves state-of-the-art True Skill Statistic (TSS) of 0.973/0.970/0.966 at 24/48/72-hour horizons for C-class flares. The model is compact, efficient to train on commodity hardware, and applicable to high-stakes domains such as industrial monitoring, weather, and satellite diagnostics. Limitations include reliance on fixed-length inputs and exclusion of image-based modalities, motivating future extensions to streaming and multimodal forecasting.
Authors: Jinkyu Sung, Myunggeum Jee, Joonseok Lee
Abstract: Link sign prediction on a signed graph is a task to determine whether the relationship represented by an edge is positive or negative. Since the presence of negative edges violates the graph homophily assumption that adjacent nodes are similar, regular graph methods have not been applicable without auxiliary structures to handle them. We aim to directly model the latent statistical dependency among edges with the Gaussian copula and its corresponding correlation matrix, extending CopulaGNN (Ma et al., 2021). However, a naive modeling of edge-edge relations is computationally intractable even for a graph with moderate scale. To address this, we propose to 1) represent the correlation matrix as a Gramian of edge embeddings, significantly reducing the number of parameters, and 2) reformulate the conditional probability distribution to dramatically reduce the inference cost. We theoretically verify scalability of our method by proving its linear convergence. Also, our extensive experiments demonstrate that it achieves significantly faster convergence than baselines, maintaining competitive prediction performance to the state-of-the-art models.
Authors: Zhizheng Jiang, Kang Zhao, Weikai Xu, Xinkui Lin, Wei Liu, Jian Luan, Shuo Shang, Peng Han
Abstract: Large reasoning models (LRMs) aim to solve diverse and complex problems through structured reasoning. Recent advances in group-based policy optimization methods have shown promise in enabling stable advantage estimation without reliance on process-level annotations. However, these methods rely on advantage gaps induced by high-quality samples within the same batch, which makes the training process fragile and inefficient when intra-group advantages collapse under challenging tasks. To address these problems, we propose a reinforcement learning mechanism named \emph{\textbf{R^3}} that along three directions: (1) a \emph{cross-context \underline{\textbf{R}}eplay} strategy that maintains the intra-group advantage by recalling valuable examples from historical trajectories of the same query, (2) an \emph{in-context self-\underline{\textbf{R}}eflection} mechanism enabling models to refine outputs by leveraging past failures, and (3) a \emph{structural entropy \underline{\textbf{R}}anking reward}, which assigns relative rewards to truncated or failed samples by ranking responses based on token-level entropy patterns, capturing both local exploration and global stability. We implement our method on Deepseek-R1-Distill-Qwen-1.5B and train it on the DeepscaleR-40k in the math domain. Experiments demonstrate our method achieves SoTA performance on several math benchmarks, representing significant improvements and fewer reasoning tokens over the base models. Code and model will be released.
Authors: Waris Gill, Ahmad Humayun, Ali Anwar, Muhammad Ali Gulzar
Abstract: Federated Learning (FL) enables collaborative training of Large Language Models (LLMs) across distributed data sources while preserving privacy. However, when federated LLMs are deployed in critical applications, it remains unclear which client(s) contributed to specific generated responses, hindering debugging, malicious client identification, fair reward allocation, and trust verification. We present ProToken, a novel Provenance methodology for Token-level attribution in federated LLMs that addresses client attribution during autoregressive text generation while maintaining FL privacy constraints. ProToken leverages two key insights to enable provenance at each token: (1) transformer architectures concentrate task-specific signals in later blocks, enabling strategic layer selection for computational tractability, and (2) gradient-based relevance weighting filters out irrelevant neural activations, focusing attribution on neurons that directly influence token generation. We evaluate ProToken across 16 configurations spanning four LLM architectures (Gemma, Llama, Qwen, SmolLM) and four domains (medical, financial, mathematical, coding). ProToken achieves 98% average attribution accuracy in correctly localizing responsible client(s), and maintains high accuracy when the number of clients are scaled, validating its practical viability for real-world deployment settings.
Authors: Chris Clifton, Bradley Malin, Anna Oganian, Ramesh Raskar, Vivek Sharma
Abstract: Government agencies collect and manage a wide range of ever-growing datasets. While such data has the potential to support research and evidence-based policy making, there are concerns that the dissemination of such data could infringe upon the privacy of the individuals (or organizations) from whom such data was collected. To appraise the current state of data sharing, as well as learn about opportunities for stimulating such sharing at a faster pace, a virtual workshop was held on May 21st and 26th, 2021, sponsored by the National Science Foundation (NSF) and National Institute of Standards and Technologies (NIST), and the White House Office of Science and Technology Policy (OSTP), where a multinational collection of researchers and practitioners were brought together to discuss their experiences and learn about recently developed technologies for managing privacy while sharing data. The workshop specifically focused on challenges and successes in government data sharing at various levels. The first day focused on successful examples of new technology applied to sharing of public data, including formal privacy techniques, synthetic data, and cryptographic approaches. Day two emphasized brainstorming sessions on some of the challenges and directions to address them.
Authors: Conghan Yue, Zhengwei Peng, Shiyan Du, Zhi Ji, Chuangjian Cai, Le Wan, Dongyu Zhang
Abstract: While many diffusion models perform well when controlling particular aspects such as style, character, and interaction, they struggle with fine-grained control due to dataset limitations and intricate model architecture design. This paper introduces a novel training-free algorithm for fine-grained generation, called Aggregation of Multiple Diffusion Models (AMDM). The algorithm integrates features in the latent data space from multiple diffusion models within the same ecosystem into a specified model, thereby activating particular features and enabling fine-grained control. Experimental results demonstrate that AMDM significantly improves fine-grained control without training, validating its effectiveness. Additionally, it reveals that diffusion models initially focus on features such as position, attributes, and style, with later stages improving generation quality and consistency. AMDM offers a new perspective for tackling the challenges of fine-grained conditional generation in diffusion models. Specifically, it allows us to fully utilize existing or develop new conditional diffusion models that control specific aspects, and then aggregate them using the AMDM algorithm. This eliminates the need for constructing complex datasets, designing intricate model architectures, and incurring high training costs. Code is available at: https://github.com/Hammour-steak/AMDM.
Authors: Bikang Pan, Qun Li, Xiaoying Tang, Wei Huang, Zhen Fang, Feng Liu, Jingya Wang, Jingyi Yu, Ye Shi
Abstract: The emergence of vision-language foundation models, such as CLIP, has revolutionized image-text representation, enabling a broad range of applications via prompt learning. Despite its promise, real-world datasets often contain noisy labels that can degrade prompt learning performance. In this paper, we demonstrate that using mean absolute error (MAE) loss in prompt learning, named PromptMAE, significantly enhances robustness against noisy labels while maintaining high accuracy. Though MAE is straightforward and recognized for its robustness, it is rarely used in noisy-label learning due to its slow convergence and poor performance outside prompt learning scenarios. To elucidate the robustness of PromptMAE, we leverage feature learning theory to show that MAE can suppress the influence of noisy samples, thereby improving the signal-to-noise ratio and enhancing overall robustness. Additionally, we introduce PromptOT, a prompt-based optimal transport data purification method to enhance the robustness further. PromptOT employs text features in vision-language models as prototypes to construct an optimal transportation matrix. This matrix effectively partitions datasets into clean and noisy subsets, allowing for the application of cross-entropy loss to the clean subset and MAE loss to the noisy subset. Our Noise-Label Prompt Learning method, named NLPrompt, offers a simple and efficient approach that leverages the expressive representations and precise alignment capabilities of vision-language models for robust prompt learning. We validate NLPrompt through extensive experiments across various noise settings, demonstrating significant performance improvements.
Authors: Yuki Kadokawa, Hirotaka Tahara, Takamitsu Matsubara
Abstract: In earthwork and construction, excavators often encounter large rocks mixed with various soil conditions, requiring skilled operators. This paper presents a framework for achieving autonomous excavation using reinforcement learning (RL) through a rock excavation simulator. In the simulation, resolution can be defined by the particle size/number in the whole soil space. Fine-resolution simulations closely mimic real-world behavior but demand significant calculation time and challenging sample collection, while coarse-resolution simulations enable faster sample collection but deviate from real-world behavior. To combine the advantages of both resolutions, we explore using policies developed in coarse-resolution simulations for pre-training in fine-resolution simulations. To this end, we propose a novel policy learning framework called Progressive-Resolution Policy Distillation (PRPD), which progressively transfers policies through some middle-resolution simulations with conservative policy transfer to avoid domain gaps that could lead to policy transfer failure. Validation in a rock excavation simulator and nine real-world rock environments demonstrated that PRPD reduced sampling time to less than 1/7 while maintaining task success rates comparable to those achieved through policy learning in a fine-resolution simulation. Additional videos and supplementary results are available on our project page: https://yuki-kadokawa.github.io/prpd/
Authors: Irshad A. Meer, Karl-Ludwig Besser, Mustafa Ozger, Dominic Schupke, H. Vincent Poor, Cicek Cavdar
Abstract: Multi-connectivity involves dynamic cluster formation among distributed access points (APs) and coordinated resource allocation from these APs, highlighting the need for efficient mobility management strategies for users with multi-connectivity. In this paper, we propose a novel mobility management scheme for unmanned aerial vehicles (UAVs) that uses dynamic cluster reconfiguration with energy-efficient power allocation in a wireless interference network. Our objective encompasses meeting stringent reliability demands, minimizing joint power consumption, and reducing the frequency of cluster reconfiguration. To achieve these objectives, we propose a hierarchical multi-agent deep reinforcement learning (H-MADRL) framework, specifically tailored for dynamic clustering and power allocation. The edge cloud connected with a set of APs through low latency optical back-haul links hosts the high-level agent responsible for the optimal clustering policy, while low-level agents reside in the APs and are responsible for the power allocation policy. To further improve the learning efficiency, we propose a novel action-observation transition-driven learning algorithm that allows the low-level agents to use the action space from the high-level agent as part of the local observation space. This allows the lower-level agents to share partial information about the clustering policy and allocate the power more efficiently. The simulation results demonstrate that our proposed distributed algorithm achieves comparable performance to the centralized algorithm. Additionally, it offers better scalability, as the decision time for clustering and power allocation increases by only 10% when doubling the number of APs, compared to a 90% increase observed with the centralized approach.
Authors: Nathan Phelps, Daniel J. Lizotte, Douglas G. Woolford
Abstract: There is a widespread and longstanding belief that machine learning models are biased towards the majority class when learning from imbalanced binary response data, leading them to neglect or ignore the minority class. Motivated by a recent simulation study that found that decision trees can be biased towards the minority class, our paper aims to reconcile the conflict between that study and other published works. First, we critically evaluate past literature on this problem, finding that failing to consider the conditional distribution of the outcome given the predictors has led to incorrect conclusions about the bias in decision trees. We then show that, under specific conditions, decision trees fit to purity are biased towards the minority class, debunking the belief that decision trees are always biased towards the majority class. This bias can be reduced by adjusting the tree-fitting process to include regularization methods like pruning and setting a maximum tree depth, and/or by using post-hoc calibration methods. Our findings have implications on the use of popular tree-based models, such as random forests. Although random forests are often composed of decision trees fit to purity, our work adds to recent literature indicating that this may not be the best approach.
Authors: Jairo Diaz-Rodriguez
Abstract: We introduce k-NLPmeans and k-LLMmeans, text-clustering variants of k-means that periodically replace numeric centroids with textual summaries. The key idea, summary-as-centroid, retains k-means assignments in embedding space while producing human-readable, auditable cluster prototypes. The method is LLM-optional: k-NLPmeans uses lightweight, deterministic summarizers, enabling offline, low-cost, and stable operation; k-LLMmeans is a drop-in upgrade that uses an LLM for summaries under a fixed per-iteration budget whose cost does not grow with dataset size. We also present a mini-batch extension for real-time clustering of streaming text. Across diverse datasets, embedding models, and summarization strategies, our approach consistently outperforms classical baselines and approaches the accuracy of recent LLM-based clustering-without extensive LLM calls. Finally, we provide a case study on sequential text streams and release a StackExchange-derived benchmark for evaluating streaming text clustering.
Authors: Simmaco Di Lillo, Domenico Marinucci, Michele Salvi, Stefano Vigogna
Abstract: We study the geometric properties of random neural networks by investigating the boundary volumes of their excursion sets for different activation functions, as the depth increases. More specifically, we show that, for activations which are not very regular (e.g., the Heaviside step function), the boundary volumes exhibit fractal behavior, with their Hausdorff dimension monotonically increasing with the depth. On the other hand, for activations which are more regular (e.g., ReLU, logistic and $\tanh$), as the depth increases, the expected boundary volumes can either converge to zero, remain constant or diverge exponentially, depending on a single spectral parameter which can be easily computed. Our theoretical results are confirmed in some numerical experiments based on Monte Carlo simulations.
Authors: Katarzyna Filus, Micha{\l} Romaszewski, Mateusz \.Zarski
Abstract: Understanding deep neural network (DNN) behavior requires more than evaluating classification accuracy alone; analyzing errors and their predictability is equally crucial. Current evaluation methodologies lack transparency, particularly in explaining the underlying causes of network misclassifications. To address this, we introduce a novel framework that investigates the relationship between the semantic hierarchy depth perceived by a network and its real-data misclassification patterns. Central to our framework is the Similarity Depth (SD) metric, which quantifies the semantic hierarchy depth perceived by a network along with a method of evaluation of how closely the network's errors align with its internally perceived similarity structure. We also propose a graph-based visualization of model semantic relationships and misperceptions. A key advantage of our approach is that leveraging class templates -- representations derived from classifier layer weights -- is applicable to already trained networks without requiring additional data or experiments. Our approach reveals that deep vision networks encode specific semantic hierarchies and that high semantic depth improves the compliance between perceived class similarities and actual errors.
Authors: Jonathan Ott, Maximilian Stahlke, Tobias Feigl, Bjoern M. Eskofier, Christopher Mutschler
Abstract: Unsupervised representation learning for wireless channel state information (CSI)reduces reliance on labeled data, thereby lowering annotation costs, and often improves performance on downstream tasks. However, state-of-the-art approaches take little or no account of domain-specific knowledge, forcing the model to learn well-known concepts solely from data. We introduce Sparse pretrained Radio Transformer (SpaRTran), a hybrid method based on the concept of compressed sensing for wireless channels. In contrast to existing work, SpaRTran builds around a wireless channel model that constrains the optimization procedure to physically meaningful solutions and induces a strong inductive bias. Compared to the state of the art, SpaRTran cuts positioning error by up to 28% and increases top-1 codebook selection accuracy for beamforming by 26 percentage points. Our results show that capturing the sparse nature of radio propagation as an unsupervised learning objective improves performance for network optimization and radio-localization tasks.
Authors: Matthew Lisondra, Beno Benhabib, Goldie Nejat
Abstract: Rapid advancements in foundation models, including Large Language Models, Vision-Language Models, Multimodal Large Language Models, and Vision-Language-Action Models, have opened new avenues for embodied AI in mobile service robotics. By combining foundation models with the principles of embodied AI, where intelligent systems perceive, reason, and act through physical interaction, mobile service robots can achieve more flexible understanding, adaptive behavior, and robust task execution in dynamic real-world environments. Despite this progress, embodied AI for mobile service robots continues to face fundamental challenges related to the translation of natural language instructions into executable robot actions, multimodal perception in human-centered environments, uncertainty estimation for safe decision-making, and computational constraints for real-time onboard deployment. In this paper, we present the first systematic review focused specifically on the integration of foundation models in mobile service robotics. We analyze how recent advances in foundation models address these core challenges through language-conditioned control, multimodal sensor fusion, uncertainty-aware reasoning, and efficient model scaling. We further examine real-world applications in domestic assistance, healthcare, and service automation, highlighting how foundation models enable context-aware, socially responsive, and generalizable robot behaviors. Beyond technical considerations, we discuss ethical, societal, and human-interaction implications associated with deploying foundation model-enabled service robots in human environments. Finally, we outline future research directions emphasizing reliability and lifelong adaptation, privacy-aware and resource-constrained deployment, and governance and human-in-the-loop frameworks required for safe, scalable, and trustworthy mobile service robotics.
Authors: Matthew C. Kelley
Abstract: Forced alignment is a common tool to align audio with orthographic and phonetic transcriptions. Most forced alignment tools provide only a single estimate of a boundary. The present project introduces a method of deriving confidence intervals for these boundaries using a neural network ensemble technique. Ten different segment classifier neural networks were previously trained, and the alignment process is repeated with each model. The alignment ensemble is then used to place the boundary at the median of the boundaries in the ensemble, and 97.85% confidence intervals are constructed using order statistics. Having confidence intervals provides an estimate of the uncertainty in the boundary placement, facilitating tasks like finding boundaries that should be reviewed. As a bonus, on the Buckeye and TIMIT corpora, the ensemble boundaries show a slight overall improvement over using just a single model. The confidence intervals can be emitted during the alignment process as JSON files and a main table for programmatic and statistical analysis. For familiarity, they are also output as Praat TextGrids using a point tier to represent the intervals.
Authors: Shibo Li, Yao Zheng
Abstract: This paper introduces the MPS (Model Prediction Set), a novel framework for online model selection for nonstationary time series. Classical model selection methods, such as information criteria and cross-validation, rely heavily on the stationarity assumption and often fail in dynamic environments which undergo gradual or abrupt changes over time. Yet real-world data are rarely stationary, and model selection under nonstationarity remains a largely open problem. To tackle this challenge, we combine conformal inference with model confidence sets to develop a procedure that adaptively selects models best suited to the evolving dynamics at any given time. Concretely, the MPS updates in real time a confidence set of candidate models that covers the best model for the next time period with a specified long-run probability, while adapting to nonstationarity of unknown forms. Through simulations and real-world data analysis, we demonstrate that MPS reliably and efficiently identifies optimal models under nonstationarity, an essential capability lacking in offline methods. Moreover, MPS frequently produces high-quality sets with small cardinality, whose evolution offers deeper insights into changing dynamics. As a generic framework, MPS accommodates any data-generating process, data structure, model class, training method, and evaluation metric, making it broadly applicable across diverse problem settings.
Authors: Richard Duong, Jannis Chemseddine, Peter K. Friz, Gabriele Steidl
Abstract: We break the mold in flow-based generative modeling by proposing a new model based on the damped wave equation, also known as telegrapher's equation. Similar to the diffusion equation and Brownian motion, there is a Feynman-Kac type relation between the telegrapher's equation and the stochastic Kac process in 1D. The Kac flow evolves stepwise linearly in time, so that the probability flow is Lipschitz continuous in the Wasserstein distance and, in contrast to diffusion flows, the norm of the velocity is globally bounded. Furthermore, the Kac model has the diffusion model as its asymptotic limit. We extend these considerations to a multi-dimensional stochastic process which consists of independent 1D Kac processes in each spatial component. We show that this process gives rise to an absolutely continuous curve in the Wasserstein space and compute the conditional velocity field starting in a Dirac point analytically. Using the framework of flow matching, we train a neural network that approximates the velocity field and use it for sample generation. Our numerical experiments demonstrate the scalability of our approach, and show its advantages over diffusion models.
Authors: Ehsan Pajouheshgar, Yitao Xu, Ali Abbasi, Alexander Mordvintsev, Wenzel Jakob, Sabine S\"usstrunk
Abstract: Neural Cellular Automata (NCAs) are bio-inspired dynamical systems in which identical cells iteratively apply a learned local update rule to self-organize into complex patterns, exhibiting regeneration, robustness, and spontaneous dynamics. Despite their success in texture synthesis and morphogenesis, NCAs remain largely confined to low-resolution outputs. This limitation stems from (1) training time and memory requirements that grow quadratically with grid size, (2) the strictly local propagation of information that impedes long-range cell communication, and (3) the heavy compute demands of real-time inference at high resolution. In this work, we overcome this limitation by pairing an NCA that evolves on a coarse grid with a lightweight implicit decoder that maps cell states and local coordinates to appearance attributes, enabling the same model to render outputs at arbitrary resolution. Moreover, because both the decoder and NCA updates are local, inference remains highly parallelizable. To supervise high-resolution outputs efficiently, we introduce task-specific losses for morphogenesis (growth from a seed) and texture synthesis with minimal additional memory and computation overhead. Our experiments across 2D/3D grids and mesh domains demonstrate that our hybrid models produce high-resolution outputs in real-time, and preserve the characteristic self-organizing behavior of NCAs.
Authors: Ruixi Zheng, Wei Zhang, Yijie Li, Xi Zhu, Zhou Lan, Jarrett Rushmore, Yogesh Rathi, Nikos Makris, Lauren J. O'Donnell, Fan Zhang
Abstract: Diffusion MRI (dMRI) tractography is currently the only method for in vivo mapping of the brain's white matter (WM) connections. Tractometry is an advanced tractography analysis technique for along-tract profiling to investigate the morphology and microstructural properties along the fiber tracts. Tractometry has become an essential tool for studying local along-tract differences between different populations (e.g., health vs disease). In this study, we propose a novel atlas-guided fine-scale tractometry method, namely AGFS-Tractometry, that leverages tract spatial information and permutation testing to enhance the along-tract statistical analysis between populations. There are two major contributions in AGFS-Tractometry. First, we create a novel atlas-guided tract profiling template that enables consistent, fine-scale, along-tract parcellation of subject-specific fiber tracts. Second, we propose a novel nonparametric permutation testing group comparison method to enable simultaneous analysis across all along-tract parcels while correcting for multiple comparisons. We perform experimental evaluations on synthetic datasets with known group differences and in vivo real data. We compare AGFS-Tractometry with two state-of-the-art tractometry methods, including Automated Fiber-tract Quantification (AFQ) and BUndle ANalytics (BUAN). Our results show that the proposed AGFS-Tractometry obtains enhanced sensitivity and specificity in detecting local WM differences. In the real data analysis experiments, AGFS-Tractometry can identify more regions with significant differences, which are anatomically consistent with the existing literature. Overall, these demonstrate the ability of AGFS-Tractometry to detect subtle or spatially localized WM group-level differences. The created tract profiling template and related code are available at: https://github.com/ZhengRuixi/AGFS-Tractometry.git.
Authors: Aneesh Rangnekar, Harini Veeraraghavan
Abstract: Accurate detection and segmentation of cancerous lesions from computed tomography (CT) scans is essential for automated treatment planning and cancer treatment response assessment. Transformer-based models with self-supervised pretraining can produce reliably accurate segmentation from in-distribution (ID) data but degrade when applied to out-of-distribution (OOD) datasets. We address this challenge with RF-Deep, a random forest classifier that utilizes deep features from a pretrained transformer encoder of the segmentation model to detect OOD scans and enhance segmentation reliability. The segmentation model comprises a Swin Transformer encoder, pretrained with masked image modeling (SimMIM) on 10,432 unlabeled 3D CT scans covering cancerous and non-cancerous conditions, with a convolution decoder, trained to segment lung cancers in 317 3D scans. Independent testing was performed on 603 3D CT public datasets that included one ID dataset and four OOD datasets comprising chest CTs with pulmonary embolism (PE) and COVID-19, and abdominal CTs with kidney cancers and healthy volunteers. RF-Deep detected OOD cases with a FPR95 of 18.26%, 27.66%, and less than 0.1% on PE, COVID-19, and abdominal CTs, consistently outperforming established OOD approaches. The RF-Deep classifier provides a simple and effective approach to enhance reliability of cancer segmentation in ID and OOD scenarios.
Authors: Daniel Ablin, Alon Cohen
Abstract: Optimization under uncertainty is a fundamental problem in learning and decision-making, particularly in multi-agent systems. Previously, Feldman, Kalai, and Tennenholtz [2010] demonstrated the ability to efficiently compete in repeated symmetric two-player matrix games without observing payoffs, as long as the opponents actions are observed. In this paper, we introduce and formalize a new class of zero-sum symmetric Markov games, which extends the notion of symmetry from matrix games to the Markovian setting. We show that even without observing payoffs, a player who knows the transition dynamics and observes only the opponents sequence of actions can still compete against an adversary who may have complete knowledge of the game. We formalize three distinct notions of symmetry in this setting and show that, under these conditions, the learning problem can be reduced to an instance of online learning, enabling the player to asymptotically match the return of the opponent despite lacking payoff observations. Our algorithms apply to both matrix and Markov games, and run in polynomial time with respect to the size of the game and the number of episodes. Our work broadens the class of games in which robust learning is possible under severe informational disadvantage and deepens the connection between online learning and adversarial game theory.
Authors: Pragun Jaswal, L. Hemanth Krishna, B. Srinivasu
Abstract: Deep Neural Networks (DNNs) require highly efficient matrix multiplication engines for complex computations. This paper presents a systolic array architecture incorporating novel exact and approximate processing elements (PEs), designed using energy-efficient positive partial product and negative partial product cells, termed as PPC and NPPC, respectively. The proposed 8-bit exact and approximate PE designs are employed in a 8x8 systolic array, which achieves a energy savings of 22% and 32%, respectively, compared to the existing design. To demonstrate their effectiveness, the proposed PEs are integrated into a systolic array (SA) for Discrete Cosine Transform (DCT) computation, achieving high output quality with a PSNR of 38.21,dB. Furthermore, in an edge detection application using convolution, the approximate PE achieves a PSNR of 30.45,dB. These results highlight the potential of the proposed design to deliver significant energy efficiency while maintaining competitive output quality, making it well-suited for error-resilient image and vision processing applications.
Authors: Chen Cheng, John Duchi
Abstract: We demonstrate that learning procedures that rely on aggregated labels, e.g., label information distilled from noisy responses, enjoy robustness properties impossible without data cleaning. This robustness appears in several ways. In the context of risk consistency -- when one takes the standard approach in machine learning of minimizing a surrogate (typically convex) loss in place of a desired task loss (such as the zero-one mis-classification error) -- procedures using label aggregation obtain stronger consistency guarantees than those even possible using raw labels. And while classical statistical scenarios of fitting perfectly-specified models suggest that incorporating all possible information -- modeling uncertainty in labels -- is statistically efficient, consistency fails for ``standard'' approaches as soon as a loss to be minimized is even slightly mis-specified. Yet procedures leveraging aggregated information still converge to optimal classifiers, highlighting how incorporating a fuller view of the data analysis pipeline, from collection to model-fitting to prediction time, can yield a more robust methodology by refining noisy signals.
Authors: Hyunjong Ok, Suho Yoo, Hyeonjun Kim, Jaeho Lee
Abstract: Even without directly hearing sounds, humans can effortlessly reason about auditory properties, such as pitch, loudness, or sound-source associations, drawing on auditory commonsense. In contrast, language models often lack this capability, limiting their effectiveness in multimodal interactions. As an initial step to address this gap, we present AuditoryBench++, a comprehensive benchmark for evaluating auditory knowledge and reasoning in text-only settings. The benchmark encompasses tasks that range from basic auditory comparisons to contextually grounded reasoning, enabling fine-grained analysis of how models process and integrate auditory concepts. In addition, we introduce AIR-CoT, a novel auditory imagination reasoning method that generates and integrates auditory information during inference through span detection with special tokens and knowledge injection. Extensive experiments with recent LLMs and Multimodal LLMs demonstrate that AIR-CoT generally outperforms both the off-the-shelf models and those augmented with auditory knowledge. The project page is available at https://auditorybenchpp.github.io.
Authors: Anil Kamber, Rahul Parhi
Abstract: Understanding the geometry of the loss landscape near a minimum is key to explaining the implicit bias of gradient-based methods in non-convex optimization problems such as deep neural network training and deep matrix factorization. A central quantity to characterize this geometry is the maximum eigenvalue of the Hessian of the loss. Currently, its precise role has been obfuscated because no exact expressions for this sharpness measure were known in general settings. In this paper, we present the first exact expression for the maximum eigenvalue of the Hessian of the squared-error loss at any minimizer in deep matrix factorization/deep linear neural network training problems, resolving an open question posed by Mulayoff & Michaeli (2020). This expression reveals a fundamental property of the loss landscape in deep matrix factorization: Having a constant product of the spectral norms of the left and right intermediate factors across layers is a sufficient condition for flatness. Most notably, in both depth-$2$ matrix and deep overparameterized scalar factorization, we show that this condition is both necessary and sufficient for flatness, which implies that flat minima are spectral-norm balanced even though they are not necessarily Frobenius-norm balanced. To complement our theory, we provide the first empirical characterization of an escape phenomenon during gradient-based training near a minimizer of a deep matrix factorization problem.
Authors: David Anugraha, Shou-Yi Hung, Zilu Tang, Annie En-Shiun Lee, Derry Tanti Wijaya, Genta Indra Winata
Abstract: Evaluation using Large Language Model (LLM) judges has been widely adopted in English and shown to be effective for automatic evaluation. However, their performance does not generalize well to non-English settings, and it remains unclear what constitutes effective multilingual training for such judges. In this paper, we introduce mR3, a massively multilingual, rubric-agnostic reward reasoning model trained on 72 languages, achieving the broadest language coverage in reward modeling to date. We present a comprehensive study of data and curriculum selection for training to identify effective strategies and data sources for building high-quality reward models, including support for reasoning in the target language. Our approach attains state-of-the-art performance on multilingual reward model benchmarks, surpassing much larger models (i.e., GPT-OSS-120B) while being up to 9x smaller, and its effectiveness is further confirmed through extensive ablation studies. Finally, we demonstrate the effectiveness of mR3 in off-policy preference optimization and validate the quality of its reasoning traces and rubric-based evaluations through human studies with 20 annotators across 12 languages, where mR3 models' reasoning is preferred, including for extremely low-resource languages that are entirely unseen during training. Our models, data, and code are available as open source at https://github.com/rubricreward/mr3.
Authors: Manish Nagaraj, Sakshi Choudhary, Utkarsh Saxena, Deepak Ravikumar, Kaushik Roy
Abstract: Instruction tuning is essential for aligning large language models (LLMs) to downstream tasks and commonly relies on large, diverse corpora. However, small, high-quality subsets, known as coresets, can deliver comparable or superior results, though curating them remains challenging. Existing methods often rely on coarse, sample-level signals like gradients, an approach that is computationally expensive and overlooks fine-grained features. To address this, we introduce TRIM (Token Relevance via Interpretable Multi-layer Attention), a forward-only, token-centric framework. Instead of using gradients, TRIM operates by matching underlying representational patterns identified via attention-based "fingerprints" from a handful of target samples. Such an approach makes TRIM highly efficient and uniquely sensitive to the structural features that define a task. Coresets selected by our method consistently outperform state-of-the-art baselines by up to 9% on downstream tasks and even surpass the performance of full-data fine-tuning in some settings. By avoiding expensive backward passes, TRIM achieves this at a fraction of the computational cost. These findings establish TRIM as a scalable and efficient alternative for building high-quality instruction-tuning datasets.
Authors: Zhaolin Hu, Kun Li, Hehe Fan, Yi Yang
Abstract: Linear attention mechanisms have emerged as efficient alternatives to full self-attention in Graph Transformers, offering linear time complexity. However, existing linear attention models often suffer from a significant drop in expressiveness due to low-rank projection structures and overly uniform attention distributions. We theoretically prove that these properties reduce the class separability of node representations, limiting the model's classification ability. To address this, we propose a novel hybrid framework that enhances both the rank and focus of attention. Specifically, we enhance linear attention by attaching a gated local graph network branch to the value matrix, thereby increasing the rank of the resulting attention map. Furthermore, to alleviate the excessive smoothing effect inherent in linear attention, we introduce a learnable log-power function into the attention scores to reduce entropy and sharpen focus. We theoretically show that this function decreases entropy in the attention distribution, enhancing the separability of learned embeddings. Extensive experiments on both homophilic and heterophilic graph benchmarks demonstrate that our method achieves competitive performance while preserving the scalability of linear attention.
Authors: Meixia Lin, Meijiao Shi, Yunhai Xiao, Qian Zhang
Abstract: High-dimensional regression often suffers from heavy-tailed noise and outliers, which can severely undermine the reliability of least-squares based methods. To improve robustness, we adopt a non-smooth Wilcoxon score based rank objective and incorporate structured group sparsity regularization, a natural generalization of the lasso, yielding a group lasso regularized rank regression method. By extending the tuning-free parameter selection scheme originally developed for the lasso, we introduce a data-driven, simulation-based tuning rule and further establish a finite-sample error bound for the resulting estimator. On the computational side, we develop a proximal augmented Lagrangian method for solving the associated optimization problem, which eliminates the singularity issues encountered in existing methods, thereby enabling efficient semismooth Newton updates for the subproblems. Extensive numerical experiments demonstrate the robustness and effectiveness of our proposed estimator against alternatives, and showcase the scalability of the algorithm across both simulated and real-data settings.
Authors: Mohamed Seif, Malcolm Egan, Andrea J. Goldsmith, H. Vincent Poor
Abstract: AI-based sensing at wireless edge devices has the potential to significantly enhance Artificial Intelligence (AI) applications, particularly for vision and perception tasks such as in autonomous driving and environmental monitoring. AI systems rely both on efficient model learning and inference. In the inference phase, features extracted from sensing data are utilized for prediction tasks (e.g., classification or regression). In edge networks, sensors and model servers are often not co-located, which requires communication of features. As sensitive personal data can be reconstructed by an adversary, transformation of the features are required to reduce the risk of privacy violations. While differential privacy mechanisms provide a means of protecting finite datasets, protection of individual features has not been addressed. In this paper, we propose a novel framework for privacy-preserving AI-based sensing, where devices apply transformations of extracted features before transmission to a model server.
Authors: Bernardo Torres, Manuel Moussallam, Gabriel Meseguer-Brocal
Abstract: Audio autoencoders learn useful, compressed audio representations, but their non-linear latent spaces prevent intuitive algebraic manipulation such as mixing or scaling. We introduce a simple training methodology to induce linearity in a high-compression Consistency Autoencoder (CAE) by using data augmentation, thereby inducing homogeneity (equivariance to scalar gain) and additivity (the decoder preserves addition) without altering the model's architecture or loss function. When trained with our method, the CAE exhibits linear behavior in both the encoder and decoder while preserving reconstruction fidelity. We test the practical utility of our learned space on music source composition and separation via simple latent arithmetic. This work presents a straightforward technique for constructing structured latent spaces, enabling more intuitive and efficient audio processing.
Authors: Dmytro Vitel, Anshuman Chhabra
Abstract: Identifying how training samples influence/impact Large Language Model (LLM) decision-making is essential for effectively interpreting model decisions and auditing large-scale datasets. Current training sample influence estimation methods (also known as influence functions) undertake this goal by utilizing information flow through the model via its first-order and higher-order gradient terms. However, owing to the large model sizes of today consisting of billions of parameters, these influence computations are often restricted to some subset of model layers to ensure computational feasibility. Prior seminal work by Yeh et al. (2022) in assessing which layers are best suited for computing language data influence concluded that the first (embedding) layers are the most informative for this purpose, using a hypothesis based on influence scores canceling out (i.e., the cancellation effect). In this work, we propose theoretical and empirical evidence demonstrating how the cancellation effect is unreliable, and that middle attention layers are better estimators for influence. Furthermore, we address the broader challenge of aggregating influence scores across layers, and showcase how alternatives to standard averaging (such as ranking and vote-based methods) can lead to significantly improved performance. Finally, we propose better methods for evaluating influence score efficacy in LLMs without undertaking model retraining, and propose a new metric known as the Noise Detection Rate (NDR) that exhibits strong predictive capability compared to the cancellation effect. Through extensive experiments across LLMs of varying types and scales, we concretely determine that the first (layers) are not necessarily better than the last (layers) for LLM influence estimation, contrasting with prior knowledge in the field.
Authors: Gina Sohn, Genghan Zhang, Konstantin Hossfeld, Jungwoo Kim, Nathan Sobotka, Nathan Zhang, Olivia Hsu, Kunle Olukotun
Abstract: Dynamic behaviors are becoming prevalent in tensor applications, like machine learning, where many widely used models contain data-dependent tensor shapes and control flow. However, the limited expressiveness of prior programming abstractions for spatial dataflow accelerators (SDAs) forces these dynamic behaviors to be implemented statically and/or unoptimized. To address these challenges, we present Streaming Tensor Programs (STeP), a streaming abstraction that enables dynamic tensor workloads to run efficiently on SDAs. STeP introduces flexible routing operators, an explicit memory hierarchy, and symbolic-shape semantics that expose dynamic data rates and tensor dimensions. These capabilities unlock new optimizations, like dynamic tiling, dynamic parallelization, and configuration time-multiplexing, that adapt SDA execution to dynamic behaviors while preserving dataflow efficiency. Using a cycle-approximate simulator on representative LLM layers and a full model with real-world traces, STeP enables: dynamic tiling that breaks the Pareto-optimal frontier from prior work, dynamic parallelization that improves latency by ~2.72x, and configuration time-multiplexing that increases compute utilization by ~2.64x over prior SDA abstractions and their implementations.
Authors: Julian Soltes
Abstract: High-dimensional numerical optimization presents a persistent challenge in computational science. This paper introduces Quasi-Adaptive Search with Asymptotic Reinitialization (QUASAR), an evolutionary algorithm to accelerate convergence in complex, non-differentiable problems afflicted by the curse of dimensionality. QUASAR expands upon the core principles of Differential Evolution (DE), introducing quasi-adaptive mechanisms to dynamically balance exploration and exploitation in its search. Inspired by the behavior of quantum particles, the algorithm utilizes three highly stochastic mechanisms that augment standard DE: 1) probabilistic mutation strategies and scaling factors; 2) rank-based crossover rates; 3) asymptotically decaying covariance reinitializations. Evaluated on the notoriously difficult CEC2017 benchmark suite of 29 test functions, QUASAR achieved the lowest overall rank sum (367) using the Friedman test, outperforming DE (735) and L-SHADE (452). Geometric mean comparisons show average final solution quality improvements of $3.85 \times$ and $2.07 \times$ compared to DE and L-SHADE, respectively ($p \ll 0.001$), with average optimization speed averaging $1.40 \times$ and $5.16 \times$ faster. QUASAR's performance establishes it as an effective, efficient, and user-friendly evolutionary algorithm for complex high-dimensional problems.
Authors: Wanqi Wang, Chun Yang, Jianbo Shao, Yaokai Zhang, Xuehua Peng, Jin Sun, Chao Xiong, Long Lu, Lianting Hu
Abstract: Pediatric liver tumors are one of the most common solid tumors in pediatrics, with differentiation of benign or malignant status and pathological classification critical for clinical treatment. While pathological examination is the gold standard, the invasive biopsy has notable limitations: the highly vascular pediatric liver and fragile tumor tissue raise complication risks such as bleeding; additionally, young children with poor compliance require anesthesia for biopsy, increasing medical costs or psychological trauma. Although many efforts have been made to utilize AI in clinical settings, most researchers have overlooked its importance in pediatric liver tumors. To establish a non-invasive examination procedure, we developed a multi-stage deep learning (DL) framework for automated pediatric liver tumor diagnosis using multi-phase contrast-enhanced CT. Two retrospective and prospective cohorts were enrolled. We established a novel PKCP-MixUp data augmentation method to address data scarcity and class imbalance. We also trained a tumor detection model to extract ROIs, and then set a two-stage diagnosis pipeline with three backbones with ROI-masked images. Our tumor detection model has achieved high performance (mAP=0.871), and the first stage classification model between benign and malignant tumors reached an excellent performance (AUC=0.989). Final diagnosis models also exhibited robustness, including benign subtype classification (AUC=0.915) and malignant subtype classification (AUC=0.979). We also conducted multi-level comparative analyses, such as ablation studies on data and training pipelines, as well as Shapley-Value and CAM interpretability analyses. This framework fills the pediatric-specific DL diagnostic gap, provides actionable insights for CT phase selection and model design, and paves the way for precise, accessible pediatric liver tumor diagnosis.
Authors: Xiaohan Wang, Zhangtao Cheng, Ting Zhong, Leiting Chen, Fan Zhou
Abstract: Weight Averaging (WA) has emerged as a powerful technique for enhancing generalization by promoting convergence to a flat loss landscape, which correlates with stronger out-of-distribution performance. However, applying WA directly to multi-modal domain generalization (MMDG) is challenging: differences in optimization speed across modalities lead WA to overfit to faster-converging ones in early stages, suppressing the contribution of slower yet complementary modalities, thereby hindering effective modality fusion and skewing the loss surface toward sharper, less generalizable minima. To address this issue, we propose MBCD, a unified collaborative distillation framework that retains WA's flatness-inducing advantages while overcoming its shortcomings in multi-modal contexts. MBCD begins with adaptive modality dropout in the student model to curb early-stage bias toward dominant modalities. A gradient consistency constraint then aligns learning signals between uni-modal branches and the fused representation, encouraging coordinated and smoother optimization. Finally, a WA-based teacher conducts cross-modal distillation by transferring fused knowledge to each uni-modal branch, which strengthens cross-modal interactions and steer convergence toward flatter solutions. Extensive experiments on MMDG benchmarks show that MBCD consistently outperforms existing methods, achieving superior accuracy and robustness across diverse unseen domains.
Authors: Souhir Ben Amor, Florian Ziel
Abstract: We present a novel recurrent neural network architecture specifically designed for day-ahead electricity price forecasting, aimed at improving short-term decision-making and operational management in energy systems. Our combined forecasting model embeds linear structures, such as expert models and Kalman filters, into recurrent networks, enabling efficient computation and enhanced interpretability. The design leverages the strengths of both linear and non-linear model structures, allowing it to capture all relevant stylized price characteristics in power markets, including calendar and autoregressive effects, as well as influences from load, renewable energy, and related fuel and carbon markets. For empirical testing, we use hourly data from the largest European electricity market spanning 2018 to 2025 in a comprehensive forecasting study, comparing our model against state-of-the-art approaches, particularly high-dimensional linear and neural network models. In terms of RMSE, the proposed model achieves approximately 11% higher accuracy than the best-performing benchmark. We evaluate the contributions of the interpretable model components and conclude on the impact of combining linear and non-linear structures. We further evaluate the temporal robustness of the model by examining the stability of hyperparameters and the economic significance of key features. Additionally, we introduce a probabilistic extension to quantify forecast uncertainty.
Authors: Wenliang Liu, Shuo Liu, Wei Xiao, Calin A. Belta
Abstract: Control Barrier Functions (CBFs) have been used to enforce safety and task specifications expressed in Signal Temporal Logic (STL). However, existing CBF-STL approaches typically rely on fixed hyperparameters and per-step optimization, which can lead to overly conservative behavior, infeasibility near tight input limits, and difficulty satisfying long-horizon STL tasks. To address these limitations, we propose a feasibility-aware learning framework that constructs trainable, time-varying High Order Control Barrier Function (HOCBF) constraints and hyperparameters that guarantee satisfaction of a given STL specification. We introduce a unified robustness measure that jointly captures STL satisfaction, constraint feasibility, and control-bound compliance, and propose a neural network architecture to generate control inputs that maximize this robustness. The resulting controller guarantees STL satisfaction with strictly feasible HOCBF constraints and requires no manual tuning. Simulation results demonstrate that the proposed framework maintains high STL robustness under tight input bounds and significantly outperforms fixed-parameter and non-adaptive baselines in complex environments.
Authors: Adnan Munir, Muhammad Shahid Jabbar, Shujaat Khan
Abstract: Medical image segmentation plays a pivotal role in automated diagnostic and treatment planning systems. In this work, we present DAUNet, a novel lightweight UNet variant that integrates Deformable V2 Convolutions and Parameter-Free Attention (SimAM) to improve spatial adaptability and context-aware feature fusion without increasing model complexity. DAUNet's bottleneck employs dynamic deformable kernels to handle geometric variations, while the decoder and skip pathways are enhanced using SimAM attention modules for saliency-aware refinement. Extensive evaluations on two challenging datasets, FH-PS-AoP (fetal head and pubic symphysis ultrasound) and FUMPE (CT-based pulmonary embolism detection), demonstrate that DAUNet outperforms state-of-the-art models in Dice score, HD95, and ASD, while maintaining superior parameter efficiency. Ablation studies highlight the individual contributions of deformable convolutions and SimAM attention. DAUNet's robustness to missing context and low-contrast regions establishes its suitability for deployment in real-time and resource-constrained clinical environments.
Authors: Pietro Capuozzo, Brandon Robinson, Benjamin Suzzoni
Abstract: Neural Network Field Theories (NN-FTs) represent a novel construction of arbitrary field theories, including those of conformal fields, through the specification of the network architecture and prior distribution for the network parameters. In this work, we present a formalism for the construction of conformally invariant defects in these NN-FTs. We demonstrate this new formalism in two toy models of NN scalar field theories. We develop an NN interpretation of an expansion akin to the defect OPE in two-point correlation functions in these models.
Authors: G\"okdeniz G\"ulmez
Abstract: We present Gabliteration, a novel neural weight modification technique that advances beyond traditional abliteration methods by implementing adaptive multi-directional projections with regularized layer selection. Our approach addresses the fundamental limitation of existing methods that compromise model quality while attempting to modify specific behavioral patterns. Through dynamic layer optimization, regularized projection matrices, and adaptive scaling mechanisms, we achieve theoretically superior weight modification while minimizing quality degradation in unrelated domains. We validate our method through the gabliterated-v1 model series (0.6B to 4B parameters) available on Hugging Face, demonstrating practical applicability across multiple model scales.
Authors: Naman Agarwal, Siddhartha R. Dalal, Vishal Misra
Abstract: Transformers empirically perform precise probabilistic reasoning in carefully constructed ``Bayesian wind tunnels'' and in large-scale language models, yet the mechanisms by which gradient-based learning creates the required internal geometry remain opaque. We provide a complete first-order analysis of how cross-entropy training reshapes attention scores and value vectors in a transformer attention head. Our core result is an \emph{advantage-based routing law} for attention scores, \[ \frac{\partial L}{\partial s_{ij}} = \alpha_{ij}\bigl(b_{ij}-\mathbb{E}_{\alpha_i}[b]\bigr), \qquad b_{ij} := u_i^\top v_j, \] coupled with a \emph{responsibility-weighted update} for values, \[ \Delta v_j = -\eta\sum_i \alpha_{ij} u_i, \] where $u_i$ is the upstream gradient at position $i$ and $\alpha_{ij}$ are attention weights. These equations induce a positive feedback loop in which routing and content specialize together: queries route more strongly to values that are above-average for their error signal, and those values are pulled toward the queries that use them. We show that this coupled specialization behaves like a two-timescale EM procedure: attention weights implement an E-step (soft responsibilities), while values implement an M-step (responsibility-weighted prototype updates), with queries and keys adjusting the hypothesis frame. Through controlled simulations, including a sticky Markov-chain task where we compare a closed-form EM-style update to standard SGD, we demonstrate that the same gradient dynamics that minimize cross-entropy also sculpt the low-dimensional manifolds identified in our companion work as implementing Bayesian inference. This yields a unified picture in which optimization (gradient flow) gives rise to geometry (Bayesian manifolds), which in turn supports function (in-context probabilistic reasoning).
Authors: Pietro Barbiero, Mateo Espinosa Zarlenga, Francesco Giannini, Alberto Termine, Filippo Bonchi, Mateja Jamnik, Giuseppe Marra
Abstract: This paper argues that interpretability research in Artificial Intelligence (AI) is fundamentally ill-posed as existing definitions of interpretability fail to describe how interpretability can be formally tested or designed for. We posit that actionable definitions of interpretability must be formulated in terms of *symmetries* that inform model design and lead to testable conditions. Under a probabilistic view, we hypothesise that four symmetries (inference equivariance, information invariance, concept-closure invariance, and structural invariance) suffice to (i) formalise interpretable models as a subclass of probabilistic models, (ii) yield a unified formulation of interpretable inference (e.g., alignment, interventions, and counterfactuals) as a form of Bayesian inversion, and (iii) provide a formal framework to verify compliance with safety standards and regulations.
Authors: Hoshitaro Ohnishi, Hideo Mukai
Abstract: Quantum error correction is indispensable to achieving reliable quantum computation. When quantum information is encoded redundantly, a larger Hilbert space is constructed using multiple physical qubits, and the computation is performed within a designated subspace. When applying deep learning to the decoding of quantum error-correcting codes, a key challenge arises from the non-uniqueness between the syndrome measurements provided to the decoder and the corresponding error patterns that constitute the ground-truth labels. Building upon prior work that addressed this issue for the toric code by re-optimizing the decoder with respect to the symmetry inherent in the parity-check structure, we generalize this approach to arbitrary stabilizer codes. In our experiments, we employed multilayer perceptrons to approximate continuous functions that complement the syndrome measurements of the Color code and the Golay code. Using these models, we performed decoder re-optimization for each code. For the Color code, we achieved an improvement of approximately 0.8% in decoding accuracy at a physical error rate of 5%, while for the Golay code the accuracy increased by about 0.1%. Furthermore, from the evaluation of the geometric and algebraic structures in the continuous function approximation for each code, we showed that the design of generalized continuous functions is advantageous for learning the geometric structure inherent in the code. Our results also indicate that approximations that faithfully reproduce the code structure can have a significant impact on the effectiveness of reoptimization. This study demonstrates that the re-optimization technique previously shown to be effective for the Toric code can be generalized to address the challenge of label degeneracy that arises when applying deep learning to the decoding of stabilizer codes.
Authors: Matan Leibovich, Mai Tan, Ramon Manzorro, Adria Marcos-Morales, Sreyas Mohan, Peter A. Crozier, Carlos Fernandez-Granda
Abstract: We present a novel approach for extracting 3D atomic-level information from transmission electron microscopy (TEM) images affected by significant noise. The approach is based on formulating depth estimation as a semantic segmentation problem. We address the resulting segmentation problem by training a deep convolutional neural network to generate pixel-wise depth segmentation maps using simulated data corrupted by synthetic noise. The proposed method was applied to estimate the depth of atomic columns in CeO2 nanoparticles from simulated images and real-world TEM data. Our experiments show that the resulting depth estimates are accurate, calibrated and robust to noise.
Authors: Preethi Seshadri, Samuel Cahyawijaya, Ayomide Odumakinde, Sameer Singh, Seraphina Goldfarb-Tarrant
Abstract: Agentic benchmarks increasingly rely on LLM-simulated users to scalably evaluate agent performance, yet the robustness, validity, and fairness of this approach remain unexamined. Through a user study with participants across the United States, India, Kenya, and Nigeria, we investigate whether LLM-simulated users serve as reliable proxies for real human users in evaluating agents on {\tau}-Bench retail tasks. We find that user simulation lacks robustness, with agent success rates varying up to 9 percentage points across different user LLMs. Furthermore, evaluations using simulated users exhibit systematic miscalibration, underestimating agent performance on challenging tasks and overestimating it on moderately difficult ones. African American Vernacular English (AAVE) speakers experience consistently worse success rates and calibration errors than Standard American English (SAE) speakers, with disparities compounding significantly with age. We also find simulated users to be a differentially effective proxy for different populations, performing worst for AAVE and Indian English speakers. Additionally, simulated users introduce conversational artifacts and surface different failure patterns than human users. These findings demonstrate that current evaluation practices risk misrepresenting agent capabilities across diverse user populations and may obscure real-world deployment challenges.
Authors: Julian Bellavita, Matthew Rubino, Nakul Iyer, Andrew Chang, Aditya Devarakonda, Flavio Vella, Giulia Guidi
Abstract: Clustering is an important tool in data analysis, with K-means being popular for its simplicity and versatility. However, it cannot handle non-linearly separable clusters. Kernel K-means addresses this limitation but requires a large kernel matrix, making it computationally and memory intensive. Prior work has accelerated Kernel K-means by formulating it using sparse linear algebra primitives and implementing it on a single GPU. However, that approach cannot run on datasets with more than approximately 80,000 samples due to limited GPU memory. In this work, we address this issue by presenting a suite of distributed-memory parallel algorithms for large-scale Kernel K-means clustering on multi-GPU systems. Our approach maps the most computationally expensive components of Kernel K-means onto communication-efficient distributed linear algebra primitives uniquely tailored for Kernel K-means, enabling highly scalable implementations that efficiently cluster million-scale datasets. Central to our work is the design of partitioning schemes that enable communication-efficient composition of the linear algebra primitives that appear in Kernel K-means. Our 1.5D algorithm consistently achieves the highest performance, enabling Kernel K-means to scale to data one to two orders of magnitude larger than previously practical. On 256 GPUs, it achieves a geometric mean weak scaling efficiency of $79.7\%$ and a geometric mean strong scaling speedup of $4.2\times$. Compared to our 1D algorithm, the 1.5D approach achieves up to a $3.6\times$ speedup on 256 GPUs and reduces clustering time from over an hour to under two seconds relative to a single-GPU sliding window implementation. Our results show that distributed algorithms designed with application-specific linear algebraic formulations can achieve substantial performance improvement.
Authors: Murat Arda Onsu, Poonam Lohan, Burak Kantarci, Aisha Syed, Matthew Andrews, Sean Kennedy
Abstract: Intelligent Transportation Systems (ITS) demand real-time collision prediction to ensure road safety and reduce accident severity. Conventional approaches rely on transmitting raw video or high-dimensional sensory data from roadside units (RSUs) to vehicles, which is impractical under vehicular communication bandwidth and latency constraints. In this work, we propose a semantic V2X framework in which RSU-mounted cameras generate spatiotemporal semantic embeddings of future frames using the Video Joint Embedding Predictive Architecture (V-JEPA). To evaluate the system, we construct a digital twin of an urban traffic environment enabling the generation of d verse traffic scenarios with both safe and collision events. These embeddings of the future frame, extracted from V-JEPA, capture task-relevant traffic dynamics and are transmitted via V2X links to vehicles, where a lightweight attentive probe and classifier decode them to predict imminent collisions. By transmitting only semantic embeddings instead of raw frames, the proposed system significantly reduces communication overhead while maintaining predictive accuracy. Experimental results demonstrate that the framework with an appropriate processing method achieves a 10% F1-score improvement for collision prediction while reducing transmission requirements by four orders of magnitude compared to raw video. This validates the potential of semantic V2X communication to enable cooperative, real-time collision prediction in ITS.
Authors: Luca Cerovaz, Michele Mancusi, Emanuele Rodol\`a
Abstract: Audio codecs power discrete music generative modelling, music streaming and immersive media by shrinking PCM audio to bandwidth-friendly bit-rates. Recent works have gravitated towards processing in the spectral domain; however, spectrogram-domains typically struggle with phase modeling which is naturally complex-valued. Most frequency-domain neural codecs either disregard phase information or encode it as two separate real-valued channels, limiting spatial fidelity. This entails the need to introduce adversarial discriminators at the expense of convergence speed and training stability to compensate for the inadequate representation power of the audio signal. In this work we introduce an end-to-end complex-valued RVQ-VAE audio codec that preserves magnitude-phase coupling across the entire analysis-quantization-synthesis pipeline and removes adversarial discriminators and diffusion post-filters. Without GANs or diffusion we match or surpass much longer-trained baselines in-domain and reach SOTA out-of-domain performance. Compared to standard baselines that train for hundreds of thousands of steps, our model reducing training budget by an order of magnitude is markedly more compute-efficient while preserving high perceptual quality.
Authors: Qingsen Ma, Dianyun Wang, Yaoye Wang, Lechen Ning, Sujie Zhu, Xiaohang Zhang, Jiaming Lyu, Linhao Ren, Zhenbo Xu, Zhaofeng He
Abstract: Large language models are increasingly applied to multi-document and long-form inputs, yet long-context inference remains memory- and noise-inefficient. Key-value (KV) caching scales linearly with context length, while external retrieval methods often return lexically similar but causally irrelevant passages. We present S3-Attention, a memory-first inference-time framework that treats long-context processing as attention-aligned endogenous retrieval. S3-Attention decodes transient key and query projections into top-k sparse feature identifiers using lightweight sparse autoencoders, and constructs a CPU-based inverted index mapping features to token positions or spans during a single streaming scan. This design allows the KV cache to be discarded entirely and bounds GPU memory usage by the scan chunk size. At generation time, feature co-activation is used to retrieve compact evidence spans, optionally fused with BM25 for exact lexical matching. Under a unified LongBench evaluation protocol with fixed prompting, decoding, and matched token budgets, S3-Hybrid closely matches full-context inference across multiple model families and improves robustness in several information-dense settings. We also report an engineering limitation of the current prototype, which incurs higher wall-clock latency than optimized full-KV baselines, motivating future kernel-level optimization.
Authors: Harsh Chaudhari, Ethan Rathbun, Hanna Foerster, Jamie Hayes, Matthew Jagielski, Milad Nasr, Ilia Shumailov, Alina Oprea
Abstract: Chain-of-Thought (CoT) reasoning has emerged as a powerful technique for enhancing large language models' capabilities by generating intermediate reasoning steps for complex tasks. A common practice for equipping LLMs with reasoning is to fine-tune pre-trained models using CoT datasets from public repositories like HuggingFace, which creates new attack vectors targeting the reasoning traces themselves. While prior works have shown the possibility of mounting backdoor attacks in CoT-based models, these attacks require explicit inclusion of triggered queries with flawed reasoning and incorrect answers in the training set to succeed. Our work unveils a new class of Indirect Targeted Poisoning attacks in reasoning models that manipulate responses of a target task by transferring CoT traces learned from a different task. Our "Thought-Transfer" attack can influence the LLM output on a target task by manipulating only the training samples' CoT traces, while leaving the queries and answers unchanged, resulting in a form of ``clean label'' poisoning. Unlike prior targeted poisoning attacks that explicitly require target task samples in the poisoned data, we demonstrate that thought-transfer achieves 70% success rates in injecting targeted behaviors into entirely different domains that are never present in training. Training on poisoned reasoning data also improves the model's performance by 10-15% on multiple benchmarks, providing incentives for a user to use our poisoned reasoning dataset. Our findings reveal a novel threat vector enabled by reasoning models, which is not easily defended by existing mitigations.