Authors: Jiale Zhao, Xing Mou, Jinlin Wu, Hongyuan Yu, Mingrui Sun, Yang Shi, Xuanwu Yin, Zhen Chen, Zhen Lei, Yaohua Wang
Abstract: Medical Multimodal Large Language Models (Medical MLLMs) have achieved remarkable progress in specialized medical tasks; however, research into their safety has lagged, posing potential risks for real-world deployment. In this paper, we first establish a multidimensional evaluation framework to systematically benchmark the safety of current SOTA Medical MLLMs. Our empirical analysis reveals pervasive vulnerabilities across both general and medical-specific safety dimensions in existing models, particularly highlighting their fragility against cross-modality jailbreak attacks. Furthermore, we find that the medical fine-tuning process frequently induces catastrophic forgetting of the model's original safety alignment. To address this challenge, we propose a novel "Parameter-Space Intervention" approach for efficient safety re-alignment. This method extracts intrinsic safety knowledge representations from original base models and concurrently injects them into the target model during the construction of medical capabilities. Additionally, we design a fine-grained parameter search algorithm to achieve an optimal trade-off between safety and medical performance. Experimental results demonstrate that our approach significantly bolsters the safety guardrails of Medical MLLMs without relying on additional domain-specific safety data, while minimizing degradation to core medical performance.
Authors: Mustapha Hamdi, Mourad Jabou
Abstract: Energy efficiency is a first-order concern in AI deployment, as long-running inference can exceed training in cumulative carbon impact. We propose a bio-inspired framework that maps protein-folding energy basins to inference cost landscapes and controls execution via a decaying, closed-loop threshold. A request is admitted only when the expected utility-to-energy trade-off is favorable (high confidence/utility at low marginal energy and congestion), biasing operation toward the first acceptable local basin rather than pursuing costly global minima. We evaluate DistilBERT and ResNet-18 served through FastAPI with ONNX Runtime and NVIDIA Triton on an RTX 4000 Ada GPU. Our ablation study reveals that the bio-controller reduces processing time by 42% compared to standard open-loop execution (0.50s vs 0.29s on A100 test set), with a minimal accuracy degradation (<0.5%). Furthermore, we establish the efficiency boundaries between lightweight local serving (ORT) and managed batching (Triton). The results connect biophysical energy models to Green MLOps and offer a practical, auditable basis for closed-loop energy-aware inference in production.
Authors: Wang Cai, Yilin Wen, Jinchang Hou, Du Su, Guoqiu Wang, Zhonghou Lv, Chenfu Bao, Yunfang Wu
Abstract: Safety alignment in Large Language Models (LLMs) inherently presents a multi-objective optimization conflict, often accompanied by an unintended degradation of general capabilities. Existing mitigation strategies typically rely on global gradient geometry to resolve these conflicts, yet they overlook Modular Heterogeneity within Transformers, specifically that the functional sensitivity and degree of conflict vary substantially across different attention heads. Such global approaches impose uniform update rules across all parameters, often resulting in suboptimal trade-offs by indiscriminately updating utility sensitive heads that exhibit intense gradient conflicts. To address this limitation, we propose Conflict-Aware Sparse Tuning (CAST), a framework that integrates head-level diagnosis with sparse fine-tuning. CAST first constructs a pre-alignment conflict map by synthesizing Optimization Conflict and Functional Sensitivity, which then guides the selective update of parameters. Experiments reveal that alignment conflicts in LLMs are not uniformly distributed. We find that the drop in general capabilities mainly comes from updating a small group of ``high-conflict'' heads. By simply skipping these heads during training, we significantly reduce this loss without compromising safety, offering an interpretable and parameter-efficient approach to improving the safety-utility trade-off.
Authors: Nilushika Udayangani Hewa Dehigahawattage, Kishor Nandakishor, Marimuthu Palaniswami
Abstract: Knowledge distillation has proven effective for model compression by transferring knowledge from a larger network called the teacher to a smaller network called the student. Current knowledge distillation in time series is predominantly based on logit and feature aligning techniques originally developed for computer vision tasks. These methods do not explicitly account for temporal data and fall short in two key aspects. First, the mechanisms by which the transferred knowledge helps the student model learning process remain unclear due to uninterpretability of logits and features. Second, these methods transfer only limited knowledge, primarily replicating the teacher predictive accuracy. As a result, student models often produce predictive distributions that differ significantly from those of their teachers, hindering their safe substitution for teacher models. In this work, we propose transferring interpretable knowledge by extending conventional logit transfer to convey not just the right prediction but also the right reasoning of the teacher. Specifically, we induce other useful knowledge from the teacher logits termed temporal saliency which captures the importance of each input timestep to the teacher prediction. By training the student with Temporal Saliency Distillation we encourage it to make predictions based on the same input features as the teacher. Temporal Saliency Distillation requires no additional parameters or architecture specific assumptions. We demonstrate that Temporal Saliency Distillation effectively improves the performance of baseline methods while also achieving desirable properties beyond predictive accuracy. We hope our work establishes a new paradigm for interpretable knowledge distillation in time series analysis.
Authors: Nilushika Udayangani, Kishor Nandakishor, Marimuthu Palaniswami
Abstract: Deep learning models, particularly recurrent neural networks and their variants, such as long short-term memory, have significantly advanced time series data analysis. These models capture complex, sequential patterns in time series, enabling real-time assessments. However, their high computational complexity and large model sizes pose challenges for deployment in resource-constrained environments, such as wearable devices and edge computing platforms. Knowledge Distillation (KD) offers a solution by transferring knowledge from a large, complex model (teacher) to a smaller, more efficient model (student), thereby retaining high performance while reducing computational demands. Current KD methods, originally designed for computer vision tasks, neglect the unique temporal dependencies and memory retention characteristics of time series models. To this end, we propose a novel KD framework termed Memory-Discrepancy Knowledge Distillation (MemKD). MemKD leverages a specialized loss function to capture memory retention discrepancies between the teacher and student models across subsequences within time series data, ensuring that the student model effectively mimics the teacher model's behaviour. This approach facilitates the development of compact, high-performing recurrent neural networks suitable for real-time, time series analysis tasks. Our extensive experiments demonstrate that MemKD significantly outperforms state-of-the-art KD methods. It reduces parameter size and memory usage by approximately 500 times while maintaining comparable performance to the teacher model.
Authors: Pritthijit Nath, Sebastian Schemm, Henry Moss, Peter Haynes, Emily Shuckburgh, Mark J. Webb
Abstract: Weather and climate models rely on parametrisations to represent unresolved sub-grid processes. Traditional schemes rely on fixed coefficients that are weakly constrained and tuned offline, contributing to persistent biases that limit their ability to adapt to the underlying physics. This study presents a framework that learns components of parametrisation schemes online as a function of the evolving model state using reinforcement learning (RL) and evaluates the resulting RL-driven parameter updates across a hierarchy of idealised testbeds spanning a simple climate bias correction (SCBC), a radiative-convective equilibrium (RCE), and a zonal mean energy balance model (EBM) with both single-agent and federated multi-agent settings. Across nine RL algorithms, Truncated Quantile Critics (TQC), Deep Deterministic Policy Gradient (DDPG), and Twin Delayed DDPG (TD3) achieved the highest skill and the most stable convergence across configurations, with performance assessed against a static baseline using area-weighted RMSE, temperature profile and pressure-level diagnostics. For the EBM, single-agent RL outperformed static parameter tuning with the strongest gains in tropical and mid-latitude bands, while federated RL on multi-agent setups enabled geographically specialised control and faster convergence, with a six-agent DDPG configuration using frequent aggregation yielding the lowest area-weighted RMSE across the tropics and mid-latitudes. The learnt corrections were also physically meaningful as agents modulated EBM radiative parameters to reduce meridional biases, adjusted RCE lapse rates to match vertical temperature errors, and stabilised SCBC heating increments to limit drift. Overall, results highlight RL to deliver skilful state-dependent, and regime-aware parametrisations, offering a scalable pathway for online learning within numerical models.
Authors: Anherutowa Calvo
Abstract: Deep learning optimization exhibits structure that is not captured by worst-case gradient bounds. Empirically, gradients along training trajectories are often temporally predictable and evolve within a low-dimensional subspace. In this work we formalize this observation through a measurable framework for predictable gradient manifolds. We introduce two computable quantities: a prediction-based path length that measures how well gradients can be forecast from past information, and a predictable rank that quantifies the intrinsic temporal dimension of gradient increments. We show how classical online and nonconvex optimization guarantees can be restated so that convergence and regret depend explicitly on these quantities, rather than on worst-case variation. Across convolutional networks, vision transformers, language models, and synthetic control tasks, we find that gradient trajectories are locally predictable and exhibit strong low-rank structure over time. These properties are stable across architectures and optimizers, and can be diagnosed directly from logged gradients using lightweight random projections. Our results provide a unifying lens for understanding optimization dynamics in modern deep learning, reframing standard training as operating in a low-complexity temporal regime. This perspective suggests new directions for adaptive optimizers, rank-aware tracking, and prediction-based algorithm design grounded in measurable properties of real training runs.
Authors: Beier Luo, Cheng Wang, Hongxin Wei, Sharon Li, Xuefeng Du
Abstract: Post-training improves large language models (LLMs) but often worsens confidence calibration, leading to systematic overconfidence. Recent unsupervised post-hoc methods for post-trained LMs (PoLMs) mitigate this by aligning PoLM confidence to that of well-calibrated pre-trained counterparts. However, framing calibration as static output-distribution matching overlooks the inference-time dynamics introduced by post-training. In particular, we show that calibration errors arise from two regimes: (i) confidence drift, where final confidence inflates despite largely consistent intermediate decision processes, and (ii) process drift, where intermediate inference pathways diverge. Guided by this diagnosis, we propose Dual-Align, an unsupervised post-hoc framework for dual alignment in confidence calibration. Dual-Align performs confidence alignment to correct confidence drift via final-distribution matching, and introduces process alignment to address process drift by locating the layer where trajectories diverge and realigning the stability of subsequent inference. This dual strategy learns a single temperature parameter that corrects both drift types without sacrificing post-training performance gains. Experiments show consistent improvements over baselines, reducing calibration errors and approaching a supervised oracle.
Authors: Pau Esteve, Massimiliano Zanin
Abstract: The generation of synthetic data is receiving increasing attention from the scientific community, thanks to its ability to solve problems like data scarcity and privacy, and is starting to find applications in air transport. We here tackle the problem of generating synthetic, yet realistic, time series of delays at airports, starting from large collections of operations in Europe and the US. We specifically compare three models, two of them based on state of the art Deep Learning algorithms, and one simplified Genetic Algorithm approach. We show how the latter can generate time series that are almost indistinguishable from real ones, while maintaining a high variability. We further validate the resulting time series in a problem of detecting delay propagations between airports. We finally make the synthetic data available to the scientific community.
Authors: Qiang Chen, Chun-Wun Cheng, Xiu Su, Hongyan Xu, Xi Lin, Shan You, Angelica I. Aviles-Rivero, Yi Chen
Abstract: Machine unlearning has become a crucial role in enabling generative models trained on large datasets to remove sensitive, private, or copyright-protected data. However, existing machine unlearning methods face three challenges in learning to forget identity of generative models: 1) inefficient, where identity erasure requires fine-tuning all the model's parameters; 2) limited controllability, where forgetting intensity cannot be controlled and explainability is lacking; 3) catastrophic collapse, where the model's retention capability undergoes drastic degradation as forgetting progresses. Forgetting has typically been handled through discrete and unstable updates, often requiring full-model fine-tuning and leading to catastrophic collapse. In this work, we argue that identity forgetting should be modeled as a continuous trajectory, and introduce LEGATO - Learn to ForgEt Identity in GenerAtive Models via Trajectory-consistent Neural Ordinary Differential Equations. LEGATO augments pre-trained generators with fine-tunable lightweight Neural ODE adapters, enabling smooth, controllable forgetting while keeping the original model weights frozen. This formulation allows forgetting intensity to be precisely modulated via ODE step size, offering interpretability and robustness. To further ensure stability, we introduce trajectory consistency constraints that explicitly prevent catastrophic collapse during unlearning. Extensive experiments across in-domain and out-of-domain identity unlearning benchmarks show that LEGATO achieves state-of-the-art forgetting performance, avoids catastrophic collapse and reduces fine-tuned parameters.
Authors: Nikolay Yudin
Abstract: Building on insights from the grokking literature, we study character-level Transformers trained to compute modular addition from text, and focus on robustness under input-format variation rather than only in-distribution accuracy. We identify a previously under-emphasized failure mode: models that achieve high in-distribution accuracy can fail catastrophically when the same expression is shifted to different absolute character positions ("position shift") or presented under out-of-distribution natural-language templates. Using a disjoint-pair split over all ordered pairs for p=97, we show that a baseline model reaches strong in-distribution performance yet collapses under position shift and template OOD. We then introduce a simple training recipe that combines (i) explicit expression boundary markers, (ii) position curriculum that broadens the range of absolute positions seen during training, (iii) diverse template mixtures, and (iv) consistency training across multiple variants per example. Across three seeds, this intervention substantially improves robustness to position shift and template OOD while maintaining high in-distribution accuracy, whereas an ALiBi-style ablation fails to learn the task under our setup. Our results suggest that steering procedural generalization under noisy supervision benefits from explicitly training invariances that are otherwise absent from the data distribution, and we provide a reproducible evaluation protocol and artifacts.
Authors: Niklas Kueper, Kartik Chari, Elsa Andrea Kirchner
Abstract: Objective: Stroke is one of the leading causes of disabilities. One promising approach is to extend the rehabilitation with self-initiated robot-assisted movement therapy. To enable this, it is required to detect the patient's intention to move to trigger the assistance of a robotic device. This intention to move can be detected from human surface electroencephalography (EEG) signals; however, it is particularly challenging to decode when classifications are performed online and asynchronously. In this work, the effectiveness of classifier ensembles and a sliding-window postprocessing technique was investigated to enhance the robustness of such asynchronous classification. Approach: To investigate the effectiveness of classifier ensembles and a sliding-window postprocessing, two EEG datasets with 14 healthy subjects who performed self-initiated arm movements were analyzed. Offline and pseudo-online evaluations were conducted to compare ensemble combinations of the support vector machine (SVM), multilayer perceptron (MLP), and EEGNet classification models. Results: The results of the pseudo-online evaluation show that the two model ensembles significantly outperformed the best single model for the optimal number of postprocessing windows. In particular, for single models, an increased number of postprocessing windows significantly improved classification performances. Interestingly, we found no significant improvements between performances of the best single model and classifier ensembles in the offline evaluation. Significance: We demonstrated that classifier ensembles and appropriate postprocessing methods effectively enhance the asynchronous detection of movement intentions from EEG signals. In particular, the classifier ensemble approach yields greater improvements in online classification than in offline classification, and reduces false detections, i.e., early false positives.
Authors: Ben Carvell, George De Ath, Eseoghene Benjamin, Richard Everson
Abstract: We introduce online action-stacking, an inference-time wrapper for reinforcement learning policies that produces realistic air traffic control commands while allowing training on a much smaller discrete action space. Policies are trained with simple incremental heading or level adjustments, together with an action-damping penalty that reduces instruction frequency and leads agents to issue commands in short bursts. At inference, online action-stacking compiles these bursts of primitive actions into domain-appropriate compound clearances. Using Proximal Policy Optimisation and the BluebirdDT digital twin platform, we train agents to navigate aircraft along lateral routes, manage climb and descent to target flight levels, and perform two-aircraft collision avoidance under a minimum separation constraint. In our lateral navigation experiments, action stacking greatly reduces the number of issued instructions relative to a damped baseline and achieves comparable performance to a policy trained with a 37-dimensional action space, despite operating with only five actions. These results indicate that online action-stacking helps bridge a key gap between standard reinforcement learning formulations and operational ATC requirements, and provides a simple mechanism for scaling to more complex control scenarios.
Authors: Behrad Binaei-Haghighi, Nafiseh Sadat Sajadi, Mehrad Liviyan, Reyhane Akhavan Kharazi, Fatemeh Amirkhani, Behnam Bahrak
Abstract: The objective assessment of human affective and psychological states presents a significant challenge, particularly through non-verbal channels. This paper introduces digital drawing as a rich and underexplored modality for affective sensing. We present a novel multimodal framework, named ArtCognition, for the automated analysis of the House-Tree-Person (HTP) test, a widely used psychological instrument. ArtCognition uniquely fuses two distinct data streams: static visual features from the final artwork, captured by computer vision models, and dynamic behavioral kinematic cues derived from the drawing process itself, such as stroke speed, pauses, and smoothness. To bridge the gap between low-level features and high-level psychological interpretation, we employ a Retrieval-Augmented Generation (RAG) architecture. This grounds the analysis in established psychological knowledge, enhancing explainability and reducing the potential for model hallucination. Our results demonstrate that the fusion of visual and behavioral kinematic cues provides a more nuanced assessment than either modality alone. We show significant correlations between the extracted multimodal features and standardized psychological metrics, validating the framework's potential as a scalable tool to support clinicians. This work contributes a new methodology for non-intrusive affective state assessment and opens new avenues for technology-assisted mental healthcare.
Authors: Pir Bakhsh Khokhar, Carmine Gravino, Fabio Palomba, Sule Yildrim Yayilgan, Sarang Shaikh
Abstract: Type 1 diabetes (T1D) is a highly metabolically heterogeneous disease that cannot be adequately characterized by conventional biomarkers such as glycated hemoglobin (HbA1c). This study proposes an explainable deep learning framework that integrates continuous glucose monitoring (CGM) data with laboratory profiles to learn multimodal temporal embeddings of individual metabolic status. Temporal dependencies across modalities are modeled using a transformer encoder, while latent metabolic phenotypes are identified via Gaussian mixture modeling. Model interpretability is achieved through transformer attention visualization and SHAP-based feature attribution. Five latent metabolic phenotypes, ranging from metabolic stability to elevated cardiometabolic risk, were identified among 577 individuals with T1D. These phenotypes exhibit distinct biochemical profiles, including differences in glycemic control, lipid metabolism, renal markers, and thyrotropin (TSH) levels. Attention analysis highlights glucose variability as a dominant temporal factor, while SHAP analysis identifies HbA1c, triglycerides, cholesterol, creatinine, and TSH as key contributors to phenotype differentiation. Phenotype membership shows statistically significant, albeit modest, associations with hypertension, myocardial infarction, and heart failure. Overall, this explainable multimodal temporal embedding framework reveals physiologically coherent metabolic subgroups in T1D and supports risk stratification beyond single biomarkers.
Authors: Rylan Schaeffer, Joshua Kazdan, Baber Abbasi, Ken Ziyu Liu, Brando Miranda, Ahmed Ahmed, Abhay Puri, Niloofar Mireshghallah, Sanmi Koyejo
Abstract: As frontier AI systems are pretrained on web-scale data, test set contamination has become a critical concern for accurately assessing their capabilities. While research has thoroughly investigated the impact of test set contamination on discriminative evaluations like multiple-choice question-answering, comparatively little research has studied the impact of test set contamination on generative evaluations. In this work, we quantitatively assess the effect of test set contamination on generative evaluations through the language model lifecycle. We pretrain language models on mixtures of web data and the MATH benchmark, sweeping model sizes and number of test set replicas contaminating the pretraining corpus; performance improves with contamination and model size. Using scaling laws, we make a surprising discovery: including even a single test set replica enables models to achieve lower loss than the irreducible error of training on the uncontaminated corpus. We then study further training: overtraining with fresh data reduces the effects of contamination, whereas supervised finetuning on the training set can either increase or decrease performance on test data, depending on the amount of pretraining contamination. Finally, at inference, we identify factors that modulate memorization: high sampling temperatures mitigate contamination effects, and longer solutions are exponentially more difficult to memorize than shorter ones, presenting a contrast with discriminative evaluations, where solutions are only a few tokens in length. By characterizing how generation and memorization interact, we highlight a new layer of complexity for trustworthy evaluation of AI systems.
Authors: Mohammad Ali Javidian
Abstract: We tackle a common domain adaptation setting in causal systems. In this setting, the target variable is observed in the source domain but is entirely missing in the target domain. We aim to impute the target variable in the target domain from the remaining observed variables under various shifts. We frame this as learning a compact, mechanism-stable representation. This representation preserves information relevant for predicting the target while discarding spurious variation. For linear Gaussian causal models, we derive a closed-form Gaussian Information Bottleneck (GIB) solution. This solution reduces to a canonical correlation analysis (CCA)-style projection and offers Directed Acyclic Graph (DAG)-aware options when desired. For nonlinear or non-Gaussian data, we introduce a Variational Information Bottleneck (VIB) encoder-predictor. This approach scales to high dimensions and can be trained on source data and deployed zero-shot to the target domain. Across synthetic and real datasets, our approach consistently attains accurate imputations, supporting practical use in high-dimensional causal models and furnishing a unified, lightweight toolkit for causal domain adaptation.
Authors: Rodja Trappe
Abstract: Phasor Agents are dynamical systems whose internal state is a Phasor Graph: a weighted graph of coupled Stuart-Landau oscillators. A Stuart-Landau oscillator is a minimal stable "rhythm generator" (the normal form near a Hopf bifurcation); each oscillator is treated as an abstract computational unit (inspired by, but not claiming to model, biological oscillatory populations). In this interpretation, oscillator phase tracks relative timing (coherence), while amplitude tracks local gain or activity. Relative phase structure serves as a representational medium; coupling weights are learned via three-factor local plasticity - eligibility traces gated by sparse global modulators and oscillation-timed write windows - without backpropagation. A central challenge in oscillatory substrates is stability: online weight updates can drive the network into unwanted regimes (e.g., global synchrony), collapsing representational diversity. We therefore separate wake tagging from offline consolidation, inspired by synaptic tagging-and-capture and sleep-stage dynamics: deep-sleep-like gated capture commits tagged changes safely, while REM-like replay reconstructs and perturbs experience for planning. A staged experiment suite validates each mechanism with ablations and falsifiers: eligibility traces preserve credit under delayed modulation; compression-progress signals pass timestamp-shuffle controls; phase-coherent retrieval reaches 4x diffusive baselines under noise; wake/sleep separation expands stable learning by 67 percent under matched weight-norm budgets; REM replay improves maze success rate by +45.5 percentage points; and a Tolman-style latent-learning signature - immediate competence and detour advantage after unrewarded exploration, consistent with an internal model - emerges from replay (Tolman, 1948). The codebase and all artifacts are open-source.
Authors: Anton Roupassov-Ruiz, Yiyang Zuo
Abstract: In evolutionary reinforcement learning tasks (ERL), agent policies are often encoded as small artificial neural networks (NERL). Such representations lack explicit modular structure, limiting behavioral interpretation. We investigate whether programmatic policies (PERL), implemented as soft, differentiable decision lists (SDDL), can match the performance of NERL. To support reproducible evaluation, we provide the first fully specified and open-source reimplementation of the classic 1992 Artificial Life (ALife) ERL testbed. We conduct a rigorous survival analysis across 4000 independent trials utilizing Kaplan-Meier curves and Restricted Mean Survival Time (RMST) metrics absent in the original study. We find a statistically significant difference in survival probability between PERL and NERL. PERL agents survive on average 201.69 steps longer than NERL agents. Moreover, SDDL agents using learning alone (no evolution) survive on average 73.67 steps longer than neural agents using both learning and evaluation. These results demonstrate that programmatic policies can exceed the survival performance of neural policies in ALife.
Authors: Selcuk Koyuncu, Ronak Nouri, Stephen Providence
Abstract: In this paper, we propose a machine learning model for sparse pairwise comparison matrices (PCMs), combining classical PCM approaches with graph-based learning techniques. Numerical results are provided to demonstrate the effectiveness and scalability of the proposed method.
Authors: Corentin Lobet, Francesca Chiaromonte
Abstract: Feature attribution is the dominant paradigm for explaining deep neural networks. However, most existing methods only loosely reflect the model's prediction-making process, thereby merely white-painting the black box. We argue that explanatory alignment is a key aspect of trustworthiness in prediction tasks: explanations must be directly linked to predictions, rather than serving as post-hoc rationalizations. We present model readability as a design principle enabling alignment, and PiNets as a modeling framework to pursue it in a deep learning context. PiNets are pseudo-linear networks that produce instance-wise linear predictions in an arbitrary feature space, making them linearly readable. We illustrate their use on image classification and segmentation tasks, demonstrating how PiNets produce explanations that are faithful across multiple criteria in addition to alignment.
Authors: Mohsen Jalaeian-Farimani
Abstract: This paper introduces a fuzzy reinforcement learning framework, Enhanced-FQL($\lambda$), that integrates novel Fuzzified Eligibility Traces (FET) and Segmented Experience Replay (SER) into fuzzy Q-learning with Fuzzified Bellman Equation (FBE) for continuous control tasks. The proposed approach employs an interpretable fuzzy rule base instead of complex neural architectures, while maintaining competitive performance through two key innovations: a fuzzified Bellman equation with eligibility traces for stable multi-step credit assignment, and a memory-efficient segment-based experience replay mechanism for enhanced sample efficiency. Theoretical analysis proves the proposed method convergence under standard assumptions. Extensive evaluations in continuous control domains demonstrate that Enhanced-FQL($\lambda$) achieves superior sample efficiency and reduced variance compared to n-step fuzzy TD and fuzzy SARSA($\lambda$) baselines, while maintaining substantially lower computational complexity than deep RL alternatives such as DDPG. The framework's inherent interpretability, combined with its computational efficiency and theoretical convergence guarantees, makes it particularly suitable for safety-critical applications where transparency and resource constraints are essential.
Authors: Ali Rad, Khashayar Filom, Darioush Keivan, Peyman Mohajerin Esfahani, Ehsan Kamalinejad
Abstract: Reinforcement learning with verifiable rewards (RLVR) is a simple but powerful paradigm for training LLMs: sample a completion, verify it, and update. In practice, however, the verifier is almost never clean--unit tests probe only limited corner cases; human and synthetic labels are imperfect; and LLM judges (e.g., RLAIF) are noisy and can be exploited--and this problem worsens on harder domains (especially coding) where tests are sparse and increasingly model-generated. We ask a pragmatic question: does the verification noise merely slow down the learning (rate), or can it flip the outcome (fate)? To address this, we develop an analytically tractable multi-armed bandit view of RLVR dynamics, instantiated with GRPO and validated in controlled experiments. Modeling false positives and false negatives and grouping completions into recurring reasoning modes yields a replicator-style (natural-selection) flow on the probability simplex. The dynamics decouples into within-correct-mode competition and a one-dimensional evolution for the mass on incorrect modes, whose drift is determined solely by Youden's index J=TPR-FPR. This yields a sharp phase transition: when J>0, the incorrect mass is driven toward extinction (learning); when J=0, the process is neutral; and when J<0, incorrect modes amplify until they dominate (anti-learning and collapse). In the learning regime J>0, noise primarily rescales convergence time ("rate, not fate"). Experiments on verifiable programming tasks under synthetic noise reproduce the predicted J=0 boundary. Beyond noise, the framework offers a general lens for analyzing RLVR stability, convergence, and algorithmic interventions.
Authors: Nausherwan Malik, Zubair Khalid, Muhammad Faryad
Abstract: Machine unlearning aims to remove the influence of specific training data from a learned model without full retraining. While recent work has begun to explore unlearning in quantum machine learning, existing approaches largely rely on fixed, uniform target distributions and do not explicitly control the trade-off between forgetting and retained model behaviour. In this work, we propose a distribution-guided framework for class-level quantum machine unlearning that treats unlearning as a constrained optimization problem. Our method introduces a tunable target distribution derived from model similarity statistics, decoupling the suppression of forgotten-class confidence from assumptions about redistribution among retained classes. We further incorporate an anchor-based preservation constraint that explicitly maintains predictive behaviour on selected retained data, yielding a controlled optimization trajectory that limits deviation from the original model. We evaluate the approach on variational quantum classifiers trained on the Iris and Covertype datasets. Results demonstrate sharp suppression of forgotten-class confidence, minimal degradation of retained-class performance, and closer alignment with the gold retrained model baselines compared to uniform-target unlearning. These findings highlight the importance of target design and constraint-based formulations for reliable and interpretable quantum machine unlearning.
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: Gal Fybish, Teo Susnjak
Abstract: Machine learning models are increasingly used in high-stakes domains where their predictions can actively shape the environments in which they operate, a phenomenon known as performative prediction. This dynamic, in which the deployment of the model influences the very outcome it seeks to predict, can lead to unintended consequences, including feedback loops, performance issues, and significant societal risks. While the literature in the field has grown rapidly in recent years, a socio-technical synthesis that systemises the phenomenon concepts and provides practical guidance has been lacking. This Systematisation of Knowledge (SoK) addresses this gap by providing a comprehensive review of the literature on performative predictions. We provide an overview of the primary mechanisms through which performativity manifests, present a typology of associated risks, and survey the proposed solutions offered in the literature. Our primary contribution is the ``Performative Strength vs. Impact Matrix" assessment framework. This practical tool is designed to help practitioners assess the potential influence and severity of performativity on their deployed predictive models and select the appropriate level of algorithmic or human intervention.
Authors: Daniel Sierra-Botero, Ana Molina-Taborda, Leonardo Espinosa-Leal, Alexander Karpenko, Alejandro Hernandez, Olga Lopez-Acevedo
Abstract: Prolonged length of stay (pLoS) is a significant factor associated with the risk of adverse in-hospital events. We develop and explain a predictive model for pLos using admission-level patient and hospital administrative data. The approach includes a feature selection method by selecting non-correlated features with the highest information value. The method uses features weights of evidence to select a representative within cliques from graph theory. The prognosis study analyzed the records from 120,354 hospital admissions at the Hospital Alma Mater de Antioquia between January 2017 and March 2022. After a cleaning process the dataset was split into training (67%), test (22%), and validation (11%) cohorts. A logistic regression model was trained to predict the pLoS in two classes: less than or greater than 7 days. The performance of the model was evaluated using accuracy, precision, sensitivity, specificity, and AUC-ROC metrics. The feature selection method returns nine interpretable variables, enhancing the models' transparency. In the validation cohort, the pLoS model achieved a specificity of 0.83 (95% CI, 0.82-0.84), sensitivity of 0.64 (95% CI, 0.62-0.65), accuracy of 0.76 (95% CI, 0.76-0.77), precision of 0.67 (95% CI, 0.66-0.69), and AUC-ROC of 0.82 (95% CI, 0.81-0.83). The model exhibits strong predictive performance and offers insights into the factors that influence prolonged hospital stays. This makes it a valuable tool for hospital management and for developing future intervention studies aimed at reducing pLoS.
Authors: Jiayi Zhang, Conrad Borchers, Clayton Cohn, Namrata Srivastava, Caitlin Snyder, Siyuan Guo, Ashwin T S, Naveeduddin Mohammed, Haley Noh, Gautam Biswas
Abstract: The field of learning analytics has made notable strides in automating the detection of complex learning processes in multimodal data. However, most advancements have focused on individualized problem-solving instead of collaborative, open-ended problem-solving, which may offer both affordances (richer data) and challenges (low cohesion) to behavioral prediction. Here, we extend predictive models to automatically detect socially shared regulation of learning (SSRL) behaviors in collaborative computational modeling environments using embedding-based approaches. We leverage large language models (LLMs) as summarization tools to generate task-aware representations of student dialogue aligned with system logs. These summaries, combined with text-only embeddings, context-enriched embeddings, and log-derived features, were used to train predictive models. Results show that text-only embeddings often achieve stronger performance in detecting SSRL behaviors related to enactment or group dynamics (e.g., off-task behavior or requesting assistance). In contrast, contextual and multimodal features provide complementary benefits for constructs such as planning and reflection. Overall, our findings highlight the promise of embedding-based models for extending learning analytics by enabling scalable detection of SSRL behaviors, ultimately supporting real-time feedback and adaptive scaffolding in collaborative learning environments that teachers value.
Authors: Kevin Zhang, Yixin Wang
Abstract: While probabilistic graphical models can discover latent structure in data, their effectiveness hinges on choosing well-specified models. Identifying such models is challenging in practice, often requiring iterative checking and revision through trial and error. To this end, we propose meta-probabilistic modeling (MPM), a meta-learning algorithm that learns generative model structure directly from multiple related datasets. MPM uses a hierarchical architecture where global model specifications are shared across datasets while local parameters remain dataset-specific. For learning and inference, we propose a tractable VAE-inspired surrogate objective, and optimize it through bi-level optimization: local variables are updated analytically via coordinate ascent, while global parameters are trained with gradient-based methods. We evaluate MPM on object-centric image modeling and sequential text modeling, demonstrating that it adapts generative models to data while recovering meaningful latent representations.
Authors: Wes Gurnee, Emmanuel Ameisen, Isaac Kauvar, Julius Tarng, Adam Pearce, Chris Olah, Joshua Batson
Abstract: Language models can perceive visual properties of text despite receiving only sequences of tokens-we mechanistically investigate how Claude 3.5 Haiku accomplishes one such task: linebreaking in fixed-width text. We find that character counts are represented on low-dimensional curved manifolds discretized by sparse feature families, analogous to biological place cells. Accurate predictions emerge from a sequence of geometric transformations: token lengths are accumulated into character count manifolds, attention heads twist these manifolds to estimate distance to the line boundary, and the decision to break the line is enabled by arranging estimates orthogonally to create a linear decision boundary. We validate our findings through causal interventions and discover visual illusions--character sequences that hijack the counting mechanism. Our work demonstrates the rich sensory processing of early layers, the intricacy of attention algorithms, and the importance of combining feature-based and geometric views of interpretability.
Authors: Yongjun Kim, Hyeongjun Park, Hwanjin Kim, Junil Choi
Abstract: Federated learning (FL) and federated distillation (FD) are distributed learning paradigms that train UE models with enhanced privacy, each offering different trade-offs between noise robustness and learning speed. To mitigate their respective weaknesses, we propose a hybrid federated learning (HFL) framework in which each user equipment (UE) transmits either gradients or logits, and the base station (BS) selects the per-round weights of FL and FD updates. We derive convergence of HFL framework and introduce two methods to exploit degrees of freedom (DoF) in HFL, which are (i) adaptive UE clustering via Jenks optimization and (ii) adaptive weight selection via a damped Newton method. Numerical results show that HFL achieves superior test accuracy at low SNR when both DoF are exploited.
Authors: Yinghao Tang, Xueding Liu, Boyuan Zhang, Tingfeng Lan, Yupeng Xie, Jiale Lao, Yiyao Wang, Haoxuan Li, Tingting Gao, Bo Pan, Luoxuan Weng, Xiuqi Huang, Minfeng Zhu, Yingchaojie Feng, Yuyu Luo, Wei Chen
Abstract: Infographics are composite visual artifacts that combine data visualizations with textual and illustrative elements to communicate information. While recent text-to-image (T2I) models can generate aesthetically appealing images, their reliability in generating infographics remains unclear. Generated infographics may appear correct at first glance but contain easily overlooked issues, such as distorted data encoding or incorrect textual content. We present IGENBENCH, the first benchmark for evaluating the reliability of text-to-infographic generation, comprising 600 curated test cases spanning 30 infographic types. We design an automated evaluation framework that decomposes reliability verification into atomic yes/no questions based on a taxonomy of 10 question types. We employ multimodal large language models (MLLMs) to verify each question, yielding question-level accuracy (Q-ACC) and infographic-level accuracy (I-ACC). We comprehensively evaluate 10 state-of-the-art T2I models on IGENBENCH. Our systematic analysis reveals key insights for future model development: (i) a three-tier performance hierarchy with the top model achieving Q-ACC of 0.90 but I-ACC of only 0.49; (ii) data-related dimensions emerging as universal bottlenecks (e.g., Data Completeness: 0.21); and (iii) the challenge of achieving end-to-end correctness across all models. We release IGENBENCH at https://igen-bench.vercel.app/.
Authors: Fang Wu, Zhengyuan Zhou, Shuting Jin, Xiangxiang Zeng, Jure Leskovec, Jinbo Xu
Abstract: Therapeutic peptides show promise in targeting previously undruggable binding sites, with recent advancements in deep generative models enabling full-atom peptide co-design for specific protein receptors. However, the critical role of molecular surfaces in protein-protein interactions (PPIs) has been underexplored. To bridge this gap, we propose an omni-design peptides generation paradigm, called SurfFlow, a novel surface-based generative algorithm that enables comprehensive co-design of sequence, structure, and surface for peptides. SurfFlow employs a multi-modality conditional flow matching (CFM) architecture to learn distributions of surface geometries and biochemical properties, enhancing peptide binding accuracy. Evaluated on the comprehensive PepMerge benchmark, SurfFlow consistently outperforms full-atom baselines across all metrics. These results highlight the advantages of considering molecular surfaces in de novo peptide discovery and demonstrate the potential of integrating multiple protein modalities for more effective therapeutic peptide discovery.
Authors: Jacob Ede Levine, Yun Lyan Luo, Sai Chandra Kosaraju
Abstract: The design of reliable, valid, and diverse molecules is fundamental to modern drug discovery, as improved molecular generation supports efficient exploration of the chemical space for potential drug candidates and reduces the cost of early design efforts. Despite these needs, current chemical language models that generate molecules as SMILES strings are vulnerable to compounding token errors: many samples are unparseable or chemically implausible, and hard constraints meant to prevent failure can restrict exploration. To address this gap, we introduce TSSR, a Two-Stage, Swap-Reward-driven reinforcement learning (RL) framework for character-level SMILES generation. Stage one rewards local token swaps that repair syntax, promoting transitions from invalid to parseable strings. Stage two provides chemistry-aware feedback from RDKit diagnostics, rewarding reductions in valence, aromaticity, and connectivity issues. The reward decomposes into interpretable terms (swap efficiency, error reduction, distance to validity), is model agnostic, and requires no task-specific labels or hand-crafted grammars. We evaluated TSSR on the MOSES benchmark using a GRU policy trained with PPO in both pure RL (P-RL) from random initialization and fine-tuning RL (F-RL) starting from a pretrained chemical language model, assessing 10,000 generated SMILES per run. In P-RL, TSSR significantly improves syntactic validity, chemical validity, and novelty. In F-RL, TSSR preserves drug-likeness and synthesizability while increasing validity and novelty. Token-level analysis shows that syntax edits and chemistry fixes act jointly to reduce RDKit detected errors. TSSR converts a sparse terminal objective into a denser and more interpretable reward, improving both syntactic and chemical quality without reducing diversity. TSSR is dataset-agnostic and can be adapted to various reinforcement learning approaches.
Authors: Tianle Wang, Zhongyuan Wu, Shenghao Jin, Hao Xu, Wei Chen, Ning Miao
Abstract: Reinforcement learning with verifiable rewards (RLVR) has become a central component of large language model (LLM) post-training. Unlike supervised fine-tuning (SFT), RLVR lets an LLM generate multiple candidate solutions and reinforces those that lead to a verifiably correct final answer. However, in practice, RLVR often requires thousands of training steps to reach strong performance, incurring substantial computation largely attributed to prolonged exploration. In this work, we make a surprising observation: during RLVR, LLMs evolve in a strongly linear manner. Specifically, both model weights and model output log-probabilities exhibit strong linear correlations with RL training steps. This suggests that RLVR predominantly amplifies trends that emerge early in training, rather than continuously discovering new behaviors throughout the entire optimization trajectory. Motivated by this linearity, we investigate whether future model states can be predicted from intermediate checkpoints via extrapolation, avoiding continued expensive training. We show that Weight Extrapolation produces models with performance comparable to standard RL training while requiring significantly less computation. Moreover, Logits Extrapolation consistently outperforms continued RL training on all four benchmarks by extrapolating beyond the step range where RL training remains stable.
Authors: Mengmeng Zhu, Yuxuan Sun, Yukuan Jia, Wei Chen, Bo Ai, Sheng Zhou
Abstract: Collaborative perception (CP) is a critical technology in applications like autonomous driving and smart cities. It involves the sharing and fusion of information among sensors to overcome the limitations of individual perception, such as blind spots and range limitations. However, CP faces two primary challenges. First, due to the dynamic nature of the environment, the timeliness of the transmitted information is critical to perception performance. Second, with limited computational power at the sensors and constrained wireless bandwidth, the communication volume must be carefully designed to ensure feature representations are both effective and sufficient. This work studies the dynamic scheduling problem in a multi-region CP scenario, and presents a Timeliness-Aware Multi-region Prioritized (TAMP) scheduling algorithm to trade-off perception accuracy and communication resource usage. Timeliness reflects the utility of information that decays as time elapses, which is manifested by the perception performance in CP tasks. We propose an empirical penalty function that maps the joint impact of Age of Information (AoI) and communication volume to perception performance. Aiming to minimize this timeliness-oriented penalty in the long-term, and recognizing that scheduling decisions have a cumulative effect on subsequent system states, we propose the TAMP scheduling algorithm. TAMP is a Lyapunov-based optimization policy that decomposes the long-term average objective into a per-slot prioritization problem, balancing the scheduling worth against resource cost. We validate our algorithm in both intersection and corridor scenarios with the real-world Roadside Cooperative perception (RCooper) dataset. Extensive simulations demonstrate that TAMP outperforms the best-performing baseline, achieving an Average Precision (AP) improvement of up to 27% across various configurations.
Authors: Zhiyan Zhou, Junjie Liao, Manho Zhang, Yingyi Liao, Ziai Wang
Abstract: With the acceleration of urbanization, intelligent transportation systems have an increasing demand for accurate traffic flow prediction. This paper proposes a novel Graph Enhanced Spatio-temporal Hierarchical Inference Network (GEnSHIN) to handle the complex spatio-temporal dependencies in traffic flow prediction. The model integrates three innovative designs: 1) An attention-enhanced Graph Convolutional Recurrent Unit (GCRU), which strengthens the modeling capability for long-term temporal dependencies by introducing Transformer modules; 2) An asymmetric dual-embedding graph generation mechanism, which leverages the real road network and data-driven latent asymmetric topology to generate graph structures that better fit the characteristics of actual traffic flow; 3) A dynamic memory bank module, which utilizes learnable traffic pattern prototypes to provide personalized traffic pattern representations for each sensor node, and introduces a lightweight graph updater during the decoding phase to adapt to dynamic changes in road network states. Extensive experiments on the public dataset METR-LA show that GEnSHIN achieves or surpasses the performance of comparative models across multiple metrics such as Mean Absolute Error (MAE), Root Mean Square Error (RMSE), and Mean Absolute Percentage Error (MAPE). Notably, the model demonstrates excellent prediction stability during peak morning and evening traffic hours. Ablation experiments further validate the effectiveness of each core module and its contribution to the final performance.
Authors: Shogo Nakayama, Masahiro Okuda
Abstract: Conventional semi-supervised contrastive learning methods assign pseudo-labels only to samples whose highest predicted class probability exceeds a predefined threshold, and then perform supervised contrastive learning using those selected samples. In this study, we propose a novel loss function that estimates the confidence of each sample based on the entropy of its predicted probability distribution and applies confidence-based adaptive weighting. This approach enables pseudo-label assignment even to samples that were previously excluded from training and facilitates contrastive learning that accounts for the confidence of both anchor and positive samples in a more principled manner. Experimental results demonstrate that the proposed method improves classification accuracy and achieves more stable learning performance even under low-label conditions.
Authors: Paul Pu Liang
Abstract: Our experience of the world is multisensory, spanning a synthesis of language, sight, sound, touch, taste, and smell. Yet, artificial intelligence has primarily advanced in digital modalities like text, vision, and audio. This paper outlines a research vision for multisensory artificial intelligence over the next decade. This new set of technologies can change how humans and AI experience and interact with one another, by connecting AI to the human senses and a rich spectrum of signals from physiological and tactile cues on the body, to physical and social signals in homes, cities, and the environment. We outline how this field must advance through three interrelated themes of sensing, science, and synergy. Firstly, research in sensing should extend how AI captures the world in richer ways beyond the digital medium. Secondly, developing a principled science for quantifying multimodal heterogeneity and interactions, developing unified modeling architectures and representations, and understanding cross-modal transfer. Finally, we present new technical challenges to learn synergy between modalities and between humans and AI, covering multisensory integration, alignment, reasoning, generation, generalization, and experience. Accompanying this vision paper are a series of projects, resources, and demos of latest advances from the Multisensory Intelligence group at the MIT Media Lab, see https://mit-mi.github.io/.
Authors: Xiaowei Mao, Huihu Ding, Yan Lin, Tingrui Wu, Shengnan Guo, Dazhuo Qiu, Feiling Fang, Jilin Hu, Huaiyu Wan
Abstract: Imputing missing values in spatial-temporal traffic data is essential for intelligent transportation systems. Among advanced imputation methods, score-based diffusion models have demonstrated competitive performance. These models generate data by reversing a noising process, using observed values as conditional guidance. However, existing diffusion models typically apply a uniform guidance scale across both spatial and temporal dimensions, which is inadequate for nodes with high missing data rates. Sparse observations provide insufficient conditional guidance, causing the generative process to drift toward the learned prior distribution rather than closely following the conditional observations, resulting in suboptimal imputation performance. To address this, we propose FENCE, a spatial-temporal feedback diffusion guidance method designed to adaptively control guidance scales during imputation. First, FENCE introduces a dynamic feedback mechanism that adjusts the guidance scale based on the posterior likelihood approximations. The guidance scale is increased when generated values diverge from observations and reduced when alignment improves, preventing overcorrection. Second, because alignment to observations varies across nodes and denoising steps, a global guidance scale for all nodes is suboptimal. FENCE computes guidance scales at the cluster level by grouping nodes based on their attention scores, leveraging spatial-temporal correlations to provide more accurate guidance. Experimental results on real-world traffic datasets show that FENCE significantly enhances imputation accuracy.
Authors: Quang-Tu Pham, Hoang-Dieu Vu, Dinh-Dat Pham, Hieu H. Pham
Abstract: This paper introduces FedKDX, a federated learning framework that addresses limitations in healthcare AI through Negative Knowledge Distillation (NKD). Unlike existing approaches that focus solely on positive knowledge transfer, FedKDX captures both target and non-target information to improve model generalization in healthcare applications. The framework integrates multiple knowledge transfer techniques--including traditional knowledge distillation, contrastive learning, and NKD--within a unified architecture that maintains privacy while reducing communication costs. Through experiments on healthcare datasets (SLEEP, UCI-HAR, and PAMAP2), FedKDX demonstrates improved accuracy (up to 2.53% over state-of-the-art methods), faster convergence, and better performance on non-IID data distributions. Theoretical analysis supports NKD's contribution to addressing statistical heterogeneity in distributed healthcare data. The approach shows promise for privacy-sensitive medical applications under regulatory frameworks like HIPAA and GDPR, offering a balanced solution between performance and practical implementation requirements in decentralized healthcare settings. The code and model are available at https://github.com/phamdinhdat-ai/Fed_2024.
Authors: Joonwon Seo, Mariana Montiel
Abstract: Classical Recurrent Neural Networks (RNNs) summarize musical context into a deterministic hidden state vector, imposing an information bottleneck that fails to capture the inherent ambiguity in music. We propose the Density Matrix RNN (DM-RNN), a novel theoretical architecture utilizing the Density Matrix. This allows the model to maintain a statistical ensemble of musical interpretations (a mixed state), capturing both classical probabilities and quantum coherences. We rigorously define the temporal dynamics using Quantum Channels (CPTP maps). Crucially, we detail a parameterization strategy based on the Choi-Jamiolkowski isomorphism, ensuring the learned dynamics remain physically valid (CPTP) by construction. We introduce an analytical framework using Von Neumann Entropy to quantify musical uncertainty and Quantum Mutual Information (QMI) to measure entanglement between voices. The DM-RNN provides a mathematically rigorous framework for modeling complex, ambiguous musical structures.
Authors: Shuhan Zhang, Zhi Wang, Rui Gao, Shuang Li
Abstract: Modeling human decision-making is central to applications such as recommendation, preference learning, and human-AI alignment. While many classic models assume context-independent choice behavior, a large body of behavioral research shows that preferences are often influenced by the composition of the choice set itself -- a phenomenon known as the context effect or Halo effect. These effects can manifest as pairwise (first-order) or even higher-order interactions among the available alternatives. Recent models that attempt to capture such effects either focus on the featureless setting or, in the feature-based setting, rely on restrictive interaction structures or entangle interactions across all orders, which limits interpretability. In this work, we propose DeepHalo, a neural modeling framework that incorporates features while enabling explicit control over interaction order and principled interpretation of context effects. Our model enables systematic identification of interaction effects by order and serves as a universal approximator of context-dependent choice functions when specialized to a featureless setting. Experiments on synthetic and real-world datasets demonstrate strong predictive performance while providing greater transparency into the drivers of choice.
Authors: Akiyoshi Tomihari
Abstract: Reinforcement learning (RL) post-training is a critical stage in modern language model development, playing a key role in improving alignment and reasoning ability. However, several phenomena remain poorly understood, including the reduction in output diversity. To gain a broader understanding of RL post-training, we analyze the learning dynamics of RL post-training from a perspective that has been studied in supervised learning but remains underexplored in RL. We adopt an empirical neural tangent kernel (NTK) framework and decompose the NTK into two components to characterize how RL updates propagate across training samples. Our analysis reveals that limited variability in feature representations can cause RL updates to systematically increase model confidence, providing an explanation for the commonly observed reduction in output diversity after RL post-training. Furthermore, we show that effective learning in this regime depends on rapidly shaping the classifier, which directly affects the gradient component of the NTK. Motivated by these insights, we propose classifier-first reinforcement learning (CF-RL), a simple two-stage training strategy that prioritizes classifier updates before standard RL optimization. Experimental results validate our theoretical analysis by demonstrating increased model confidence and accelerated optimization under CF-RL. Additional analysis shows that the mechanism underlying CF-RL differs from that of linear-probing-then-fine-tuning in supervised learning. Overall, our study formalizes the learning dynamics of RL post-training and motivates further analysis and improvement.
Authors: Aurghya Maiti, Prateek Jain
Abstract: Estimating causal effects from observational data remains a fundamental challenge in causal inference, especially in the presence of latent confounders. This paper focuses on estimating causal effects in Gaussian Linear Structural Causal Models (GL-SCMs), which are widely used due to their analytical tractability. However, parameter estimation in GL-SCMs is often infeasible with finite data, primarily due to overparameterization. To address this, we introduce the class of Centralized Gaussian Linear SCMs (CGL-SCMs), a simplified yet expressive subclass where exogenous variables follow standardized distributions. We show that CGL-SCMs are equally expressive in terms of causal effect identifiability from observational distributions and present a novel EM-based estimation algorithm that can learn CGL-SCM parameters and estimate identifiable causal effects from finite observational samples. Our theoretical analysis is validated through experiments on synthetic data and benchmark causal graphs, demonstrating that the learned models accurately recover causal distributions.
Authors: Oluwatosin Oseni, Shengjie Wang, Jun Zhu, Micah Corah
Abstract: Reinforcement Learning (RL) has shown remarkable success in real-world applications, particularly in robotics control. However, RL adoption remains limited due to insufficient safety guarantees. We introduce Nightmare Dreamer, a model-based Safe RL algorithm that addresses safety concerns by leveraging a learned world model to predict potential safety violations and plan actions accordingly. Nightmare Dreamer achieves nearly zero safety violations while maximizing rewards. Nightmare Dreamer outperforms model-free baselines on Safety Gymnasium tasks using only image observations, achieving nearly a 20x improvement in efficiency.
Authors: Mir Rayat Imtiaz Hossain, Leo Feng, Leonid Sigal, Mohamed Osama Ahmed
Abstract: Large Language Models (LLMs) have emerged as promising recommendation systems, offering novel ways to model user preferences through generative approaches. However, many existing methods often rely solely on text semantics or incorporate collaborative signals in a limited manner, typically using only user or item embeddings. These methods struggle to handle multiple item embeddings representing user history, reverting to textual semantics and neglecting richer collaborative information. In this work, we propose a simple yet effective solution that projects user and item embeddings, learned from collaborative filtering, into the LLM token space via separate lightweight projector modules. A finetuned LLM then conditions on these projected embeddings alongside textual tokens to generate recommendations. Preliminary results show that this design effectively leverages structured user-item interaction data, improves recommendation performance over text-only LLM baselines, and offers a practical path for bridging traditional recommendation systems with modern LLMs.
Authors: \`Angel Ruiz-Fas, Carlos Granell, Jos\'e Francisco Ramos, Joaqu\'in Huerta, Sergio Trilles
Abstract: Rapid e-commerce growth has pushed last-mile delivery networks to their limits, where small routing gains translate into lower costs, faster service, and fewer emissions. Classical heuristics struggle to adapt when travel times are highly asymmetric (e.g., one-way streets, congestion). A deep learning-based approach to the last-mile routing problem is presented to generate geographical zones composed of stop sequences to minimize last-mile delivery times. The presented approach is an encoder-decoder architecture. Each route is represented as a complete directed graph whose nodes are stops and whose edge weights are asymmetric travel times. A Graph Neural Network encoder produces node embeddings that captures the spatial relationships between stops. A Pointer Network decoder then takes the embeddings and the route's start node to sequentially select the next stops, assigning a probability to each unvisited node as the next destination. Cells of a Discrete Global Grid System which contain route stops in the training data are obtained and clustered to generate geographical zones of similar size in which the process of training and inference are divided. Subsequently, a different instance of the model is trained per zone only considering the stops of the training routes which are included in that zone. This approach is evaluated using the Los Angeles routes from the 2021 Amazon Last Mile Routing Challenge. Results from general and zone-based training are compared, showing a reduction in the average predicted route length in the zone-based training compared to the general training. The performance improvement of the zone-based approach becomes more pronounced as the number of stops per route increases.
Authors: Irfan Ullah, Young-Koo Lee
Abstract: Graph Neural Networks (GNNs) are powerful tools for learning graph-structured data, but their scalability is hindered by inefficient mini-batch generation, data transfer bottlenecks, and costly inter-GPU synchronization. Existing training frameworks fail to overlap these stages, leading to suboptimal resource utilization. This paper proposes MQ-GNN, a multi-queue pipelined framework that maximizes training efficiency by interleaving GNN training stages and optimizing resource utilization. MQ-GNN introduces Ready-to-Update Asynchronous Consistent Model (RaCoM), which enables asynchronous gradient sharing and model updates while ensuring global consistency through adaptive periodic synchronization. Additionally, it employs global neighbor sampling with caching to reduce data transfer overhead and an adaptive queue-sizing strategy to balance computation and memory efficiency. Experiments on four large-scale datasets and ten baseline models demonstrate that MQ-GNN achieves up to \boldmath $\bm{4.6\,\times}$ faster training time and 30% improved GPU utilization while maintaining competitive accuracy. These results establish MQ-GNN as a scalable and efficient solution for multi-GPU GNN training.
Authors: Maanas Taneja, Purab Shingvi
Abstract: The key-value (KV) cache in large language models presents a significant memory bottleneck during inference, growing linearly with sequence length and often exceeding the memory footprint of model weights themselves. We implement and evaluate GPU-accelerated INT8 quantization for KV cache compression, achieving 4$\times$ memory reduction with minimal accuracy degradation. We develop four CUDA kernel variants -- naive, tiled, coarsened, and vectorized -- and benchmark them across realistic workload sizes up to 1 billion elements. Our vectorized kernel achieves up to 1,694$\times$ speedup over CPU baselines while maintaining reconstruction error below 0.004 and attention score error below 0.1 even for 8K-dimensional heads. These results demonstrate that INT8 quantization provides a practical approach for reducing memory pressure in LLM inference with negligible computational overhead (6--58ms) and minimal impact on downstream model behavior
Authors: Elizabeth Donoway, Hailey Joren, Fabien Roger, Jan Leike
Abstract: Understanding whether fine-tuning elicits latent capabilities or teaches new ones is a fundamental question for language model evaluation and safety. We develop a formal information-theoretic framework for quantifying how much predictive structure fine-tuning extracts from the train dataset and writes into a model's parameters. Our central quantity, Excess Description Length (EDL), is defined via prequential coding and measures the gap between the bits required to encode training labels sequentially using an evolving model (trained online) and the residual encoding cost under the final trained model. We establish that EDL is non-negative in expectation, converges to surplus description length in the infinite-data limit, and provides bounds on expected generalization gain. Through a series of toy models, we clarify common confusions about information in learning: why random labels yield EDL near zero, how a single example can eliminate many bits of uncertainty about the underlying rule(s) that describe the data distribution, why structure learned on rare inputs contributes proportionally little to expected generalization, and how format learning creates early transients distinct from capability acquisition. This framework provides rigorous foundations for the empirical observation that capability elicitation and teaching exhibit qualitatively distinct scaling signatures.
Authors: Kota Nakamura, Koki Kawabata, Yasuko Matsubara, Yasushi Sakurai
Abstract: Given real-time sensor data streams obtained from machines, how can we continuously predict when a machine failure will occur? This work aims to continuously forecast the timing of future events by analyzing multi-sensor data streams. A key characteristic of real-world data streams is their dynamic nature, where the underlying patterns evolve over time. To address this, we present TimeCast, a dynamic prediction framework designed to adapt to these changes and provide accurate, real-time predictions of future event time. Our proposed method has the following properties: (a) Dynamic: it identifies the distinct time-evolving patterns (i.e., stages) and learns individual models for each, enabling us to make adaptive predictions based on pattern shifts. (b) Practical: it finds meaningful stages that capture time-varying interdependencies between multiple sensors and improve prediction performance; (c) Scalable: our algorithm scales linearly with the input size and enables online model updates on data streams. Extensive experiments on real datasets demonstrate that TimeCast provides higher prediction accuracy than state-of-the-art methods while finding dynamic changes in data streams with a great reduction in computational time.
Authors: Luca Lanzilao, Angela Meyer
Abstract: We present a novel framework for spatiotemporal photovoltaic (PV) power forecasting and use it to evaluate the reliability, sharpness, and overall performance of seven intraday PV power nowcasting models. The model suite includes satellite-based deep learning and optical-flow approaches and physics-based numerical weather prediction models, covering both deterministic and probabilistic formulations. Forecasts are first validated against satellite-derived surface solar irradiance (SSI). Irradiance fields are then converted into PV power using station-specific machine learning models, enabling comparison with production data from 6434 PV stations across Switzerland. To our knowledge, this is the first study to investigate spatiotemporal PV forecasting at a national scale. We additionally provide the first visualizations of how mesoscale cloud systems shape national PV production on hourly and sub-hourly timescales. Our results show that satellite-based approaches outperform the Integrated Forecast System (IFS-ENS), particularly at short lead times. Among them, SolarSTEPS and SHADECast deliver the most accurate SSI and PV power predictions, with SHADECast providing the most reliable ensemble spread. The deterministic model IrradianceNet achieves the lowest root mean square error, while probabilistic forecasts of SolarSTEPS and SHADECast provide better-calibrated uncertainty. Forecast skill generally decreases with elevation. At a national scale, satellite-based models forecast the daily total PV generation with relative errors below 10% for 82% of the days in 2019-2020, demonstrating robustness and their potential for operational use.
Authors: Rupsa Rani Mishra, D. Chandrasekhar Rao, Ajaya Kumar Tripathy
Abstract: Manual observation and monitoring of individual cows for disease detection present significant challenges in large-scale farming operations, as the process is labor-intensive, time-consuming, and prone to reduced accuracy. The reliance on human observation often leads to delays in identifying symptoms, as the sheer number of animals can hinder timely attention to each cow. Consequently, the accuracy and precision of disease detection are significantly compromised, potentially affecting animal health and overall farm productivity. Furthermore, organizing and managing human resources for the manual observation and monitoring of cow health is a complex and economically demanding task. It necessitates the involvement of skilled personnel, thereby contributing to elevated farm maintenance costs and operational inefficiencies. Therefore, the development of an automated, low-cost, and reliable smart system is essential to address these challenges effectively. Although several studies have been conducted in this domain, very few have simultaneously considered the detection of multiple common diseases with high prediction accuracy. However, advancements in Internet of Things (IoT), Machine Learning (ML), and Cyber-Physical Systems have enabled the automation of cow health monitoring with enhanced accuracy and reduced operational costs. This study proposes an IoT-enabled Cyber-Physical System framework designed to monitor the daily activities and health status of cow. A novel ML algorithm is proposed for the diagnosis of common cow diseases using collected physiological and behavioral data. The algorithm is designed to predict multiple diseases by analyzing a comprehensive set of recorded physiological and behavioral features, enabling accurate and efficient health assessment.
Authors: Lang Feng, Fuchao Yang, Feng Chen, Xin Cheng, Haiyang Xu, Zhenglin Wan, Ming Yan, Bo An
Abstract: Recent advances in large language models (LLMs) enable agentic systems trained with reinforcement learning (RL) over multi-turn interaction trajectories, but practical deployment is bottlenecked by rapidly growing textual histories that inflate token budgets and memory usage. We introduce AgentOCR, a framework that exploits the superior information density of visual tokens by representing the accumulated observation-action history as a compact rendered image. To make multi-turn rollouts scalable, AgentOCR proposes segment optical caching. By decomposing history into hashable segments and maintaining a visual cache, this mechanism eliminates redundant re-rendering. Beyond fixed rendering, AgentOCR introduces agentic self-compression, where the agent actively emits a compression rate and is trained with compression-aware reward to adaptively balance task success and token efficiency. We conduct extensive experiments on challenging agentic benchmarks, ALFWorld and search-based QA. Remarkably, results demonstrate that AgentOCR preserves over 95\% of text-based agent performance while substantially reducing token consumption (>50\%), yielding consistent token and memory efficiency. Our further analysis validates a 20x rendering speedup from segment optical caching and the effective strategic balancing of self-compression.
Authors: Marios Thoma, Vassilis Vassiliades, Loizos Michael
Abstract: Neural-Symbolic (NeSy) Artificial Intelligence has emerged as a promising approach for combining the learning capabilities of neural networks with the interpretable reasoning of symbolic systems. However, existing NeSy frameworks typically require either predefined symbolic policies or policies that are differentiable, limiting their applicability when domain expertise is unavailable or when policies are inherently non-differentiable. We propose a framework that addresses this limitation by enabling the concurrent learning of both non-differentiable symbolic policies and neural network weights through an evolutionary process. Our approach casts NeSy systems as organisms in a population that evolve through mutations (both symbolic rule additions and neural weight changes), with fitness-based selection guiding convergence toward hidden target policies. The framework extends the NEUROLOG architecture to make symbolic policies trainable, adapts Valiant's Evolvability framework to the NeSy context, and employs Machine Coaching semantics for mutable symbolic representations. Neural networks are trained through abductive reasoning from the symbolic component, eliminating differentiability requirements. Through extensive experimentation, we demonstrate that NeSy systems starting with empty policies and random neural weights can successfully approximate hidden non-differentiable target policies, achieving median correct performance approaching 100%. This work represents a step toward enabling NeSy research in domains where the acquisition of symbolic knowledge from experts is challenging or infeasible.
Authors: Oscar Llorente, Jaime Boal, Eugenio F. S\'anchez-\'Ubeda, Antonio Diaz-Cano, Miguel Familiar
Abstract: Graph Neural Networks (GNNs) have demonstrated remarkable performance in a wide range of tasks, such as node classification, link prediction, and graph classification, by exploiting the structural information in graph-structured data. However, in node classification, computing node-level explainability becomes extremely time-consuming as the size of the graph increases, while batching strategies often degrade explanation quality. This paper introduces a novel approach to parallelizing node-level explainability in GNNs through graph partitioning. By decomposing the graph into disjoint subgraphs, we enable parallel computation of explainability for node neighbors, significantly improving the scalability and efficiency without affecting the correctness of the results, provided sufficient memory is available. For scenarios where memory is limited, we further propose a dropout-based reconstruction mechanism that offers a controllable trade-off between memory usage and explanation fidelity. Experimental results on real-world datasets demonstrate substantial speedups, enabling scalable and transparent explainability for large-scale GNN models.
Authors: Francesco Ferrini, Veronica Lachi, Antonio Longa, Bruno Lepri, Matono Akiyoshi, Andrea Passerini, Xin Liu, Manfred Jaeger
Abstract: Handling missing node features is a key challenge for deploying Graph Neural Networks (GNNs) in real-world domains such as healthcare and sensor networks. Existing studies mostly address relatively benign scenarios, namely benchmark datasets with (a) high-dimensional but sparse node features and (b) incomplete data generated under Missing Completely At Random (MCAR) mechanisms. For (a), we theoretically prove that high sparsity substantially limits the information loss caused by missingness, making all models appear robust and preventing a meaningful comparison of their performance. To overcome this limitation, we introduce one synthetic and three real-world datasets with dense, semantically meaningful features. For (b), we move beyond MCAR and design evaluation protocols with more realistic missingness mechanisms. Moreover, we provide a theoretical background to state explicit assumptions on the missingness process and analyze their implications for different methods. Building on this analysis, we propose GNNmim, a simple yet effective baseline for node classification with incomplete feature data. Experiments show that GNNmim is competitive with respect to specialized architectures across diverse datasets and missingness regimes.
Authors: Elisa Roldan, Kirstie Andrews, Stephen M. Richardson, Reyhaneh Fatahian, Glen Cooper, Rasool Erfani, Tasneem Sabir, Neil D. Reeves
Abstract: Electrospinning is a scalable technique for producing fibrous scaffolds with tunable micro- and nanoscale architectures for applications in tissue engineering, drug delivery, and wound care. While machine learning (ML) has been used to support electrospinning process optimisation, most existing approaches predict only mean fibre diameters, neglecting the full diameter distribution that governs scaffold performance. This work presents FibreCastML, an open, distribution-aware ML framework that predicts complete fibre diameter spectra from routinely reported electrospinning parameters and provides interpretable insights into process structure relationships. A meta-dataset comprising 68538 individual fibre diameter measurements extracted from 1778 studies across 16 biomedical polymers was curated. Six standard processing parameters, namely solution concentration, applied voltage, flow rate, tip to collector distance, needle diameter, and collector rotation speed, were used to train seven ML models using nested cross validation with leave one study out external folds. Model interpretability was achieved using variable importance analysis, SHapley Additive exPlanations, correlation matrices, and three dimensional parameter maps. Non linear models consistently outperformed linear baselines, achieving coefficients of determination above 0.91 for several widely used polymers. Solution concentration emerged as the dominant global driver of fibre diameter distributions. Experimental validation across different electrospinning systems demonstrated close agreement between predicted and measured distributions. FibreCastML enables more reproducible and data driven optimisation of electrospun scaffold architectures.
Authors: Maksim Velikanov, Ilyas Chahed, Jingwei Zuo, Dhia Eddine Rhaiem, Younes Belkada, Hakim Hacid
Abstract: Applying weight decay (WD) to matrix layers is standard practice in large-language-model pretraining. Prior work suggests that stochastic gradient noise induces a Brownian-like expansion of the weight matrices W, whose growth is counteracted by WD, leading to a WD-noise equilibrium with a certain weight norm ||W||. In this work, we view the equilibrium norm as a harmful artifact of the training procedure, and address it by introducing learnable multipliers to learn the optimal scale. First, we attach a learnable scalar multiplier to W and confirm that the WD-noise equilibrium norm is suboptimal: the learned scale adapts to data and improves performance. We then argue that individual row and column norms are similarly constrained, and free their scale by introducing learnable per-row and per-column multipliers. Our method can be viewed as a learnable, more expressive generalization of muP multipliers. It outperforms a well-tuned muP baseline, reduces the computational overhead of multiplier tuning, and surfaces practical questions such as forward-pass symmetries and the width-scaling of the learned multipliers. Finally, we validate learnable multipliers with both Adam and Muon optimizers, where it shows improvement in downstream evaluations matching the improvement of the switching from Adam to Muon.
Authors: Sifan Yang, Wenhao Yang, Wei Jiang, Lijun Zhang
Abstract: We investigate distributed online convex optimization with compressed communication, where $n$ learners connected by a network collaboratively minimize a sequence of global loss functions using only local information and compressed data from neighbors. Prior work has established regret bounds of $O(\max\{\omega^{-2}\rho^{-4}n^{1/2},\omega^{-4}\rho^{-8}\}n\sqrt{T})$ and $O(\max\{\omega^{-2}\rho^{-4}n^{1/2},\omega^{-4}\rho^{-8}\}n\ln{T})$ for convex and strongly convex functions, respectively, where $\omega\in(0,1]$ is the compression quality factor ($\omega=1$ means no compression) and $\rho<1$ is the spectral gap of the communication matrix. However, these regret bounds suffer from a \emph{quadratic} or even \emph{quartic} dependence on $\omega^{-1}$. Moreover, the \emph{super-linear} dependence on $n$ is also undesirable. To overcome these limitations, we propose a novel algorithm that achieves improved regret bounds of $\tilde{O}(\omega^{-1/2}\rho^{-1}n\sqrt{T})$ and $\tilde{O}(\omega^{-1}\rho^{-2}n\ln{T})$ for convex and strongly convex functions, respectively. The primary idea is to design a \emph{two-level blocking update framework} incorporating two novel ingredients: an online gossip strategy and an error compensation scheme, which collaborate to \emph{achieve a better consensus} among learners. Furthermore, we establish the first lower bounds for this problem, justifying the optimality of our results with respect to both $\omega$ and $T$. Additionally, we consider the bandit feedback scenario, and extend our method with the classic gradient estimators to enhance existing regret bounds.
Authors: Miguel O'Malley
Abstract: Class imbalance is a common and pernicious issue for the training of neural networks. Often, an imbalanced majority class can dominate training to skew classifier performance towards the majority outcome. To address this problem we introduce cardinality augmented loss functions, derived from cardinality-like invariants in modern mathematics literature such as magnitude and the spread. These invariants enrich the concept of cardinality by evaluating the `effective diversity' of a metric space, and as such represent a natural solution to overly homogeneous training data. In this work, we establish a methodology for applying cardinality augmented loss functions in the training of neural networks and report results on both artificially imbalanced datasets as well as a real-world imbalanced material science dataset. We observe significant performance improvement among minority classes, as well as improvement in overall performance metrics.
Authors: Yirong Zeng, Yufei Liu, Xiao Ding, Yutai Hou, Yuxian Wang, Haonan Song, Wu Ning, Dandan Tu, Qixun Zhang, Bibo Cai, Yuxiang He, Ting Liu
Abstract: A central belief in scaling reinforcement learning with verifiable rewards for instruction following (IF) tasks is that, a diverse mixture of verifiable hard and unverifiable soft constraints is essential for generalizing to unseen instructions. In this work, we challenge this prevailing consensus through a systematic empirical investigation. Counter-intuitively, we find that models trained on hard-only constraints consistently outperform those trained on mixed datasets. Extensive experiments reveal that reward precision, rather than constraint diversity, is the primary driver of effective alignment. The LLM judge suffers from a low recall rate in detecting false response, which leads to severe reward hacking, thereby undermining the benefits of diversity. Furthermore, analysis of the attention mechanism reveals that high-precision rewards develop a transferable meta-skill for IF. Motivated by these insights, we propose a simple yet effective data-centric refinement strategy that prioritizes reward precision. Evaluated on five benchmarks, our approach outperforms competitive baselines by 13.4\% in performance while achieving a 58\% reduction in training time, maintaining strong generalization beyond instruction following. Our findings advocate for a paradigm shift: moving away from the indiscriminate pursuit of data diversity toward high-precision rewards.
Authors: James Hinns, Sofie Goethals, Stephan Van der Veeken, Theodoros Evgeniou, David Martens
Abstract: Counterfactual explanations are widely used to communicate how inputs must change for a model to alter its prediction. For a single instance, many valid counterfactuals can exist, which leaves open the possibility for an explanation provider to cherry-pick explanations that better suit a narrative of their choice, highlighting favourable behaviour and withholding examples that reveal problematic behaviour. We formally define cherry-picking for counterfactual explanations in terms of an admissible explanation space, specified by the generation procedure, and a utility function. We then study to what extent an external auditor can detect such manipulation. Considering three levels of access to the explanation process: full procedural access, partial procedural access, and explanation-only access, we show that detection is extremely limited in practice. Even with full procedural access, cherry-picked explanations can remain difficult to distinguish from non cherry-picked explanations, because the multiplicity of valid counterfactuals and flexibility in the explanation specification provide sufficient degrees of freedom to mask deliberate selection. Empirically, we demonstrate that this variability often exceeds the effect of cherry-picking on standard counterfactual quality metrics such as proximity, plausibility, and sparsity, making cherry-picked explanations statistically indistinguishable from baseline explanations. We argue that safeguards should therefore prioritise reproducibility, standardisation, and procedural constraints over post-hoc detection, and we provide recommendations for algorithm developers, explanation providers, and auditors.
Authors: Aleksandar Fontana, Marco Simoni, Giulio Rossolini, Andrea Saracino, Paolo Mori
Abstract: Group-based reinforcement learning methods, like Group Relative Policy Optimization (GRPO), are widely used nowadays to post-train large language models. Despite their empirical success, they exhibit structural mismatches between reward optimization and the underlying training objective. In this paper, we present a theoretical analysis of GRPO style methods by studying them within a unified surrogate formulation. This perspective reveals recurring properties that affect all the methods under analysis: (i) non-uniform group weighting induces systematic gradient biases on shared prefix tokens; (ii) interactions with the AdamW optimizer make training dynamics largely insensitive to reward scaling; and (iii) optimizer momentum can push policy updates beyond the intended clipping region under repeated optimization steps. We believe that these findings highlight fundamental limitations of current approaches and provide principled guidance for the design of future formulations.
Authors: Xiaopeng Luo, Zexi Tan, Zhuowei Wang
Abstract: Missing value imputation is a fundamental challenge in machine intelligence, heavily dependent on data completeness. Current imputation methods often handle numerical and categorical attributes independently, overlooking critical interdependencies among heterogeneous features. To address these limitations, we propose a novel imputation approach that explicitly models cross-type feature dependencies within a unified framework. Our method leverages both complete and incomplete instances to ensure accurate and consistent imputation in tabular data. Extensive experimental results demonstrate that the proposed approach achieves superior performance over existing techniques and significantly enhances downstream machine learning tasks, providing a robust solution for real-world systems with missing data.
Authors: Torben Berndt, Jan St\"uhmer
Abstract: Equivariance is a powerful inductive bias in neural networks, improving generalisation and physical consistency. Recently, however, non-equivariant models have regained attention, due to their better runtime performance and imperfect symmetries that might arise in real-world applications. This has motivated the development of approximately equivariant models that strike a middle ground between respecting symmetries and fitting the data distribution. Existing approaches in this field usually apply sample-based regularisers which depend on data augmentation at training time, incurring a high sample complexity, in particular for continuous groups such as $SO(3)$. This work instead approaches approximate equivariance via a projection-based regulariser which leverages the orthogonal decomposition of linear layers into equivariant and non-equivariant components. In contrast to existing methods, this penalises non-equivariance at an operator level across the full group orbit, rather than point-wise. We present a mathematical framework for computing the non-equivariance penalty exactly and efficiently in both the spatial and spectral domain. In our experiments, our method consistently outperforms prior approximate equivariance approaches in both model performance and efficiency, achieving substantial runtime gains over sample-based regularisers.
Authors: Anees Fatima, Mohammad Abdus Salam
Abstract: Demand forecasting in supply chain management (SCM) is critical for optimizing inventory, reducing waste, and improving customer satisfaction. Conventional approaches frequently neglect external influences like weather, festivities, and equipment breakdowns, resulting in inefficiencies. This research investigates the use of machine learning (ML) algorithms to improve demand prediction in retail and vending machine sectors. Four machine learning algorithms. Extreme Gradient Boosting (XGBoost), Autoregressive Integrated Moving Average (ARIMA), Facebook Prophet (Fb Prophet), and Support Vector Regression (SVR) were used to forecast inventory requirements. Ex-ternal factors like weekdays, holidays, and sales deviation indicators were methodically incorporated to enhance precision. XGBoost surpassed other models, reaching the lowest Mean Absolute Error (MAE) of 22.7 with the inclusion of external variables. ARIMAX and Fb Prophet demonstrated noteworthy enhancements, whereas SVR fell short in performance. Incorporating external factors greatly improves the precision of demand forecasting models, and XGBoost is identified as the most efficient algorithm. This study offers a strong framework for enhancing inventory management in retail and vending machine systems.
Authors: Saumya Gupta, Scott Biggs, Moritz Laber, Zohair Shafi, Robin Walters, Ayan Paul
Abstract: Building efficient and effective generative models for neural network weights has been a research focus of significant interest that faces challenges posed by the high-dimensional weight spaces of modern neural networks and their symmetries. Several prior generative models are limited to generating partial neural network weights, particularly for larger models, such as ResNet and ViT. Those that do generate complete weights struggle with generation speed or require finetuning of the generated models. In this work, we present DeepWeightFlow, a Flow Matching model that operates directly in weight space to generate diverse and high-accuracy neural network weights for a variety of architectures, neural network sizes, and data modalities. The neural networks generated by DeepWeightFlow do not require fine-tuning to perform well and can scale to large networks. We apply Git Re-Basin and TransFusion for neural network canonicalization in the context of generative weight models to account for the impact of neural network permutation symmetries and to improve generation efficiency for larger model sizes. The generated networks excel at transfer learning, and ensembles of hundreds of neural networks can be generated in minutes, far exceeding the efficiency of diffusion-based methods. DeepWeightFlow models pave the way for more efficient and scalable generation of diverse sets of neural networks.
Authors: Jianlong Chen, Daocheng Fu, Shengze Xu, Jiawei Chen, Yuan Feng, Yue Yang, Junchi Yan, Hongyuan Zha, Renqiu Xia
Abstract: Multimodal Large Language Models (MLLMs) struggle with complex geometric reasoning, largely because "black box" outcome-based supervision fails to distinguish between lucky guesses and rigorous deduction. To address this, we introduce a paradigm shift towards subgoal-level evaluation and learning. We first construct GeoGoal, a benchmark synthesized via a rigorous formal verification data engine, which converts abstract proofs into verifiable numeric subgoals. This structure reveals a critical divergence between reasoning quality and outcome accuracy. Leveraging this, we propose the Sub-Goal Verifiable Reward (SGVR) framework, which replaces sparse signals with dense rewards based on the Skeleton Rate. Experiments demonstrate that SGVR not only enhances geometric performance (+9.7%) but also exhibits strong generalization, transferring gains to general math (+8.0%) and other general reasoning tasks (+2.8%), demonstrating broad applicability across diverse domains.
Authors: Hayk Asatryan, Basile Tousside, Janis Mohr, Malte Neugebauer, Hildo Bijl, Paul Spiegelberg, Claudia Frohn-Schauf, J\"org Frochte
Abstract: Learning Analytics (LA) is nowadays ubiquitous in many educational systems, providing the ability to collect and analyze student data in order to understand and optimize learning and the environments in which it occurs. On the other hand, the collection of data requires to comply with the growing demand regarding privacy legislation. In this paper, we use the Student Expectation of Learning Analytics Questionnaire (SELAQ) to analyze the expectations and confidence of students from different faculties regarding the processing of their data for Learning Analytics purposes. This allows us to identify four clusters of students through clustering algorithms: Enthusiasts, Realists, Cautious and Indifferents. This structured analysis provides valuable insights into the acceptance and criticism of Learning Analytics among students.
Authors: Polina Dolgova, Sebastian U. Stich
Abstract: Certified unlearning based on differential privacy offers strong guarantees but remains largely impractical: the noisy fine-tuning approaches proposed so far achieve these guarantees but severely reduce model accuracy. We propose sequential noise scheduling, which distributes the noise budget across orthogonal subspaces of the parameter space, rather than injecting it all at once. This simple modification mitigates the destructive effect of noise while preserving the original certification guarantees. We extend the analysis of noisy fine-tuning to the subspace setting, proving that the same $(\varepsilon,\delta)$ privacy budget is retained. Empirical results on image classification benchmarks show that our approach substantially improves accuracy after unlearning while remaining robust to membership inference attacks. These results show that certified unlearning can achieve both rigorous guarantees and practical utility.
Authors: Timofey Tomashevskiy
Abstract: This work provides a state-of-the-art survey of continual safe online reinforcement learning (COSRL) methods. We discuss theoretical aspects, challenges, and open questions in building continual online safe reinforcement learning algorithms. We provide the taxonomy and the details of continual online safe reinforcement learning methods based on the type of safe learning mechanism that takes adaptation to nonstationarity into account. We categorize safety constraints formulation for online reinforcement learning algorithms, and finally, we discuss prospects for creating reliable, safe online learning algorithms. Keywords: safe RL in nonstationary environments, safe continual reinforcement learning under nonstationarity, HM-MDP, NSMDP, POMDP, safe POMDP, constraints for continual learning, safe continual reinforcement learning review, safe continual reinforcement learning survey, safe continual reinforcement learning, safe online learning under distribution shift, safe continual online adaptation, safe reinforcement learning, safe exploration, safe adaptation, constrained Markov decision processes, safe reinforcement learning, partially observable Markov decision process, safe reinforcement learning and hidden Markov decision processes, Safe Online Reinforcement Learning, safe online reinforcement learning, safe online reinforcement learning, safe meta-learning, safe meta-reinforcement learning, safe context-based reinforcement learning, formulating safety constraints for continual learning
Authors: Yiji Zhao, Zihao Zhong, Ao Wang, Haomin Wen, Ming Jin, Yuxuan Liang, Huaiyu Wan, Hao Wu
Abstract: Spatial-Temporal Graph (STG) forecasting on large-scale networks has garnered significant attention. However, existing models predominantly focus on short-horizon predictions and suffer from notorious computational costs and memory consumption when scaling to long-horizon predictions and large graphs. Targeting the above challenges, we present FaST, an effective and efficient framework based on heterogeneity-aware Mixture-of-Experts (MoEs) for long-horizon and large-scale STG forecasting, which unlocks one-week-ahead (672 steps at a 15-minute granularity) prediction with thousands of nodes. FaST is underpinned by two key innovations. First, an adaptive graph agent attention mechanism is proposed to alleviate the computational burden inherent in conventional graph convolution and self-attention modules when applied to large-scale graphs. Second, we propose a new parallel MoE module that replaces traditional feed-forward networks with Gated Linear Units (GLUs), enabling an efficient and scalable parallel structure. Extensive experiments on real-world datasets demonstrate that FaST not only delivers superior long-horizon predictive accuracy but also achieves remarkable computational efficiency compared to state-of-the-art baselines. Our source code is available at: https://github.com/yijizhao/FaST.
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: Zain Iqbal, Lorenzo Valerio
Abstract: Pervasive AI increasingly depends on on-device learning systems that deliver low-latency and energy-efficient computation under strict resource constraints. Liquid State Machines (LSMs) offer a promising approach for low-power temporal processing in pervasive and neuromorphic systems, but their deployment remains challenging due to high hyperparameter sensitivity and the computational cost of traditional optimization methods that ignore energy constraints. This work presents EARL, an energy-aware reinforcement learning framework that integrates Bayesian optimization with an adaptive reinforcement learning based selection policy to jointly optimize accuracy and energy consumption. EARL employs surrogate modeling for global exploration, reinforcement learning for dynamic candidate prioritization, and an early termination mechanism to eliminate redundant evaluations, substantially reducing computational overhead. Experiments on three benchmark datasets demonstrate that EARL achieves 6 to 15 percent higher accuracy, 60 to 80 percent lower energy consumption, and up to an order of magnitude reduction in optimization time compared to leading hyperparameter tuning frameworks. These results highlight the effectiveness of energy-aware adaptive search in improving the efficiency and scalability of LSMs for resource-constrained on-device AI applications.
Authors: Ilmo Sung
Abstract: Large language models suffer from "hallucinations"-logical inconsistencies induced by semantic noise. We propose that current architectures operate in a "Metric Phase," where causal order is vulnerable to spontaneous symmetry breaking. Here, we identify robust inference as an effective Symmetry-Protected Topological phase, where logical operations are formally isomorphic to non-Abelian anyon braiding, replacing fragile geometric interpolation with robust topological invariants. Empirically, we demonstrate a sharp topological phase transition: while Transformers and RNNs exhibit gapless decay, our Holonomic Network reveals a macroscopic "mass gap," maintaining invariant fidelity below a critical noise threshold. Furthermore, in a variable-binding task on $S_{10}$ ($3.6 \times 10^6$ states) representing symbolic manipulation, we demonstrate holonomic generalization: the topological model maintains perfect fidelity extrapolating $100\times$ beyond training ($L=50 \to 5000$), consistent with a theoretically indefinite causal horizon, whereas Transformers lose logical coherence. Ablation studies indicate this protection emerges strictly from non-Abelian gauge symmetry. This provides strong evidence for a new universality class for logical reasoning, linking causal stability to the topology of the semantic manifold.
Authors: Natalie Collina, Jiuyao Lu, Georgy Noarov, Aaron Roth
Abstract: We prove tight lower bounds for online multicalibration, establishing an information-theoretic separation from marginal calibration. In the general setting where group functions can depend on both context and the learner's predictions, we prove an $\Omega(T^{2/3})$ lower bound on expected multicalibration error using just three disjoint binary groups. This matches the upper bounds of Noarov et al. (2025) up to logarithmic factors and exceeds the $O(T^{2/3-\varepsilon})$ upper bound for marginal calibration (Dagan et al., 2025), thereby separating the two problems. We then turn to lower bounds for the more difficult case of group functions that may depend on context but not on the learner's predictions. In this case, we establish an $\widetilde{\Omega}(T^{2/3})$ lower bound for online multicalibration via a $\Theta(T)$-sized group family constructed using orthogonal function systems, again matching upper bounds up to logarithmic factors.
Authors: Anas Ezzakri, Nicola Piovesan, Mohamed Sana, Antonio De Domenico, Fadhel Ayed, Haozhe Zhang
Abstract: Language Models (LLMs) are increasingly explored in the telecom industry to support engineering tasks, accelerate troubleshooting, and assist in interpreting complex technical documents. However, recent studies show that LLMs perform poorly on telecom standards, particularly 3GPP specifications. We argue that a key reason is that these standards densely include tables to present essential information, yet the LLM knowledge and interpretation ability of such tables remains largely unexamined. To address this gap, we introduce TeleTables, a benchmark designed to evaluate both the implicit knowledge LLMs have about tables in technical specifications and their explicit ability to interpret them. TeleTables is built through a novel multi-stage data generation pipeline that extracts tables from 3GPP standards and uses multimodal and reasoning-oriented LLMs to generate and validate questions. The resulting dataset, which is publicly available, comprises 500 human-verified question-answer pairs, each associated with the corresponding table in multiple formats. Our evaluation shows that, smaller models (under 10B parameters) struggle both to recall 3GPP knowledge and to interpret tables, indicating the limited exposure to telecom standards in their pretraining and the insufficient inductive biases for navigating complex technical material. Larger models, on the other hand, show stronger reasoning on table interpretation. Overall, TeleTables highlights the need for domain-specialized fine-tuning to reliably interpret and reason over telecom standards.
Authors: Xueqing Wu, Zihan Xue, Da Yin, Shuyan Zhou, Kai-Wei Chang, Nanyun Peng, Yeming Wen
Abstract: We present FronTalk, a benchmark for front-end code generation that pioneers the study of a unique interaction dynamic: conversational code generation with multi-modal feedback. In front-end development, visual artifacts such as sketches, mockups and annotated creenshots are essential for conveying design intent, yet their role in multi-turn code generation remains largely unexplored. To address this gap, we focus on the front-end development task and curate FronTalk, a collection of 100 multi-turn dialogues derived from real-world websites across diverse domains such as news, finance, and art. Each turn features both a textual instruction and an equivalent visual instruction, each representing the same user intent. To comprehensively evaluate model performance, we propose a novel agent-based evaluation framework leveraging a web agent to simulate users and explore the website, and thus measuring both functional correctness and user experience. Evaluation of 20 models reveals two key challenges that are under-explored systematically in the literature: (1) a significant forgetting issue where models overwrite previously implemented features, resulting in task failures, and (2) a persistent challenge in interpreting visual feedback, especially for open-source vision-language models (VLMs). We propose a strong baseline to tackle the forgetting issue with AceCoder, a method that critiques the implementation of every past instruction using an autonomous web agent. This approach significantly reduces forgetting to nearly zero and improves the performance by up to 9.3% (56.0% to 65.3%). Overall, we aim to provide a solid foundation for future research in front-end development and the general interaction dynamics of multi-turn, multi-modal code generation. Code and data are released at https://github.com/shirley-wu/frontalk
Authors: Adel Daoud
Abstract: When sociologists and other social scientist ask whether the return to college differs by race and gender, they face a choice between two fundamentally different modes of inquiry. Traditional interaction models follow deductive logic: the researcher specifies which variables moderate effects and tests these hypotheses. Machine learning methods follow inductive logic: algorithms search across vast combinatorial spaces to discover patterns of heterogeneity. This article develops a framework for navigating between these approaches. We show that the choice between deduction and induction reflects a tradeoff between interpretability and flexibility, and we demonstrate through simulation when each approach excels. Our framework is particularly relevant for inequality research, where understanding how treatment effects vary across intersecting social subpopulation is substantively central.
Authors: Thijs Snelleman, Peter Lundestad Lawrence, Holger H. Hoos, Odd Erik Gundersen
Abstract: Background. Reproducibility is essential to the scientific method, but reproduction is often a laborious task. Recent works have attempted to automate this process and relieve researchers of this workload. However, due to varying definitions of reproducibility, a clear problem statement is missing. Objectives. Create a generalisable problem statement, applicable to any empirical study. We hypothesise that we can represent any empirical study using a structure based on the scientific method and that this representation can be automatically extracted from any publication, and captures the essence of the study. Methods. We apply our definition of reproducibility as a problem statement for the automatisation of reproducibility by automatically extracting the hypotheses, experiments and interpretations of 20 studies and assess the quality based on assessments by the original authors of each study. Results. We create a dataset representing the reproducibility problem, consisting of the representation of 20 studies. The majority of author feedback is positive, for all parts of the representation. In a few cases, our method failed to capture all elements of the study. We also find room for improvement at capturing specific details, such as results of experiments. Conclusions. We conclude that our formulation of the problem is able to capture the concept of reproducibility in empirical AI studies across a wide range of subfields. Authors of original publications generally agree that the produced structure is representative of their work; we believe improvements can be achieved by applying our findings to create a more structured and fine-grained output in future work.
Authors: Basab Jha, Firoj Paudel, Ujjwal Puri, Ethan Henkel, Zhang Yuting, Mateusz Kowalczyk, Mei Huang, Choi Donghyuk, Wang Junhao
Abstract: We demonstrate SAGE-32B, a 32 billion parameter language model that focuses on agentic reasoning and long range planning tasks. Unlike chat models that aim for general conversation fluency, SAGE-32B is designed to operate in an agentic loop, emphasizing task decomposition, tool usage, and error recovery. The model is initialized from the Qwen2.5-32B pretrained model and fine tuned using Iterative Distillation, a two stage training process that improves reasoning performance through rigorously tested feedback loops. SAGE-32B also introduces an inverse reasoning approach, which uses a meta cognition head to forecast potential failures in the planning process before execution. On agentic reasoning benchmarks including MMLU-Pro, AgentBench, and MATH-500, SAGE-32B achieves higher success rates in multi tool usage scenarios compared to similarly sized baseline models, while remaining competitive on standard reasoning evaluations. Model weights are publicly released at https://huggingface.co/sagea-ai/sage-reasoning-32b
Authors: Brady Steele, Micah Katz
Abstract: We present a controlled study of multi-hop contextual reasoning in large language models, providing a clean demonstration of the task-method dissociation: rule-based pattern matching achieves 100% success on structured information retrieval but only 6.7% on tasks requiring cross-document reasoning, while LLM-based multi-agent systems show the inverse pattern, achieving up to 80% on reasoning tasks where rule-based methods fail. Using a synthetic evaluation framework with 120 trials across four models (LLaMA-3 8B, LLaMA-2 13B, Mixtral 8x7B, DeepSeek-V2 16B), we report three key findings: (1) Multi-agent amplification depends on base capability: statistically significant gains occur only for models with sufficient reasoning ability (p < 0.001 for LLaMA-3 8B, p = 0.014 for Mixtral), with improvements of up to 46.7 percentage points, while weaker models show no benefit, suggesting amplification rather than compensation; (2) Active parameters predict reasoning performance: Mixtral's performance aligns with its ~12B active parameters rather than 47B total, consistent with the hypothesis that inference-time compute drives reasoning capability in MoE architectures; (3) Architecture quality matters: LLaMA-3 8B outperforms LLaMA-2 13B despite fewer parameters, consistent with known training improvements. Our results provide controlled quantitative evidence for intuitions about multi-agent coordination and MoE scaling, while highlighting the dependence of multi-agent benefits on base model capability. We release our evaluation framework to support reproducible research on reasoning in mid-scale models.
Authors: Danchun Chen, Qiyao Yan, Liangming Pan
Abstract: Understanding how Large Language Models (LLMs) perform logical reasoning internally remains a fundamental challenge. While prior mechanistic studies focus on identifying taskspecific circuits, they leave open the question of what computational strategies LLMs employ for propositional reasoning. We address this gap through comprehensive analysis of Qwen3 (8B and 14B) on PropLogic-MI, a controlled dataset spanning 11 propositional logic rule categories across one-hop and two-hop reasoning. Rather than asking ''which components are necessary,'' we ask ''how does the model organize computation?'' Our analysis reveals a coherent computational architecture comprising four interlocking mechanisms: Staged Computation (layer-wise processing phases), Information Transmission (information flow aggregation at boundary tokens), Fact Retrospection (persistent re-access of source facts), and Specialized Attention Heads (functionally distinct head types). These mechanisms generalize across model scales, rule types, and reasoning depths, providing mechanistic evidence that LLMs employ structured computational strategies for logical reasoning.
Authors: Ji Guo, Wenbo Jiang, Yansong Lin, Yijing Liu, Ruichen Zhang, Guomin Lu, Aiguo Chen, Xinshuo Han, Hongwei Li, Dusit Niyato
Abstract: Vision-Language-Action (VLA) models are widely deployed in safety-critical embodied AI applications such as robotics. However, their complex multimodal interactions also expose new security vulnerabilities. In this paper, we investigate a backdoor threat in VLA models, where malicious inputs cause targeted misbehavior while preserving performance on clean data. Existing backdoor methods predominantly rely on inserting visible triggers into visual modality, which suffer from poor robustness and low insusceptibility in real-world settings due to environmental variability. To overcome these limitations, we introduce the State Backdoor, a novel and practical backdoor attack that leverages the robot arm's initial state as the trigger. To optimize trigger for insusceptibility and effectiveness, we design a Preference-guided Genetic Algorithm (PGA) that efficiently searches the state space for minimal yet potent triggers. Extensive experiments on five representative VLA models and five real-world tasks show that our method achieves over 90% attack success rate without affecting benign task performance, revealing an underexplored vulnerability in embodied AI systems.
Authors: Sean Niklas Semmler
Abstract: This paper proposes a conceptual framework in which intelligence and consciousness emerge from relational structure rather than from prediction or domain-specific mechanisms. Intelligence is defined as the capacity to form and integrate causal connections between signals, actions, and internal states. Through context enrichment, systems interpret incoming information using learned relational structure that provides essential context in an efficient representation that the raw input itself does not contain, enabling efficient processing under metabolic constraints. Building on this foundation, we introduce the systems-explaining-systems principle, where consciousness emerges when recursive architectures allow higher-order systems to learn and interpret the relational patterns of lower-order systems across time. These interpretations are integrated into a dynamically stabilized meta-state and fed back through context enrichment, transforming internal models from representations of the external world into models of the system's own cognitive processes. The framework reframes predictive processing as an emergent consequence of contextual interpretation rather than explicit forecasting and suggests that recursive multi-system architectures may be necessary for more human-like artificial intelligence.
Authors: Xiaoyu Xu, Minxin Du, Zitong Li, Zi Liang, Zhibiao Guo, Shiyu Zhang, Peizhao Hu, Qingqing Ye, Haibo Hu
Abstract: Although machine unlearning is essential for removing private, harmful, or copyrighted content from LLMs, current benchmarks often fail to faithfully represent the true "forgetting scope" learned by the model. We formalize two distinct unlearning granularities, domain-level and instance-level, and propose BiForget, an automated framework for synthesizing high-quality forget sets. Unlike prior work relying on external generators, BiForget exploits the target model per se to elicit data that matches its internal knowledge distribution through seed-guided and adversarial prompting. Our experiments across diverse benchmarks show that it achieves a superior balance of relevance, diversity, and efficiency. Quantitatively, in the Harry Potter domain, it improves relevance by ${\sim}20$ and diversity by ${\sim}$0.05 while halving the total data size compared to SOTAs. Ultimately, it facilitates more robust forgetting and better utility preservation, providing a more rigorous foundation for evaluating LLM unlearning.
Authors: Paul Kent, George De Ath, Martin Layton, Allen Hart, Richard Everson, Ben Carvell
Abstract: Escalating air traffic demand is driving the adoption of automation to support air traffic controllers, but existing approaches face a trade-off between safety assurance and interpretability. Optimisation-based methods such as reinforcement learning offer strong performance but are difficult to verify and explain, while rules-based systems are transparent yet rarely check safety under uncertainty. This paper outlines Agent Mallard, a forward-planning, rules-based agent for tactical control in systemised airspace that embeds a stochastic digital twin directly into its conflict-resolution loop. Mallard operates on predefined GPS-guided routes, reducing continuous 4D vectoring to discrete choices over lanes and levels, and constructs hierarchical plans from an expert-informed library of deconfliction strategies. A depth-limited backtracking search uses causal attribution, topological plan splicing, and monotonic axis constraints to seek a complete safe plan for all aircraft, validating each candidate manoeuvre against uncertain execution scenarios (e.g., wind variation, pilot response, communication loss) before commitment. Preliminary walkthroughs with UK controllers and initial tests in the BluebirdDT airspace digital twin indicate that Mallard's behaviour aligns with expert reasoning and resolves conflicts in simplified scenarios. The architecture is intended to combine model-based safety assessment, interpretable decision logic, and tractable computational performance in future structured en-route environments.
Authors: Ben Carvell, Marc Thomas, Andrew Pace, Christopher Dorney, George De Ath, Richard Everson, Nick Pepper, Adam Keane, Samuel Tomlinson, Richard Cannon
Abstract: We present a rigorous, human-in-the-loop evaluation framework for assessing the performance of AI agents on the task of Air Traffic Control, grounded in a regulator-certified simulator-based curriculum used for training and testing real-world trainee controllers. By leveraging legally regulated assessments and involving expert human instructors in the evaluation process, our framework enables a more authentic and domain-accurate measurement of AI performance. This work addresses a critical gap in the existing literature: the frequent misalignment between academic representations of Air Traffic Control and the complexities of the actual operational environment. It also lays the foundations for effective future human-machine teaming paradigms by aligning machine performance with human assessment targets.
Authors: Minglei Yin (Fanny), Chuanbo Hu (Fanny), Bin Liu (Fanny), Neil Zhenqiang Gong (Fanny), Yanfang (Fanny), Ye, Xin Li
Abstract: Learning from implicit feedback has become the standard paradigm for modern recommender systems. However, this setting is fraught with the persistent challenge of false negatives, where unobserved user-item interactions are not necessarily indicative of negative preference. To address this issue, this paper introduces a novel and principled loss function, named Corrected and Weighted (CW) loss, that systematically corrects for the impact of false negatives within the training objective. Our approach integrates two key techniques. First, inspired by Positive-Unlabeled learning, we debias the negative sampling process by re-calibrating the assumed negative distribution. By theoretically approximating the true negative distribution (p-) using the observable general data distribution (p) and the positive interaction distribution (p^+), our method provides a more accurate estimate of the likelihood that a sampled unlabeled item is truly negative. Second, we introduce a dynamic re-weighting mechanism that modulates the importance of each negative instance based on the model's current prediction. This scheme encourages the model to enforce a larger ranking margin between positive items and confidently predicted (i.e., easy) negative items, while simultaneously down-weighting the penalty on uncertain negatives that have a higher probability of being false negatives. A key advantage of our approach is its elegance and efficiency; it requires no complex modifications to the data sampling process or significant computational overhead, making it readily applicable to a wide array of existing recommendation models. Extensive experiments conducted on four large-scale, sparse benchmark datasets demonstrate the superiority of our proposed loss. The results show that our method consistently and significantly outperforms a suite of state-of-the-art loss functions across multiple ranking-oriented metrics.
Authors: Ibrahim Tanvir (University of Dhaka), Alif Ruslan (University of Dhaka), Sartaj Solaiman (University of Dhaka)
Abstract: This study presents a comprehensive comparative analysis of custom-built Convolutional Neural Networks (CNNs) against popular pre-trained architectures (ResNet-18 and VGG-16) using both feature extraction and transfer learning approaches. We evaluated these models across five diverse image classification datasets from Bangladesh: Footpath Vision, Auto Rickshaw Detection, Mango Image Classification, Paddy Variety Recognition, and Road Damage Detection. Our experimental results demonstrate that transfer learning with fine-tuning consistently outperforms both custom CNNs built from scratch and feature extraction methods, achieving accuracy improvements ranging from 3% to 76% across different datasets. Notably, ResNet-18 with fine-tuning achieved perfect 100% accuracy on the Road Damage BD dataset. While custom CNNs offer advantages in model size (3.4M parameters vs. 11-134M for pre-trained models) and training efficiency on simpler tasks, pre-trained models with transfer learning provide superior performance, particularly on complex classification tasks with limited training data. This research provides practical insights for practitioners in selecting appropriate deep learning approaches based on dataset characteristics, computational resources, and performance requirements.
Authors: Dongqi Liu, Hang Ding, Qiming Feng, Jian Li, Xurong Xie, Zhucun Xue, Chengjie Wang, Jiangning Zhang, Yabiao Wang
Abstract: Retrieval-Augmented Generation (RAG) has emerged as an important means of enhancing the performance of large language models (LLMs) in knowledge-intensive tasks. However, most existing RAG strategies treat retrieved passages in a flat and unstructured way, which prevents the model from capturing structural cues and constrains its ability to synthesize knowledge from dispersed evidence across documents. To overcome these limitations, we propose Disco-RAG, a discourse-aware framework that explicitly injects discourse signals into the generation process. Our method constructs intra-chunk discourse trees to capture local hierarchies and builds inter-chunk rhetorical graphs to model cross-passage coherence. These structures are jointly integrated into a planning blueprint that conditions the generation. Experiments on question answering and long-document summarization benchmarks show the efficacy of our approach. Disco-RAG achieves state-of-the-art results on the benchmarks without fine-tuning. These findings underscore the important role of discourse structure in advancing RAG systems.
Authors: Arsyi Aziz, Peng Wei
Abstract: Conventional optimization-based metering depends on strict adherence to precomputed schedules, which limits the flexibility required for the stochastic operations of Advanced Air Mobility (AAM). In contrast, multi-agent reinforcement learning (MARL) offers a decentralized, adaptive framework that can better handle uncertainty, required for safe aircraft separation assurance. Despite this advantage, current MARL approaches often overfit to specific airspace structures, limiting their adaptability to new configurations. To improve generalization, we recast the MARL problem in a relative polar state space and train a transformer encoder model across diverse traffic patterns and intersection angles. The learned model provides speed advisories to resolve conflicts while maintaining aircraft near their desired cruising speeds. In our experiments, we evaluated encoder depths of 1, 2, and 3 layers in both structured and unstructured airspaces, and found that a single encoder configuration outperformed deeper variants, yielding near-zero near mid-air collision rates and shorter loss-of-separation infringements than the deeper configurations. Additionally, we showed that the same configuration outperforms a baseline model designed purely with attention. Together, our results suggest that the newly formulated state representation, novel design of neural network architecture, and proposed training strategy provide an adaptable and scalable decentralized solution for aircraft separation assurance in both structured and unstructured airspaces.
Authors: Flavio Chierichetti, Mirko Giacchini, Ravi Kumar, Silvio Lattanzi, Alessandro Panconesi, Erasmo Tani, Andrew Tomkins
Abstract: A Multinomial Logit (MNL) model is composed of a finite universe of items $[n]=\{1,..., n\}$, each assigned a positive weight. A query specifies an admissible subset -- called a slate -- and the model chooses one item from that slate with probability proportional to its weight. This query model is also known as the Plackett-Luce model or conditional sampling oracle in the literature. Although MNLs have been studied extensively, a basic computational question remains open: given query access to slates, how efficiently can we learn weights so that, for every slate, the induced choice distribution is within total variation distance $\varepsilon$ of the ground truth? This question is central to MNL learning and has direct implications for modern recommender system interfaces. We provide two algorithms for this task, one with adaptive queries and one with non-adaptive queries. Each algorithm outputs an MNL $M'$ that induces, for each slate $S$, a distribution $M'_S$ on $S$ that is within $\varepsilon$ total variation distance of the true distribution. Our adaptive algorithm makes $O\left(\frac{n}{\varepsilon^{3}}\log n\right)$ queries, while our non-adaptive algorithm makes $O\left(\frac{n^{2}}{\varepsilon^{3}}\log n \log\frac{n}{\varepsilon}\right)$ queries. Both algorithms query only slates of size two and run in time proportional to their query complexity. We complement these upper bounds with lower bounds of $\Omega\left(\frac{n}{\varepsilon^{2}}\log n\right)$ for adaptive queries and $\Omega\left(\frac{n^{2}}{\varepsilon^{2}}\log n\right)$ for non-adaptive queries, thus proving that our adaptive algorithm is optimal in its dependence on the support size $n$, while the non-adaptive one is tight within a $\log n$ factor.
Authors: Ahmad Mohammad Saber, Saeed Jafari, Zhengmao Ouyang, Paul Budnarain, Amr Youssef, Deepa Kundur
Abstract: This paper presents a large language model (LLM)-based framework for detecting cyberattacks on transformer current differential relays (TCDRs), which, if undetected, may trigger false tripping of critical transformers. The proposed approach adapts and fine-tunes compact LLMs such as DistilBERT to distinguish cyberattacks from actual faults using textualized multidimensional TCDR current measurements recorded before and after tripping. Our results demonstrate that DistilBERT detects 97.6% of cyberattacks without compromising TCDR dependability and achieves inference latency below 6 ms on a commercial workstation. Additional evaluations confirm the framework's robustness under combined time-synchronization and false-data-injection attacks, resilience to measurement noise, and stability across prompt formulation variants. Furthermore, GPT-2 and DistilBERT+LoRA achieve comparable performance, highlighting the potential of LLMs for enhancing smart grid cybersecurity. We provide the full dataset used in this study for reproducibility.
Authors: Dmitriy Poteryayev, Pietro Novelli, Annalisa Coriolano, Riccardo Dettori, Valentina Tozzini, Fabio Beltram, Massimiliano Pontil, Antonio Rossi, Stiven Forti, Camilla Coletti
Abstract: Raman spectroscopy is a key tool for graphene characterization, yet its application to graphene grown on silicon carbide (SiC) is strongly limited by the intense and variable second-order Raman response of the substrate. This limitation is critical for buffer layer graphene, a semiconducting interfacial phase, whose vibrational signatures are overlapped with the SiC background and challenging to be reliably accessed using conventional reference-based subtraction, due to strong spatial and experimental variability of the substrate signal. Here we present SpectraFormer, a transformer-based deep learning model that reconstructs the SiC Raman substrate contribution directly from post-growth partially masked spectroscopic data without relying on explicit reference measurements. By learning global correlations across the entire Raman shift range, the model captures the statistical structure of the SiC background and enables accurate reconstruction of its contribution in mixed spectra. Subtraction of the reconstructed substrate signal reveals weak vibrational features associated with ZLG that are inaccessible through conventional analysis methods. The extracted spectra are validated by ab initio vibrational calculations, allowing assignment of the resolved features to specific modes and confirming their physical consistency. By leveraging a state-of-the-art attention-based deep learning architecture, this approach establishes a robust, reference-free framework for Raman analysis of graphene on SiC and provides a foundation, compatible with real-time data acquisition, to its integration into automated, closed-loop AI-assisted growth optimization.
Authors: Chuan Meng, Jiqun Liu, Mohammad Aliannejadi, Fengran Mo, Jeff Dalton, Maarten de Rijke
Abstract: Using large language models (LLMs) to predict relevance judgments has shown promising results. Most studies treat this task as a distinct research line, e.g., focusing on prompt design for predicting relevance labels given a query and passage. However, predicting relevance judgments is essentially a form of relevance prediction, a problem extensively studied in tasks such as re-ranking. Despite this potential overlap, little research has explored reusing or adapting established re-ranking methods to predict relevance judgments, leading to potential resource waste and redundant development. To bridge this gap, we reproduce re-rankers in a re-ranker-as-relevance-judge setup. We design two adaptation strategies: (i) using binary tokens (e.g., "true" and "false") generated by a re-ranker as direct judgments, and (ii) converting continuous re-ranking scores into binary labels via thresholding. We perform extensive experiments on TREC-DL 2019 to 2023 with 8 re-rankers from 3 families, ranging from 220M to 32B, and analyse the evaluation bias exhibited by re-ranker-based judges. Results show that re-ranker-based relevance judges, under both strategies, can outperform UMBRELA, a state-of-the-art LLM-based relevance judge, in around 40% to 50% of the cases; they also exhibit strong self-preference towards their own and same-family re-rankers, as well as cross-family bias.
Authors: Ignacio Sastre, Aiala Ros\'a
Abstract: We propose Concept Tokens, a lightweight method that adds a new special token to a pretrained LLM and learns only its embedding from multiple natural language definitions of a target concept, where occurrences of the concept are replaced by the new token. The LLM is kept frozen and the embedding is optimized with the standard language-modeling objective. We evaluate Concept Tokens in three settings. First, we study hallucinations in closed-book question answering on HotpotQA and find a directional effect: negating the hallucination token reduces hallucinated answers mainly by increasing abstentions, whereas asserting it increases hallucinations and lowers precision. Second, we induce recasting, a pedagogical feedback strategy for second language teaching, and observe the same directional effect. Moreover, compared to providing the full definitional corpus in-context, concept tokens better preserve compliance with other instructions (e.g., asking follow-up questions). Finally, we include a qualitative study with the Eiffel Tower and a fictional "Austral Tower" to illustrate what information the learned embeddings capture and where their limitations emerge. Overall, Concept Tokens provide a compact control signal learned from definitions that can steer behavior in frozen LLMs.
Authors: Iaroslav Chelombitko, Ekaterina Chelombitko, Aleksey Komissarov
Abstract: The quality of subword tokenization is critical for Large Language Models, yet evaluating tokenizers for morphologically rich Uralic languages is hampered by the lack of clean morpheme lexicons. We introduce SampoNLP, a corpus-free toolkit for morphological lexicon creation using MDL-inspired Self-Referential Atomicity Scoring, which filters composite forms through internal structural cues - suited for low-resource settings. Using the high-purity lexicons generated by SampoNLP for Finnish, Hungarian, and Estonian, we conduct a systematic evaluation of BPE tokenizers across a range of vocabulary sizes (8k-256k). We propose a unified metric, the Integrated Performance Score (IPS), to navigate the trade-off between morpheme coverage and over-splitting. By analyzing the IPS curves, we identify the "elbow points" of diminishing returns and provide the first empirically grounded recommendations for optimal vocabulary sizes (k) in these languages. Our study not only offers practical guidance but also quantitatively demonstrates the limitations of standard BPE for highly agglutinative languages. The SampoNLP library and all generated resources are made publicly available: https://github.com/AragonerUA/SampoNLP
Authors: Jiaheng Chen, Daniel Sanz-Alonso
Abstract: This paper establishes convergence rates for learning elliptic pseudo-differential operators, a fundamental operator class in partial differential equations and mathematical physics. In a wavelet-Galerkin framework, we formulate learning over this class as a structured infinite-dimensional regression problem with multiscale sparsity. Building on this structure, we propose a sparse, data- and computation-efficient estimator, which leverages a novel matrix compression scheme tailored to the learning task and a nested-support strategy to balance approximation and estimation errors. In addition to obtaining convergence rates for the estimator, we show that the learned operator induces an efficient and stable Galerkin solver whose numerical error matches its statistical accuracy. Our results therefore contribute to bringing together operator learning, data-driven solvers, and wavelet methods in scientific computing.
Authors: Shadeeb Hossain
Abstract: Bioelectrical properties of cells such as relative permittivity, conductivity, and characteristic time constants vary significantly between healthy and malignant cells across different frequencies. These distinctions provide a promising foundation for diagnostic and classification applications. This study systematically reviewed 33 scholarly articles to compile datasets of quantitative bioelectric parameters and evaluated their utility in predictive modeling. Three supervised machine learning algorithms- Random Forest (RF), Support Vector Machine (SVM), and K-Nearest Neighbor (KNN) were implemented and tuned using key hyperparameters to assess classification performance. Model effectiveness was evaluated using accuracy and F1 score as performance metrics. Results demonstrate that Random Forest achieved the highest predictive accuracy of ~ 90% when configured with a maximum depth of 4 and 100 estimators. These findings highlight the potential of integrating bioelectrical property analysis with machine learning for improved diagnostic decision-making. Similarly, for KNN and SVM, the F1 score peaked at approximately 78% and 76.5%, respectively. Future work will explore incorporating additional discriminative features, leveraging stimulated datasets, and optimizing hyperparameter through advanced search strategies. Ultimately, hardware prototype with embedded micro-electrodes and real-time control systems could pave the path for practical diagnostic tools capable of in-situ cell classification.
Authors: Daniel Connor, Colin Defant
Abstract: We introduce Minary, a computational framework designed as a candidate for the first formally provable autopoietic primitive. Minary represents interacting probabilistic events as multi-dimensional vectors and combines them via linear superposition rather than multiplicative scalar operations, thereby preserving uncertainty and enabling constructive and destructive interference in the range $[-1,1]$. A fixed set of ``perspectives'' evaluates ``semantic dimensions'' according to hidden competencies, and their interactions drive two discrete-time stochastic processes. We model this system as an iterated random affine map and use the theory of iterated random functions to prove that it converges in distribution to a unique stationary law; we moreover obtain an explicit closed form for the limiting expectation in terms of row, column, and global averages of the competency matrix. We then derive exact formulas for the mean and variance of the normalized consensus conditioned on the activation of a given semantic dimension, revealing how consensus depends on competency structure rather than raw input signals. Finally, we argue that Minary is organizationally closed yet operationally open in the sense of Maturana and Varela, and we discuss implications for building self-maintaining, distributed, and parallelizable computational systems that house a uniquely subjective notion of identity.
Authors: Christophe Bonneville, Nathan Bieberdorf, Pieterjan Robbe, Mark Asta, Habib Najm, Laurent Capolungo, Cosmin Safta
Abstract: Phase-field simulations of liquid metal dealloying (LMD) can capture complex microstructural evolutions but can be prohibitively expensive for large domains and long time horizons. In this paper, we introduce a fully convolutional, conditionally parameterized U-Net surrogate designed to extrapolate far beyond its training data in both space and time. The architecture integrates convolutional self-attention, physically informed padding, and a flood-fill corrector method to maintain accuracy under extreme extrapolation, while conditioning on simulation parameters allows for flexible time-step skipping and adaptation to varying alloy compositions. To remove the need for costly solver-based initialization, we couple the surrogate with a conditional diffusion model that generates synthetic, physically consistent initial conditions. We train our surrogate on simulations generated over small domain sizes and short time spans, but, by taking advantage of the convolutional nature of U-Nets, we are able to run and extrapolate surrogate simulations for longer time horizons than what would be achievable with classic numerical solvers. Across multiple alloy compositions, the framework is able to reproduce the LMD physics accurately. It predicts key quantities of interest and spatial statistics with relative errors typically below 5% in the training regime and under 15% during large-scale, long time-horizon extrapolations. Our framework can also deliver speed-ups of up to 36,000 times, bringing the time to run weeks-long simulations down to a few seconds. This work is a first stepping stone towards high-fidelity extrapolation in both space and time of phase-field simulation for LMD.
Authors: Zhenglong Luo, Zhiyong Chen, Aoxiang Liu
Abstract: Multiagent reinforcement learning, as a prominent intelligent paradigm, enables collaborative decision-making within complex systems. However, existing approaches often rely on explicit action exchange between agents to evaluate action value functions, which is frequently impractical in real-world engineering environments due to communication constraints, latency, energy consumption, and reliability requirements. From an artificial intelligence perspective, this paper proposes an enhanced multiagent reinforcement learning framework that employs action estimation neural networks to infer agent behaviors. By integrating a lightweight action estimation module, each agent infers neighboring agents' behaviors using only locally observable information, enabling collaborative policy learning without explicit action sharing. This approach is fully compatible with standard TD3 algorithms and scalable to larger multiagent systems. At the engineering application level, this framework has been implemented and validated in dual-arm robotic manipulation tasks: two robotic arms collaboratively lift objects. Experimental results demonstrate that this approach significantly enhances the robustness and deployment feasibility of real-world robotic systems while reducing dependence on information infrastructure. Overall, this research advances the development of decentralized multiagent artificial intelligence systems while enabling AI to operate effectively in dynamic, information-constrained real-world environments.
Authors: Zimo Yan, Zheng Xie, Runfan Duan, Chang Liu, Wumei Du
Abstract: Molecular graph learning benefits from positional signals that capture both local neighborhoods and global topology. Two widely used families are spectral encodings derived from Laplacian or diffusion operators and anchor-based distance encodings built from shortest-path information, yet their precise relationship is poorly understood. We interpret distance encodings as a low-rank surrogate of diffusion geometry and derive an explicit trilateration map that reconstructs truncated diffusion coordinates from transformed anchor distances and anchor spectral positions, with pointwise and Frobenius-gap guarantees on random regular graphs. On DrugBank molecular graphs using a shared GNP-based DDI prediction backbone, a distance-driven Nystr\"om scheme closely recovers diffusion geometry, and both Laplacian and distance encodings substantially outperform a no-encoding baseline.
Authors: Shogo Nakayama, Masahiro Okuda
Abstract: The advancement of deep learning has greatly improved supervised image classification. However, labeling data is costly, prompting research into unsupervised learning methods such as contrastive learning. In real-world scenarios, fully unlabeled datasets are rare, making semi-supervised learning (SSL) highly relevant in scenarios where a small amount of labeled data coexists with a large volume of unlabeled data. A well-known semi-supervised contrastive learning approach involves assigning pseudo-labels to unlabeled data. This study aims to enhance pseudo-label-based SSL by incorporating distribution matching between labeled and unlabeled feature embeddings to improve image classification accuracy across multiple datasets.
Authors: Noah Eckstein, Manoj Srinivasan
Abstract: In recurrent neural networks (RNNs) used to model biological neural networks, noise is typically introduced during training to emulate biological variability and regularize learning. The expectation is that removing the noise at test time should preserve or improve performance. Contrary to this intuition, we find that continuous-time recurrent neural networks (CTRNNs) often perform best at a nonzero noise level, specifically, the same level used during training. This noise preference typically arises when noise is injected inside the neural activation function; networks trained with noise injected outside the activation function perform best with zero noise. Through analyses of simple function approximation, maze navigation, and single neuron regulator tasks, we show that the phenomenon stems from noise-induced shifts of fixed points (stationary distributions) in the underlying stochastic dynamics of the RNNs. These fixed point shifts are noise-level dependent and bias the network outputs when the noise is removed, degrading performance. Analytical and numerical results show that the bias arises when neural states operate near activation function nonlinearities, where noise is asymmetrically attenuated, and that performance optimization incentivizes operation near these nonlinearities. Thus, networks can overfit to the stochastic training environment itself rather than just to the input-output data. The phenomenon is distinct from stochastic resonance, wherein nonzero noise enhances signal processing. Our findings reveal that training noise can become an integral part of the computation learned by recurrent networks, with implications for understanding neural population dynamics and for the design of robust artificial RNNs.
Authors: Yash Saxena, Manas Gaur
Abstract: Retrieval Augmented Generation (RAG) has made significant strides in overcoming key limitations of large language models, such as hallucination, lack of contextual grounding, and issues with transparency. However, traditional RAG systems consist of three interconnected neural components - the retriever, re-ranker, and generator - whose internal reasoning processes remain opaque. This lack of transparency complicates interpretability, hinders debugging efforts, and erodes trust, especially in high-stakes domains where clear decision-making is essential. To address these challenges, we introduce the concept of Neurosymbolic RAG, which integrates symbolic reasoning using a knowledge graph with neural retrieval techniques. This new framework aims to answer two primary questions: (a) Can retrievers provide a clear and interpretable basis for document selection? (b) Can symbolic knowledge enhance the clarity of the retrieval process? We propose three methods to improve this integration. First is MAR (Knowledge Modulation Aligned Retrieval) that employs modulation networks to refine query embeddings using interpretable symbolic features, thereby making document matching more explicit. Second, KG-Path RAG enhances queries by traversing knowledge graphs to improve overall retrieval quality and interpretability. Lastly, Process Knowledge-infused RAG utilizes domain-specific tools to reorder retrieved content based on validated workflows. Preliminary results from mental health risk assessment tasks indicate that this neurosymbolic approach enhances both transparency and overall performance
Authors: Jiachen Liu, Maestro Harmon, Zechen Zhang
Abstract: While AI innovation accelerates rapidly, the intellectual process behind breakthroughs -- how researchers identify gaps, synthesize prior work, and generate insights -- remains poorly understood. The lack of structured data on scientific reasoning hinders systematic analysis and development of AI research agents. We introduce Sci-Reasoning, the first dataset capturing the intellectual synthesis behind high-quality AI research. Using community-validated quality signals and an LLM-accelerated, human-verified pipeline, we trace Oral and Spotlight papers across NeurIPS, ICML, and ICLR (2023-2025) to its key predecessors, articulating specific reasoning links in a structured format. Our analysis identifies 15 distinct thinking patterns, with three dominant strategies accounting for 52.7%: Gap-Driven Reframing (24.2%), Cross-Domain Synthesis (18.0%), and Representation Shift (10.5%). The most powerful innovation recipes combine multiple patterns: Gap-Driven Reframing + Representation Shift, Cross-Domain Synthesis + Representation Shift, and Gap-Driven Reframing + Cross-Domain Synthesis. This dataset enables quantitative studies of scientific progress and provides structured reasoning trajectories for training the next generation AI research agents.
Authors: Zhiyuan He, Binghan Chen, Tianxiang Xiong, Ziyang Sun, Mozhao Zhu, Xi Chen
Abstract: Recent advances in Knowledge Editing (KE), particularly Rank-One Model Editing (ROME), show superior efficiency over fine-tuning and in-context learning for updating single-hop facts in transformers. However, these methods face significant challenges when applied to multi-hop reasoning tasks requiring knowledge chaining. In this work, we study the effect of editing knowledge with ROME on different layer depths and identify three key failure modes. First, the "hopping-too-late" problem occurs as later layers lack access to necessary intermediate representations. Second, generalization ability deteriorates sharply when editing later layers. Third, the model overfits to edited knowledge, incorrectly prioritizing edited-hop answers regardless of context. To mitigate the issues of "hopping-too-late" and generalisation decay, we propose Redundant Editing, a simple yet effective strategy that enhances multi-hop reasoning. Our experiments demonstrate that this approach can improve accuracy on 2-hop questions by at least 15.5 percentage points, representing a 96% increase over the previous single-edit strategy, while trading off some specificity and language naturalness.
Authors: Osman Goni Ridwan, Gilles Frapper, Hongfei Xue, Qiang Zhu
Abstract: We present a materials generation framework that couples a symmetry-conditioned variational autoencoder (CVAE) with a differentiable SO(3) power spectrum objective to steer candidates toward a specified local environment under the crystallographic constraints. In particular, we implement a fully differentiable pipeline that performs batch-wise optimization on both direct and latent crystallographic representations. Using the GPU acceleration, the implementation achieves about fivefold speed compared to our previous CPU workflow, while yielding comparable outcomes. In addition, we introduce the optimization strategy that alternatively performs optimization on the direct and latent crystal representations. This dual-level relaxation approach can effectively overcome local barrier defined by different objective gradients, thus increasing the success rate of generating complex structures satisfying the targe local environments. This framework can be extended to systems consisting of multi-components and multi-environments, providing a scalable route to generate material structures with the target local environment.
Authors: Lionel Z. Wang, Yusheng Zhao, Jiabin Luo, Xinfeng Li, Lixu Wang, Yinan Peng, Haoyang Li, XiaoFeng Wang, Wei Dong
Abstract: The deployment of Machine-Generated Text (MGT) detection systems necessitates processing sensitive user data, creating a fundamental conflict between authorship verification and privacy preservation. Standard anonymization techniques often disrupt linguistic fluency, while rigorous Differential Privacy (DP) mechanisms typically degrade the statistical signals required for accurate detection. To resolve this dilemma, we propose \textbf{DP-MGTD}, a framework incorporating an Adaptive Differentially Private Entity Sanitization algorithm. Our approach utilizes a two-stage mechanism that performs noisy frequency estimation and dynamically calibrates privacy budgets, applying Laplace and Exponential mechanisms to numerical and textual entities respectively. Crucially, we identify a counter-intuitive phenomenon where the application of DP noise amplifies the distinguishability between human and machine text by exposing distinct sensitivity patterns to perturbation. Extensive experiments on the MGTBench-2.0 dataset show that our method achieves near-perfect detection accuracy, significantly outperforming non-private baselines while satisfying strict privacy guarantees.
Authors: Prateek Jain, Shabari S Nair, Ritesh Goru, Prakhar Agarwal, Ajay Yadav, Yoga Sri Varshan Varadharajan, Constantine Caramanis
Abstract: Large-scale multi-tenant retrieval systems amass vast user query logs yet critically lack the curated relevance labels required for effective domain adaptation. This "dark data" problem is exacerbated by the operational cost of model updates: jointly fine-tuning query and document encoders requires re-indexing the entire corpus, which is prohibitive in multi-tenant environments with thousands of isolated indices. To address these dual challenges, we introduce \textbf{DevRev Search}, a passage retrieval benchmark for technical customer support constructed through a fully automatic pipeline. We employ a \textbf{fusion-based candidate generation} strategy, pooling results from diverse sparse and dense retrievers, and utilize an LLM-as-a-Judge to perform rigorous \textbf{consistency filtering} and relevance assignment. We further propose a practical \textbf{Index-Preserving Adaptation} strategy: by fine-tuning only the query encoder via Low-Rank Adaptation (LoRA), we achieve competitive performance improvements while keeping the document index frozen. Our experiments on DevRev Search and SciFact demonstrate that targeting specific transformer layers in the query encoder yields optimal quality-efficiency trade-offs, offering a scalable path for personalized enterprise search.
Authors: Xiang Li, Bing Luo, Jianwei Huang, Yuan Luo
Abstract: Mechanism design is pivotal to federated learning (FL) for maximizing social welfare by coordinating self-interested clients. Existing mechanisms, however, often overlook the network effects of client participation and the diverse model performance requirements (i.e., generalization error) across applications, leading to suboptimal incentives and social welfare, or even inapplicability in real deployments. To address this gap, we explore incentive mechanism design for FL with network effects and application-specific requirements of model performance. We develop a theoretical model to quantify the impact of network effects on heterogeneous client participation, revealing the non-monotonic nature of such effects. Based on these insights, we propose a Model Trading and Sharing (MoTS) framework, which enables clients to obtain FL models through either participation or purchase. To further address clients' strategic behaviors, we design a Social Welfare maximization with Application-aware and Network effects (SWAN) mechanism, exploiting model customer payments for incentivization. Experimental results on a hardware prototype demonstrate that our SWAN mechanism outperforms existing FL mechanisms, improving social welfare by up to $352.42\%$ and reducing extra incentive costs by $93.07\%$.
Authors: Enze Pan
Abstract: We present Tape, a controlled reinforcement-learning benchmark designed to isolate out-of-distribution (OOD) failure under latent rule shifts.Tape is derived from one-dimensional cellular automata, enabling precise train/test splits where observation and action spaces are held fixed while transition rules change. Using a reproducible evaluation pipeline, we compare model-free baselines, model-based planning with learned world models, and task-inference (meta-RL) methods. A consistent pattern emerges: methods that are strong in-distribution (ID) can collapse under heldout-rule OOD, and high-variance OOD evaluation can make rankings unstable unless experiments are sufficiently replicated.We provide (i) standardized OOD protocols, (ii) statistical reporting requirements (seeds, confidence intervals, and hypothesis tests), and (iii) information-theoretic identities connecting entropy reduction to conditional mutual information and expected posterior KL divergence, clarifying what "uncertainty reduction" objectives can and cannot guarantee under rule shifts.
Authors: Yinuo Wang, Mining Tan, Wenxiang Jiao, Xiaoxi Li, Hao Wang, Xuanyu Zhang, Yuan Lu, Weiming Dong
Abstract: Travel planning is a sophisticated decision-making process that requires synthesizing multifaceted information to construct itineraries. However, existing travel planning approaches face several challenges: (1) Pruning candidate points of interest (POIs) while maintaining a high recall rate; (2) A single reasoning path restricts the exploration capability within the feasible solution space for travel planning; (3) Simultaneously optimizing hard constraints and soft constraints remains a significant difficulty. To address these challenges, we propose TourPlanner, a comprehensive framework featuring multi-path reasoning and constraint-gated reinforcement learning. Specifically, we first introduce a Personalized Recall and Spatial Optimization (PReSO) workflow to construct spatially-aware candidate POIs' set. Subsequently, we propose Competitive consensus Chain-of-Thought (CCoT), a multi-path reasoning paradigm that improves the ability of exploring the feasible solution space. To further refine the plan, we integrate a sigmoid-based gating mechanism into the reinforcement learning stage, which dynamically prioritizes soft-constraint satisfaction only after hard constraints are met. Experimental results on travel planning benchmarks demonstrate that TourPlanner achieves state-of-the-art performance, significantly surpassing existing methods in both feasibility and user-preference alignment.
Authors: Feihu Jin, Shipeng Cen, Ying Tan
Abstract: Fine-tuning large language models (LLMs) has achieved remarkable success across various NLP tasks, but the substantial memory overhead during backpropagation remains a critical bottleneck, especially as model scales grow. Zeroth-order (ZO) optimization alleviates this issue by estimating gradients through forward passes and Gaussian sampling, avoiding the need for backpropagation. However, conventional ZO methods suffer from high variance in gradient estimation due to their reliance on random perturbations, leading to slow convergence and suboptimal performance. We propose a simple plug-and-play method that incorporates prior-informed perturbations to refine gradient estimation. Our method dynamically computes a guiding vector from Gaussian samples, which directs perturbations toward more informative directions, significantly accelerating convergence compared to standard ZO approaches. We further investigate a greedy perturbation strategy to explore the impact of prior knowledge on gradient estimation. Theoretically, we prove that our gradient estimator achieves stronger alignment with the true gradient direction, enhancing optimization efficiency. Extensive experiments across LLMs of varying scales and architectures demonstrate that our proposed method could seamlessly integrate into existing optimization methods, delivering faster convergence and superior performance. Notably, on the OPT-13B model, our method outperforms traditional ZO optimization across all 11 benchmark tasks and surpasses gradient-based baselines on 9 out of 11 tasks, establishing a robust balance between efficiency and accuracy.
Authors: Dominik Freinberger, Philipp Moser
Abstract: Quantum machine learning has emerged as a promising application domain for near-term quantum hardware, particularly through hybrid quantum-classical models that leverage both classical and quantum processing. Although numerous hybrid architectures have been proposed and demonstrated successfully on benchmark tasks, a significant open question remains regarding the specific contribution of quantum components to the overall performance of these models. In this work, we aim to shed light on the impact of quantum processing within hybrid quantum-classical neural network architectures through a rigorous statistical study. We systematically assess common hybrid models on medical signal data as well as planar and volumetric images, examining the influence attributable to classical and quantum aspects such as encoding schemes, entanglement, and circuit size. We find that in best-case scenarios, hybrid models show performance comparable to their classical counterparts, however, in most cases, performance metrics deteriorate under the influence of quantum components. Our multi-modal analysis provides realistic insights into the contributions of quantum components and advocates for cautious claims and design choices for hybrid models in near-term applications.
Authors: Santiago Acevedo, Alessandro Laio, Marco Baroni
Abstract: We study how syntactic and semantic information is encoded in inner layer representations of Large Language Models (LLMs), focusing on the very large DeepSeek-V3. We find that, by averaging hidden-representation vectors of sentences sharing syntactic structure or meaning, we obtain vectors that capture a significant proportion of the syntactic and semantic information contained in the representations. In particular, subtracting these syntactic and semantic ``centroids'' from sentence vectors strongly affects their similarity with syntactically and semantically matched sentences, respectively, suggesting that syntax and semantics are, at least partially, linearly encoded. We also find that the cross-layer encoding profiles of syntax and semantics are different, and that the two signals can to some extent be decoupled, suggesting differential encoding of these two types of linguistic information in LLM representations.
Authors: Lee Hyoseok, Sohwi Lim, Eunju Cha, Tae-Hyun Oh
Abstract: With recent advances in generative models, diffusion models have emerged as powerful priors for solving inverse problems in each domain. Since Latent Diffusion Models (LDMs) provide generic priors, several studies have explored their potential as domain-agnostic zero-shot inverse solvers. Despite these efforts, existing latent diffusion inverse solvers suffer from their instability, exhibiting undesirable artifacts and degraded quality. In this work, we first identify the instability as a discrepancy between the solver's and true reverse diffusion dynamics, and show that reducing this gap stabilizes the solver. Building on this, we introduce Measurement-Consistent Langevin Corrector (MCLC), a theoretically grounded plug-and-play correction module that remedies the LDM-based inverse solvers through measurement-consistent Langevin updates. Compared to prior approaches that rely on linear manifold assumptions, which often do not hold in latent space, MCLC operates without this assumption, leading to more stable and reliable behavior. We experimentally demonstrate the effectiveness of MCLC and its compatibility with existing solvers across diverse image restoration tasks. Additionally, we analyze blob artifacts and offer insights into their underlying causes. We highlight that MCLC is a key step toward more robust zero-shot inverse problem solvers.
Authors: Lei Xu, Shanshan Wang, Chenglong Xiao
Abstract: High-Level Synthesis (HLS) design space exploration (DSE) seeks Pareto-optimal designs within expansive pragma configuration spaces. To accelerate HLS DSE, graph neural networks (GNNs) are commonly employed as surrogates for HLS tools to predict quality of results (QoR) metrics, while multi-objective optimization algorithms expedite the exploration. However, GNN-based prediction methods may not fully capture the rich semantic features inherent in behavioral descriptions, and conventional multi-objective optimization algorithms often do not explicitly account for the domain-specific knowledge regarding how pragma directives influence QoR. To address these limitations, this paper proposes the MPM-LLM4DSE framework, which incorporates a multimodal prediction model (MPM) that simultaneously fuses features from behavioral descriptions and control and data flow graphs. Furthermore, the framework employs a large language model (LLM) as an optimizer, accompanied by a tailored prompt engineering methodology. This methodology incorporates pragma impact analysis on QoR to guide the LLM in generating high-quality configurations (LLM4DSE). Experimental results demonstrate that our multimodal predictive model significantly outperforms state-of-the-art work ProgSG by up to 10.25$\times$. Furthermore, in DSE tasks, the proposed LLM4DSE achieves an average performance gain of 39.90\% over prior methods, validating the effectiveness of our prompting methodology. Code and models are available at https://github.com/wslcccc/MPM-LLM4DSE.
Authors: Muhammad Shoaib, Zaka Ur Rehman, Muhammad Qasim
Abstract: The aim of this paper is to use Maximum Likelihood (ML) Classification on multispectral data by means of qualitative and quantitative approaches. Maximum Likelihood is a supervised classification algorithm which is based on the Classical Bayes theorem. It makes use of a discriminant function to assign pixel to the class with the highest likelihood. Class means vector and covariance matrix are the key inputs to the function and can be estimated from training pixels of a particular class. As Maximum Likelihood need some assumptions before it has to be applied on the data. In this paper we will compare the results of Maximum Likelihood Classification (ML) before apply the Weierstrass Transform and apply Weierstrass Transform and will see the difference between the accuracy on training pixels of high resolution Quickbird satellite image. Principle Component analysis (PCA) is also used for dimension reduction and also used to check the variation in bands. The results shows that the separation between mean of the classes in the decision space is to be the main factor that leads to the high classification accuracy of Maximum Likelihood (ML) after using Weierstrass Transform than without using it.
Authors: Matan Kleiner, Lior Michaeli, Tomer Michaeli
Abstract: Diffractive neural networks have recently emerged as a promising framework for all-optical computing. However, these networks are typically trained for a single task, limiting their potential adoption in systems requiring multiple functionalities. Existing approaches to achieving multi-task functionality either modify the mechanical configuration of the network per task or use a different illumination wavelength or polarization state for each task. In this work, we propose a new control mechanism, which is based on the illumination's angular spectrum. Specifically, we shape the illumination using an amplitude mask that selectively controls its angular spectrum. We employ different illumination masks for achieving different network functionalities, so that the mask serves as a unique task encoder. Interestingly, we show that effective control can be achieved over a very narrow angular range, within the paraxial regime. We numerically illustrate the proposed approach by training a single diffractive network to perform multiple image-to-image translation tasks. In particular, we demonstrate translating handwritten digits into typeset digits of different values, and translating handwritten English letters into typeset numbers and typeset Greek letters, where the type of the output is determined by the illumination's angular components. As we show, the proposed framework can work under different coherence conditions, and can be combined with existing control strategies, such as different wavelengths. Our results establish the illumination angular spectrum as a powerful degree of freedom for controlling diffractive networks, enabling a scalable and versatile framework for multi-task all-optical computing.
Authors: Oshri Naparstek
Abstract: Autoregressive language models are conventionally defined over discrete token sequences, committing to a specific token at every generation step. This early discretization forces uncertainty to be resolved through token-level sampling, often leading to instability, repetition, and sensitivity to decoding heuristics. In this work, we introduce a continuous autoregressive formulation of language generation in which tokens are represented as continuous vectors that \emph{mature} over multiple update steps before being discretized. Rather than sampling tokens, the model evolves continuous token representations through a deterministic dynamical process, committing to a discrete token only when the representation has sufficiently converged. Discrete text is recovered via hard decoding, while uncertainty is maintained and resolved in the continuous space. We show that this maturation process alone is sufficient to produce coherent and diverse text using deterministic decoding (argmax), without reliance on token-level sampling, diffusion-style denoising, or auxiliary stabilization mechanisms. Additional perturbations, such as stochastic dynamics or history smoothing, can be incorporated naturally but are not required for the model to function. To our knowledge, this is the first autoregressive language model that generates text by evolving continuous token representations to convergence prior to discretization, enabling stable generation without token-level sampling.
Authors: Alistair Carson, Alec Wright, Stefan Bilbao
Abstract: Modulation effects such as phasers, flangers and chorus effects are heavily used in conjunction with the electric guitar. Machine learning based emulation of analog modulation units has been investigated in recent years, but most methods have either been limited to one class of effect or suffer from a high computational cost or latency compared to canonical digital implementations. Here, we build on previous work and present a framework for modelling flanger, chorus and phaser effects based on differentiable digital signal processing. The model is trained in the time-frequency domain, but at inference operates in the time-domain, requiring zero latency. We investigate the challenges associated with gradient-based optimisation of such effects, and show that low-frequency weighting of loss functions avoids convergence to local minima when learning delay times. We show that when trained against analog effects units, sound output from the model is in some cases perceptually indistinguishable from the reference, but challenges still remain for effects with long delay times and feedback.
Authors: Isabella A. Stewart, Markus J. Buehler
Abstract: Scientific inquiry requires systems-level reasoning that integrates heterogeneous experimental data, cross-domain knowledge, and mechanistic evidence into coherent explanations. While Large Language Models (LLMs) offer inferential capabilities, they often depend on retrieval-augmented contexts that lack structural depth. Traditional Knowledge Graphs (KGs) attempt to bridge this gap, yet their pairwise constraints fail to capture the irreducible higher-order interactions that govern emergent physical behavior. To address this, we introduce a methodology for constructing hypergraph-based knowledge representations that faithfully encode multi-entity relationships. Applied to a corpus of ~1,100 manuscripts on biocomposite scaffolds, our framework constructs a global hypergraph of 161,172 nodes and 320,201 hyperedges, revealing a scale-free topology (power law exponent ~1.23) organized around highly connected conceptual hubs. This representation prevents the combinatorial explosion typical of pairwise expansions and explicitly preserves the co-occurrence context of scientific formulations. We further demonstrate that equipping agentic systems with hypergraph traversal tools, specifically using node-intersection constraints, enables them to bridge semantically distant concepts. By exploiting these higher-order pathways, the system successfully generates grounded mechanistic hypotheses for novel composite materials, such as linking cerium oxide to PCL scaffolds via chitosan intermediates. This work establishes a "teacherless" agentic reasoning system where hypergraph topology acts as a verifiable guardrail, accelerating scientific discovery by uncovering relationships obscured by traditional graph methods.
Authors: Ao Sun, Xiaoyu Wang, Zhe Tan, Yu Li, Jiachen Zhu, Shu Su, Yuheng Jia
Abstract: As Large Language Models (LLMs) serve a global audience, alignment must transition from enforcing universal consensus to respecting cultural pluralism. We demonstrate that dense models, when forced to fit conflicting value distributions, suffer from \textbf{Mean Collapse}, converging to a generic average that fails to represent diverse groups. We attribute this to \textbf{Cultural Sparsity}, where gradient interference prevents dense parameters from spanning distinct cultural modes. To resolve this, we propose \textbf{\textsc{CuMA}} (\textbf{Cu}ltural \textbf{M}ixture of \textbf{A}dapters), a framework that frames alignment as a \textbf{conditional capacity separation} problem. By incorporating demographic-aware routing, \textsc{CuMA} internalizes a \textit{Latent Cultural Topology} to explicitly disentangle conflicting gradients into specialized expert subspaces. Extensive evaluations on WorldValuesBench, Community Alignment, and PRISM demonstrate that \textsc{CuMA} achieves state-of-the-art performance, significantly outperforming both dense baselines and semantic-only MoEs. Crucially, our analysis confirms that \textsc{CuMA} effectively mitigates mean collapse, preserving cultural diversity. Our code is available at https://github.com/Throll/CuMA.
Authors: Suyash Mishra, Qiang Li, Srikanth Patil, Satyanarayan Pati, Baddu Narendra
Abstract: Vision Language Models (VLMs) have shown strong performance on multimodal reasoning tasks, yet most evaluations focus on short videos and assume unconstrained computational resources. In industrial settings such as pharmaceutical content understanding, practitioners must process long-form videos under strict GPU, latency, and cost constraints, where many existing approaches fail to scale. In this work, we present an industrial GenAI framework that processes over 200,000 PDFs, 25,326 videos across eight formats (e.g., MP4, M4V, etc.), and 888 multilingual audio files in more than 20 languages. Our study makes three contributions: (i) an industrial large-scale architecture for multimodal reasoning in pharmaceutical domains; (ii) empirical analysis of over 40 VLMs on two leading benchmarks (Video-MME and MMBench) and proprietary dataset of 25,326 videos across 14 disease areas; and (iii) four findings relevant to long-form video reasoning: the role of multimodality, attention mechanism trade-offs, temporal reasoning limits, and challenges of video splitting under GPU constraints. Results show 3-8 times efficiency gains with SDPA attention on commodity GPUs, multimodality improving up to 8/12 task domains (especially length-dependent tasks), and clear bottlenecks in temporal alignment and keyframe detection across open- and closed-source VLMs. Rather than proposing a new "A+B" model, this paper characterizes practical limits, trade-offs, and failure patterns of current VLMs under realistic deployment constraints, and provide actionable guidance for both researchers and practitioners designing scalable multimodal systems for long-form video understanding in industrial domains.
Authors: Ziteng Wang, Yujie He, Guanliang Li, Siqi Yang, Jiaqi Xiong, Songxiang Liu
Abstract: Recent advancements in Multimodal Large Language Models (MLLMs) have demonstrated impressive performance on standard visual reasoning benchmarks. However, there is growing concern that these models rely excessively on linguistic shortcuts rather than genuine visual grounding, a phenomenon we term Text Bias. In this paper, we investigate the fundamental tension between visual perception and linguistic priors. We decouple the sources of this bias into two dimensions: Internal Corpus Bias, stemming from statistical correlations in pretraining, and External Instruction Bias, arising from the alignment-induced tendency toward sycophancy. To quantify this effect, we introduce V-FAT (Visual Fidelity Against Text-bias), a diagnostic benchmark comprising 4,026 VQA instances across six semantic domains. V-FAT employs a Three-Level Evaluation Framework that systematically increases the conflict between visual evidence and textual information: (L1) internal bias from atypical images, (L2) external bias from misleading instructions, and (L3) synergistic bias where both coincide. We introduce the Visual Robustness Score (VRS), a metric designed to penalize "lucky" linguistic guesses and reward true visual fidelity. Our evaluation of 12 frontier MLLMs reveals that while models excel in existing benchmarks, they experience significant visual collapse under high linguistic dominance.
Authors: Hongyi Li, William Ward Armstrong, Jun Xu
Abstract: We study rotation-robust learning for image inputs using Convolutional Model Trees (CMTs) [1], whose split and leaf coefficients can be structured on the image grid and transformed geometrically at deployment time. In a controlled MNIST setting with a rotation-invariant regression target, we introduce three geometry-aware inductive biases for split directions -- convolutional smoothing, a tilt dominance constraint, and importance-based pruning -- and quantify their impact on robustness under in-plane rotations. We further evaluate a deployment-time orientation search that selects a discrete rotation maximizing a forest-level confidence proxy without updating model parameters. Orientation search improves robustness under severe rotations but can be harmful near the canonical orientation when confidence is misaligned with correctness. Finally, we observe consistent trends on MNIST digit recognition implemented as one-vs-rest regression, highlighting both the promise and limitations of confidence-based orientation selection for model-tree ensembles.
Authors: Karim El Khoury, Maxime Zanella, Tiffanie Godelaine, Christophe De Vleeschouwer, Benoit Macq
Abstract: Audio-language models have recently demonstrated strong zero-shot capabilities by leveraging natural-language supervision to classify audio events without labeled training data. Yet, their performance is highly sensitive to the wording of text prompts, with small variations leading to large fluctuations in accuracy. Prior work has mitigated this issue through prompt learning or prompt ensembling. However, these strategies either require annotated data or fail to account for the fact that some prompts may negatively impact performance. In this work, we present an entropy-guided prompt weighting approach that aims to find a robust combination of prompt contributions to maximize prediction confidence. To this end, we formulate a tailored objective function that minimizes prediction entropy to yield new prompt weights, utilizing low-entropy as a proxy for high confidence. Our approach can be applied to individual samples or a batch of audio samples, requiring no additional labels and incurring negligible computational overhead. Experiments on five audio classification datasets covering environmental, urban, and vocal sounds, demonstrate consistent gains compared to classical prompt ensembling methods in a zero-shot setting, with accuracy improvements 5-times larger across the whole benchmark.
Authors: Milan Liepelt, Julien Baglio
Abstract: Quantum generative modeling is a very active area of research in looking for practical advantage in data analysis. Quantum generative adversarial networks (QGANs) are leading candidates for quantum generative modeling and have been applied to diverse areas, from high-energy physics to image generation. The latent style-based QGAN, relying on a classical variational autoencoder to encode the input data into a latent space and then using a style-based QGAN for data generation has been proven to be efficient for image generation or drug design, hinting at the use of far less trainable parameters than their classical counterpart to achieve comparable performance, however this advantage has never been systematically studied. We present in this work the first comprehensive experimental analysis of this advantage of QGANS applied to SAT4 image generation, obtaining an exponential advantage in capacity scaling for a quantum generator in the hybrid latent style-based QGAN architecture. Careful tuning of the autoencoder is crucial to obtain stable, reliable results. Once this tuning is performed and defining training optimality as when the training is stable and the FID score is low and stable as well, the optimal capacity (or number of trainable parameters) of the classical discriminator scales exponentially with respect to the capacity of the quantum generator, and the same is true for the capacity of the classical generator. This hints toward a type of quantum advantage for quantum generative modeling.
Authors: Xiaoyu Ma, David Patterson
Abstract: Large Language Model (LLM) inference is hard. The autoregressive Decode phase of the underlying Transformer model makes LLM inference fundamentally different from training. Exacerbated by recent AI trends, the primary challenges are memory and interconnect rather than compute. To address these challenges, we highlight four architecture research opportunities: High Bandwidth Flash for 10X memory capacity with HBM-like bandwidth; Processing-Near-Memory and 3D memory-logic stacking for high memory bandwidth; and low-latency interconnect to speedup communication. While our focus is datacenter AI, we also review their applicability for mobile devices.
Authors: Suyash Mishra, Qiang Li, Srikanth Patil, Anubhav Girdhar
Abstract: Vision Language Models (VLMs) are poised to revolutionize the digital transformation of pharmacyceutical industry by enabling intelligent, scalable, and automated multi-modality content processing. Traditional manual annotation of heterogeneous data modalities (text, images, video, audio, and web links), is prone to inconsistencies, quality degradation, and inefficiencies in content utilization. The sheer volume of long video and audio data further exacerbates these challenges, (e.g. long clinical trial interviews and educational seminars). Here, we introduce a domain adapted Video to Video Clip Generation framework that integrates Audio Language Models (ALMs) and Vision Language Models (VLMs) to produce highlight clips. Our contributions are threefold: (i) a reproducible Cut & Merge algorithm with fade in/out and timestamp normalization, ensuring smooth transitions and audio/visual alignment; (ii) a personalization mechanism based on role definition and prompt injection for tailored outputs (marketing, training, regulatory); (iii) a cost efficient e2e pipeline strategy balancing ALM/VLM enhanced processing. Evaluations on Video MME benchmark (900) and our proprietary dataset of 16,159 pharmacy videos across 14 disease areas demonstrate 3 to 4 times speedup, 4 times cost reduction, and competitive clip quality. Beyond efficiency gains, we also report our methods improved clip coherence scores (0.348) and informativeness scores (0.721) over state of the art VLM baselines (e.g., Gemini 2.5 Pro), highlighting the potential of transparent, custom extractive, and compliance supporting video summarization for life sciences.
Authors: Gorjan Radevski, Kiril Gashteovski, Giwon Hong, Carolin Lawrence, Goran Glava\v{s}
Abstract: Deploying LLMs in real-world applications requires controllable output that satisfies multiple desiderata at the same time. While existing work extensively addresses LLM steering for a single behavior, \textit{compositional steering} -- i.e., steering LLMs simultaneously towards multiple behaviors -- remains an underexplored problem. In this work, we propose \emph{compositional steering tokens} for multi-behavior steering. We first embed individual behaviors, expressed as natural language instructions, into dedicated tokens via self-distillation. Contrary to most prior work, which operates in the activation space, our behavior steers live in the space of input tokens, enabling more effective zero-shot composition. We then train a dedicated \textit{composition token} on pairs of behaviors and show that it successfully captures the notion of composition: it generalizes well to \textit{unseen} compositions, including those with unseen behaviors as well as those with an unseen \textit{number} of behaviors. Our experiments across different LLM architectures show that steering tokens lead to superior multi-behavior control compared to competing approaches (instructions, activation steering, and LoRA merging). Moreover, we show that steering tokens complement natural language instructions, with their combination resulting in further gains.
Authors: Jelmer van Lune, Stefano Mandija, Oscar van der Heide, Matteo Maspero, Martin B. Schilder, Jan Willem Dankbaar, Cornelis A. T. van den Berg, Alessandro Sbrizzi
Abstract: Magnetic resonance imaging (MRI) is a cornerstone of clinical neuroimaging, yet conventional MRIs provide qualitative information heavily dependent on scanner hardware and acquisition settings. While quantitative MRI (qMRI) offers intrinsic tissue parameters, the requirement for specialized acquisition protocols and reconstruction algorithms restricts its availability and impedes large-scale biomarker research. This study presents a self-supervised physics-guided deep learning framework to infer quantitative T1, T2, and proton-density (PD) maps directly from widely available clinical conventional T1-weighted, T2-weighted, and FLAIR MRIs. The framework was trained and evaluated on a large-scale, clinically heterogeneous dataset comprising 4,121 scan sessions acquired at our institution over six years on four different 3 T MRI scanner systems, capturing real-world clinical variability. The framework integrates Bloch-based signal models directly into the training objective. Across more than 600 test sessions, the generated maps exhibited white matter and gray matter values consistent with literature ranges. Additionally, the generated maps showed invariance to scanner hardware and acquisition protocol groups, with inter-group coefficients of variation $\leq$ 1.1%. Subject-specific analyses demonstrated excellent voxel-wise reproducibility across scanner systems and sequence parameters, with Pearson $r$ and concordance correlation coefficients exceeding 0.82 for T1 and T2. Mean relative voxel-wise differences were low across all quantitative parameters, especially for T2 ($<$ 6%). These results indicate that the proposed framework can robustly transform diverse clinical conventional MRI data into quantitative maps, potentially paving the way for large-scale quantitative biomarker research.
Authors: Aashi Garg, Aneshya Das, Arshi Arya, Anushka Goyal, Aditi
Abstract: The effectiveness of brand monitoring in India is increasingly challenged by the rise of Hinglish--a hybrid of Hindi and English--used widely in user-generated content on platforms like Twitter. Traditional Natural Language Processing (NLP) models, built for monolingual data, often fail to interpret the syntactic and semantic complexity of this code-mixed language, resulting in inaccurate sentiment analysis and misleading market insights. To address this gap, we propose a high-performance sentiment classification framework specifically designed for Hinglish tweets. Our approach fine-tunes mBERT (Multilingual BERT), leveraging its multilingual capabilities to better understand the linguistic diversity of Indian social media. A key component of our methodology is the use of subword tokenization, which enables the model to effectively manage spelling variations, slang, and out-of-vocabulary terms common in Romanized Hinglish. This research delivers a production-ready AI solution for brand sentiment tracking and establishes a strong benchmark for multilingual NLP in low-resource, code-mixed environments.
Authors: Nuoya Xiong, Yuhang Zhou, Hanqing Zeng, Zhaorun Chen, Furong Huang, Shuchao Bi, Lizhu Zhang, Zhuokai Zhao
Abstract: Large language models (LLMs) exhibit strengths across diverse domains. However, achieving strong performance across these domains with a single general-purpose model typically requires scaling to sizes that are prohibitively expensive to train and deploy. On the other hand, while smaller domain-specialized models are much more efficient, they struggle to generalize beyond their training distributions. To address this dilemma, we propose FusionRoute, a robust and effective token-level multi-LLM collaboration framework in which a lightweight router simultaneously (i) selects the most suitable expert at each decoding step and (ii) contributes a complementary logit that refines or corrects the selected expert's next-token distribution via logit addition. Unlike existing token-level collaboration methods that rely solely on fixed expert outputs, we provide a theoretical analysis showing that pure expert-only routing is fundamentally limited: unless strong global coverage assumptions hold, it cannot in general realize the optimal decoding policy. By augmenting expert selection with a trainable complementary generator, FusionRoute expands the effective policy class and enables recovery of optimal value functions under mild conditions. Empirically, across both Llama-3 and Gemma-2 families and diverse benchmarks spanning mathematical reasoning, code generation, and instruction following, FusionRoute outperforms both sequence- and token-level collaboration, model merging, and direct fine-tuning, while remaining competitive with domain experts on their respective tasks.
Authors: Knut Vanderbush, Melanie Weber
Abstract: Node coloring is the task of assigning colors to the nodes of a graph such that no two adjacent nodes have the same color, while using as few colors as possible. It is the most widely studied instance of graph coloring and of central importance in graph theory; major results include the Four Color Theorem and work on the Hadwiger-Nelson Problem. As an abstraction of classical combinatorial optimization tasks, such as scheduling and resource allocation, it is also rich in practical applications. Here, we focus on a relaxed version, approximate $k$-coloring, which is the task of assigning at most $k$ colors to the nodes of a graph such that the number of edges whose vertices have the same color is approximately minimized. While classical approaches leverage mathematical programming or SAT solvers, recent studies have explored the use of machine learning. We follow this route and explore the use of graph neural networks (GNNs) for node coloring. We first present an optimized differentiable algorithm that improves a prior approach by Schuetz et al. with orthogonal node feature initialization and a loss function that penalizes conflicting edges more heavily when their endpoints have higher degree; the latter inspired by the classical result that a graph is $k$-colorable if and only if its $k$-core is $k$-colorable. Next, we introduce a lightweight greedy local search algorithm and show that it may be improved by recursively computing a $(k-1)$-coloring to use as a warm start. We then show that applying such recursive warm starts to the GNN approach leads to further improvements. Numerical experiments on a range of different graph structures show that while the local search algorithms perform best on small inputs, the GNN exhibits superior performance at scale. The recursive warm start may be of independent interest beyond graph coloring for local search methods for combinatorial optimization.
Authors: Maximilian Alber, Timo Milbich, Alexandra Carpen-Amarie, Stephan Tietz, Jonas Dippel, Lukas Muttenthaler, Beatriz Perez Cancer, Alessandro Benetti, Panos Korfiatis, Elias Eulig, J\'er\^ome L\"uscher, Jiasen Wu, Sayed Abid Hashimi, Gabriel Dernbach, Simon Schallenberg, Neelay Shah, Moritz Kr\"ugener, Aniruddh Jammoria, Jake Matras, Patrick Duffy, Matt Redlon, Philipp Jurmeister, David Horst, Lukas Ruff, Klaus-Robert M\"uller, Frederick Klauschen, Andrew Norgan
Abstract: Pathology foundation models substantially advanced the possibilities in computational pathology -- yet tradeoffs in terms of performance, robustness, and computational requirements remained, which limited their clinical deployment. In this report, we present Atlas 2, Atlas 2-B, and Atlas 2-S, three pathology vision foundation models which bridge these shortcomings by showing state-of-the-art performance in prediction performance, robustness, and resource efficiency in a comprehensive evaluation across eighty public benchmarks. Our models were trained on the largest pathology foundation model dataset to date comprising 5.5 million histopathology whole slide images, collected from three medical institutions Charit\'e - Universt\"atsmedizin Berlin, LMU Munich, and Mayo Clinic.
Authors: Anastasiia Bakhmach, Paul Dufoss\'e, Andrea Vaglio, Florence Monville, Laurent Greillier, Fabrice Barl\'esi, S\'ebastien Benzekry
Abstract: Feature selection (FS) is essential for biomarker discovery and in the analysis of biomedical datasets. However, challenges such as high-dimensional feature space, low sample size, multicollinearity, and missing values make FS non-trivial. Moreover, FS performances vary across datasets and predictive tasks. We propose roofs, a Python package available at https://gitlab.inria.fr/compo/roofs, designed to help researchers in the choice of FS method adapted to their problem. Roofs benchmarks multiple FS methods on the user's data and generates reports that summarize a comprehensive set of evaluation metrics, including downstream predictive performance estimated using optimism correction, stability, reliability of individual features, and true positive and false positive rates assessed on semi-synthetic data with a simulated outcome. We demonstrate the utility of roofs on data from the PIONeeR clinical trial, aimed at identifying predictors of resistance to anti-PD-(L)1 immunotherapy in lung cancer. The PIONeeR dataset contained 374 multi-source blood and tumor biomarkers from 435 patients. A reduced subset of 214 features was obtained through iterative variance inflation factor pre-filtering. Of the 34 FS methods gathered in roofs, we evaluated 23 in combination with 11 classifiers (253 models in total) and identified a filter based on the union of Benjamini-Hochberg false discovery rate-adjusted p-values from t-test and logistic regression as the optimal approach, outperforming other methods including the widely used LASSO. We conclude that comprehensive benchmarking with roofs has the potential to improve the robustness and reproducibility of FS discoveries and increase the translational value of clinical models.
Authors: Alkis Kalavasis, Pravesh K. Kothari, Shuchen Li, Manolis Zampetakis
Abstract: In this work, we give a ${\rm poly}(d,k)$ time and sample algorithm for efficiently learning the parameters of a mixture of $k$ spherical distributions in $d$ dimensions. Unlike all previous methods, our techniques apply to heavy-tailed distributions and include examples that do not even have finite covariances. Our method succeeds whenever the cluster distributions have a characteristic function with sufficiently heavy tails. Such distributions include the Laplace distribution but crucially exclude Gaussians. All previous methods for learning mixture models relied implicitly or explicitly on the low-degree moments. Even for the case of Laplace distributions, we prove that any such algorithm must use super-polynomially many samples. Our method thus adds to the short list of techniques that bypass the limitations of the method of moments. Somewhat surprisingly, our algorithm does not require any minimum separation between the cluster means. This is in stark contrast to spherical Gaussian mixtures where a minimum $\ell_2$-separation is provably necessary even information-theoretically [Regev and Vijayaraghavan '17]. Our methods compose well with existing techniques and allow obtaining ''best of both worlds" guarantees for mixtures where every component either has a heavy-tailed characteristic function or has a sub-Gaussian tail with a light-tailed characteristic function. Our algorithm is based on a new approach to learning mixture models via efficient high-dimensional sparse Fourier transforms. We believe that this method will find more applications to statistical estimation. As an example, we give an algorithm for consistent robust mean estimation against noise-oblivious adversaries, a model practically motivated by the literature on multiple hypothesis testing. It was formally proposed in a recent Master's thesis by one of the authors, and has already inspired follow-up works.
Authors: Chengsong Huang, Tong Zheng, Langlin Huang, Jinyuan Li, Haolin Liu, Jiaxin Huang
Abstract: Large Language Models (LLMs) for complex reasoning is often hindered by high computational costs and latency, while resource-efficient Small Language Models (SLMs) typically lack the necessary reasoning capacity. Existing collaborative approaches, such as cascading or routing, operate at a coarse granularity by offloading entire queries to LLMs, resulting in significant computational waste when the SLM is capable of handling the majority of reasoning steps. To address this, we propose RelayLLM, a novel framework for efficient reasoning via token-level collaborative decoding. Unlike routers, RelayLLM empowers the SLM to act as an active controller that dynamically invokes the LLM only for critical tokens via a special command, effectively "relaying" the generation process. We introduce a two-stage training framework, including warm-up and Group Relative Policy Optimization (GRPO) to teach the model to balance independence with strategic help-seeking. Empirical results across six benchmarks demonstrate that RelayLLM achieves an average accuracy of 49.52%, effectively bridging the performance gap between the two models. Notably, this is achieved by invoking the LLM for only 1.07% of the total generated tokens, offering a 98.2% cost reduction compared to performance-matched random routers.
Authors: Zuhair Ahmed Khan Taha, Mohammed Mudassir Uddin, Shahnawaz Alam
Abstract: When researchers deploy large language models for autonomous tasks like reviewing literature or generating hypotheses, the computational bills add up quickly. A single research session using a 70-billion parameter model can cost around $127 in cloud fees, putting these tools out of reach for many academic labs. We developed AgentCompress to tackle this problem head-on. The core idea came from a simple observation during our own work: writing a novel hypothesis clearly demands more from the model than reformatting a bibliography. Why should both tasks run at full precision? Our system uses a small neural network to gauge how hard each incoming task will be, based only on its opening words, then routes it to a suitably compressed model variant. The decision happens in under a millisecond. Testing across 500 research workflows in four scientific fields, we cut compute costs by 68.3% while keeping 96.2% of the original success rate. For labs watching their budgets, this could mean the difference between running experiments and sitting on the sidelines
Authors: Navin Chhibber, Suneel Khemka, Navneet Kumar Tyagi, Rohit Tewari, Bireswar Banerjee, Piyush Ranjan
Abstract: Stock market price prediction is a significant interdisciplinary research domain that depends at the intersection of finance, statistics, and economics. Forecasting Accurately predicting stock prices has always been a focal point for various researchers. However, existing statistical approaches for time-series prediction often fail to effectively forecast the probability range of future stock prices. Hence, to solve this problem, the Neural Prophet with a Deep Neural Network (NP-DNN) is proposed to predict stock market prices. The preprocessing technique used in this research is Z-score normalization, which normalizes stock price data by removing scale differences, making patterns easier to detect. Missing value imputation fills gaps in historical data, enhancing the models use of complete information for more accurate predictions. The Multi-Layer Perceptron (MLP) learns complex nonlinear relationships among stock market prices and extracts hidden patterns from the input data, thereby creating meaningful feature representations for better prediction accuracy. The proposed NP-DNN model achieved an accuracy of 99.21% compared with other approaches using the Fused Large Language Model. Keywords: deep neural network, forecasting stock prices, multi-layer perceptron, neural prophet, stock market price prediction.
Authors: Maja Waldron
Abstract: One-shot prediction enables rapid adaptation of pretrained foundation models to new tasks using only one labeled example, but lacks principled uncertainty quantification. While conformal prediction provides finite-sample coverage guarantees, standard split conformal methods are inefficient in the one-shot setting due to data splitting and reliance on a single predictor. We propose Conformal Aggregation of One-Shot Predictors (CAOS), a conformal framework that adaptively aggregates multiple one-shot predictors and uses a leave-one-out calibration scheme to fully exploit scarce labeled data. Despite violating classical exchangeability assumptions, we prove that CAOS achieves valid marginal coverage using a monotonicity-based argument. Experiments on one-shot facial landmarking and RAFT text classification tasks show that CAOS produces substantially smaller prediction sets than split conformal baselines while maintaining reliable coverage.
Authors: James Rice
Abstract: I propose a novel framework that integrates stochastic differential equations (SDEs) with deep generative models to improve uncertainty quantification in machine learning applications involving structured and temporal data. This approach, termed Stochastic Latent Differential Inference (SLDI), embeds an It\^o SDE in the latent space of a variational autoencoder, allowing for flexible, continuous-time modeling of uncertainty while preserving a principled mathematical foundation. The drift and diffusion terms of the SDE are parameterized by neural networks, enabling data-driven inference and generalizing classical time series models to handle irregular sampling and complex dynamic structure. A central theoretical contribution is the co-parameterization of the adjoint state with a dedicated neural network, forming a coupled forward-backward system that captures not only latent evolution but also gradient dynamics. I introduce a pathwise-regularized adjoint loss and analyze variance-reduced gradient flows through the lens of stochastic calculus, offering new tools for improving training stability in deep latent SDEs. My paper unifies and extends variational inference, continuous-time generative modeling, and control-theoretic optimization, providing a rigorous foundation for future developments in stochastic probabilistic machine learning.
Authors: P. Gilda (Columbia University), P. Dungarwal (Columbia University), A. Thongkham (Columbia University), E. T. Ajayi (St John's University), S. Choudhary (Columbia University), T. M. Terol (Columbia University), C. Lam (Columbia University), J. P. Araujo (Columbia University), M. McFadyen-Mungalln (Columbia University), L. S. Liebovitch (Columbia University), P. T. Coleman (Columbia University), H. West (Columbia University), K. Sieck (Toyota Research Institute), S. Carter (Toyota Research Institute)
Abstract: We used machine learning and artificial intelligence: 1) to measure levels of peace in countries from news and social media and 2) to develop on-line tools that promote peace by helping users better understand their own media diet. For news media, we used neural networks to measure levels of peace from text embeddings of on-line news sources. The model, trained on one news media dataset also showed high accuracy when used to analyze a different news dataset. For social media, such as YouTube, we developed other models to measure levels of social dimensions important in peace using word level (GoEmotions) and context level (Large Language Model) methods. To promote peace, we note that 71% of people 20-40 years old daily view most of their news through short videos on social media. Content creators of these videos are biased towards creating videos with emotional activation, making you angry to engage you, to increase clicks. We developed and tested a Chrome extension, MirrorMirror, which provides real-time feedback to YouTube viewers about the peacefulness of the media they are watching. Our long term goal is for MirrorMirror to evolve into an open-source tool for content creators, journalists, researchers, platforms, and individual users to better understand the tone of their media creation and consumption and its effects on viewers. Moving beyond simple engagement metrics, we hope to encourage more respectful, nuanced, and informative communication.
Authors: Shih-Yang Liu, Xin Dong, Ximing Lu, Shizhe Diao, Peter Belcak, Mingjie Liu, Min-Hung Chen, Hongxu Yin, Yu-Chiang Frank Wang, Kwang-Ting Cheng, Yejin Choi, Jan Kautz, Pavlo Molchanov
Abstract: As language models become increasingly capable, users expect them to provide not only accurate responses but also behaviors aligned with diverse human preferences across a variety of scenarios. To achieve this, Reinforcement learning (RL) pipelines have begun incorporating multiple rewards, each capturing a distinct preference, to guide models toward these desired behaviors. However, recent work has defaulted to apply Group Relative Policy Optimization (GRPO) under multi-reward setting without examining its suitability. In this paper, we demonstrate that directly applying GRPO to normalize distinct rollout reward combinations causes them to collapse into identical advantage values, reducing the resolution of the training signal and resulting in suboptimal convergence and, in some cases, early training failure. We then introduce Group reward-Decoupled Normalization Policy Optimization (GDPO), a new policy optimization method to resolve these issues by decoupling the normalization of individual rewards, more faithfully preserving their relative differences and enabling more accurate multi-reward optimization, along with substantially improved training stability. We compare GDPO with GRPO across three tasks: tool calling, math reasoning, and coding reasoning, evaluating both correctness metrics (accuracy, bug ratio) and constraint adherence metrics (format, length). Across all settings, GDPO consistently outperforms GRPO, demonstrating its effectiveness and generalizability for multi-reward reinforcement learning optimization.
Authors: Zhen Qin, Zhishuai Liu, Pan Xu
Abstract: signSGD is popular in nonconvex optimization due to its communication efficiency. Yet, existing analyses typically assume data are sampled with replacement in each iteration, contradicting a common practical implementation where data are randomly reshuffled and sequentially fed into the algorithm. This gap leaves the theoretical understanding of the more practical algorithm, signSGD with random reshuffling (SignRR), largely unexplored. We develop the first analysis of SignRR to identify the core technical challenge that prevents a thorough convergence analysis of this method. In particular, given a dataset of size $n$ and $T$ epochs, we show that the expected gradient norm of SignRR is upper bounded by $O(\log(nT)/\sqrt{nT} + \sigma)$, where $\sigma$ is the averaged conditional mean square error that may not vanish. To tackle this limitation, we develop two new sign-based algorithms under random reshuffling: SignRVR, which incorporates variance-reduced gradients, and SignRVM, which integrates momentum-based updates. Both algorithms achieve a faster convergence rate of ${O}(\log(nT)/\sqrt{nT} +\log(nT)\sqrt{n}/\sqrt{T})$. We further extend our algorithms to a distributed setting, with a convergence rate of ${O}(\log(n_0T)/\sqrt{n_0T} +\log (n_0T)\sqrt{n_0}/\sqrt{T})$, where $n_0$ is the size of the dataset of a single machine. These results mark the first step towards the theoretical understanding of practical implementation of sign-based optimization algorithms. Finally, we back up our theoretical findings through experiments on simulated and real-world problems, verifying that randomly reshuffled sign methods match or surpass existing baselines.
Authors: Yun Young Choi, Sun Woo Park, Minho Lee, Youngho Woo
Abstract: Transformers have revolutionized performance in Natural Language Processing and Vision, paving the way for their integration with Graph Neural Networks (GNNs). One key challenge in enhancing graph transformers is strengthening the discriminative power of distinguishing isomorphisms of graphs, which plays a crucial role in boosting their predictive performances. To address this challenge, we introduce 'Topology-Informed Graph Transformer (TIGT)', a novel transformer enhancing both discriminative power in detecting graph isomorphisms and the overall performance of Graph Transformers. TIGT consists of four components: A topological positional embedding layer using non-isomorphic universal covers based on cyclic subgraphs of graphs to ensure unique graph representation: A dual-path message-passing layer to explicitly encode topological characteristics throughout the encoder layers: A global attention mechanism: And a graph information layer to recalibrate channel-wise graph features for better feature representation. TIGT outperforms previous Graph Transformers in classifying synthetic dataset aimed at distinguishing isomorphism classes of graphs. Additionally, mathematical analysis and empirical evaluations highlight our model's competitive edge over state-of-the-art Graph Transformers across various benchmark datasets.
Authors: Anuj Kumar Sirohi, Anjali Gupta, Sandeep Kumar, Amitabha Bagchi, Sayan Ranu
Abstract: Graph Neural Networks (GNNs) have demonstrated impressive performance across various tasks, leading to their increased adoption in high-stakes decision-making systems. However, concerns have arisen about GNNs potentially generating unfair decisions for underprivileged groups or individuals when lacking fairness constraints. This work addresses this issue by introducing GraphGini, a novel approach that incorporates the Gini coefficient to enhance both individual and group fairness within the GNN framework. We rigorously establish that the Gini coefficient offers greater robustness and promotes equal opportunity among GNN outcomes, advantages not afforded by the prevailing Lipschitz constant methodology. Additionally, we employ the Nash social welfare program to ensure our solution yields a Pareto optimal distribution of group fairness. Extensive experimentation on real-world datasets demonstrates GraphGini's efficacy in significantly improving individual fairness compared to state-of-the-art methods while maintaining utility and group fairness.
Authors: Martin Haugh, Raghav Singal
Abstract: The dishonest casino is a well-known hidden Markov model (HMM) often used in education to introduce HMMs and graphical models. A sequence of die rolls is observed with the casino switching between a fair and a loaded die. Instead of recovering the latent regime through filtering, smoothing, or the Viterbi algorithm, we ask a counterfactual question: how much of the gambler's winnings are caused by the casino's cheating? We introduce a class of structural causal models (SCMs) consistent with the HMM and define the expected winnings attributable to cheating (EWAC). Because EWAC is only partially identifiable, we bound it via linear programs (LPs). Numerical experiments help to develop intuition using benchmark SCMs based on independence, comonotonic, and countermonotonic copulas. Imposing a time homogeneity condition on the SCM yields tighter bounds, whereas relaxing it produces looser bounds that admit an explicit LP solution. Domain knowledge such as pathwise monotonicity or counterfactual stability can be incorporated through additional linear constraints. Finally, we show the time-averaged EWAC becomes fully identifiable as the number of time periods tends to infinity. Our work is the first to develop LP bounds for counterfactuals in an HMM setting, benefiting educational contexts where counterfactual inference is taught.
Authors: Liyi Zhang, Michael Y. Li, R. Thomas McCoy, Theodore R. Sumers, Jian-Qiao Zhu, Thomas L. Griffiths
Abstract: Autoregressive language models have demonstrated a remarkable ability to extract latent structure from text. The embeddings from large language models have been shown to capture aspects of the syntax and semantics of language. But what should embeddings represent? We connect the autoregressive prediction objective to the idea of constructing predictive sufficient statistics to summarize the information contained in a sequence of observations, and use this connection to identify three settings where the optimal content of embeddings can be identified: independent identically distributed data, where the embedding should capture the sufficient statistics of the data; latent state models, where the embedding should encode the posterior distribution over states given the data; and discrete hypothesis spaces, where the embedding should reflect the posterior distribution over hypotheses given the data. We then conduct empirical probing studies to show that transformers encode these three kinds of latent generating distributions, and that they perform well in out-of-distribution cases and without token memorization in these settings.
Authors: Mirko Nardi, Lorenzo Valerio, Andrea Passarella
Abstract: Federated Learning (FL) enables decentralized machine learning while preserving data privacy, making it ideal for sensitive applications where data cannot be shared. While FL has been widely studied in supervised contexts, its application to unsupervised learning remains underdeveloped. This work introduces FedCRef, a novel unsupervised federated learning method designed to uncover all underlying data distributions across decentralized clients without requiring labels. This task, known as Federated Clustering, presents challenges due to heterogeneous, non-uniform data distributions and the lack of centralized coordination. Unlike previous methods that assume a one-cluster-per-client setup or require prior knowledge of the number of clusters, FedCRef generalizes to multi-cluster-per-client scenarios. Clients iteratively refine their data partitions while discovering all distinct distributions in the system. The process combines local clustering, model exchange and evaluation via reconstruction error analysis, and collaborative refinement within federated groups of similar distributions to enhance clustering accuracy. Extensive evaluations on four public datasets (EMNIST, KMNIST, Fashion-MNIST and KMNIST49) show that FedCRef successfully identifies true global data distributions, achieving an average local accuracy of up to 95%. The method is also robust to noisy conditions, scalable, and lightweight, making it suitable for resource-constrained edge devices.
Authors: Georgios Is. Detorakis
Abstract: Deep learning has become a popular tool across many scientific fields, including the study of differential equations, particularly partial differential equations. This work introduces the basic principles of deep learning and the Deep Galerkin method, which uses deep neural networks to solve differential equations. This primer aims to provide technical and practical insights into the Deep Galerkin method and its implementation. We demonstrate how to solve the one-dimensional heat equation step-by-step. We also show how to apply the Deep Galerkin method to solve systems of ordinary differential equations and integral equations, such as the Fredholm of the second kind. Additionally, we provide code snippets within the text and the complete source code on Github. The examples are designed so that one can run them on a simple computer without needing a GPU.
Authors: Matthew Landers, Taylor W. Killian, Hugo Barnes, Thomas Hartvigsen, Afsaneh Doryab
Abstract: Offline reinforcement learning in high-dimensional, discrete action spaces is challenging due to the exponential scaling of the joint action space with the number of sub-actions and the complexity of modeling sub-action dependencies. Existing methods either exhaustively evaluate the action space, making them computationally infeasible, or factorize Q-values, failing to represent joint sub-action effects. We propose Branch Value Estimation (BraVE), a value-based method that uses tree-structured action traversal to evaluate a linear number of joint actions while preserving dependency structure. BraVE outperforms prior offline RL methods by up to $20\times$ in environments with over four million actions.
Authors: Kevin Black, Noah Brown, Danny Driess, Adnan Esmail, Michael Equi, Chelsea Finn, Niccolo Fusai, Lachy Groom, Karol Hausman, Brian Ichter, Szymon Jakubczak, Tim Jones, Liyiming Ke, Sergey Levine, Adrian Li-Bell, Mohith Mothukuri, Suraj Nair, Karl Pertsch, Lucy Xiaoyang Shi, James Tanner, Quan Vuong, Anna Walling, Haohuan Wang, Ury Zhilinsky
Abstract: Robot learning holds tremendous promise to unlock the full potential of flexible, general, and dexterous robot systems, as well as to address some of the deepest questions in artificial intelligence. However, bringing robot learning to the level of generality required for effective real-world systems faces major obstacles in terms of data, generalization, and robustness. In this paper, we discuss how generalist robot policies (i.e., robot foundation models) can address these challenges, and how we can design effective generalist robot policies for complex and highly dexterous tasks. We propose a novel flow matching architecture built on top of a pre-trained vision-language model (VLM) to inherit Internet-scale semantic knowledge. We then discuss how this model can be trained on a large and diverse dataset from multiple dexterous robot platforms, including single-arm robots, dual-arm robots, and mobile manipulators. We evaluate our model in terms of its ability to perform tasks in zero shot after pre-training, follow language instructions from people and from a high-level VLM policy, and its ability to acquire new skills via fine-tuning. Our results cover a wide variety of tasks, such as laundry folding, table cleaning, and assembling boxes.
Authors: Md Abrar Jahin, M. F. Mridha, Nilanjan Dey, Md. Jakir Hossen
Abstract: Feature selection is critical for improving the performance and interpretability of machine learning models, particularly in high-dimensional spaces where complex feature interactions can reduce accuracy and increase computational demands. Existing approaches often rely on static feature subsets or manual intervention, limiting adaptability and scalability. However, dynamic, per-instance feature selection methods and model-specific interpretability in reinforcement learning remain underexplored. This study proposes a human-in-the-loop (HITL) feature selection framework integrated into a Double Deep Q-Network (DDQN) using a Kolmogorov-Arnold Network (KAN). Our novel approach leverages simulated human feedback and stochastic distribution-based sampling, specifically Beta, to iteratively refine feature subsets per data instance, improving flexibility in feature selection. The KAN-DDQN achieved notable test accuracies of 93% on MNIST and 83% on FashionMNIST, outperforming conventional MLP-DDQN models by up to 9%. The KAN-based model provided high interpretability via symbolic representation while using 4 times fewer neurons in the hidden layer than MLPs did. Comparatively, the models without feature selection achieved test accuracies of only 58% on MNIST and 64% on FashionMNIST, highlighting significant gains with our framework. We further validate scalability on CIFAR-10 and CIFAR-100, achieving up to 30% relative macro F1 improvement on MNIST and 5% on CIFAR-10, while reducing calibration error by 25%. Complexity analysis confirms real-time feasibility with latency below 1 ms and parameter counts under 0.02M. Pruning and visualization further enhanced model transparency by elucidating decision pathways. These findings present a scalable, interpretable solution for feature selection that is suitable for applications requiring real-time, adaptive decision-making with minimal human oversight.
Authors: William Cappelletti, Pascal Frossard
Abstract: Representing and exploiting multivariate signals requires capturing relations between variables, which we can represent by graphs. Graph dictionaries allow to describe complex relational information as a sparse sum of simpler structures, but no prior model exists to infer such underlying structure elements from data. We define a novel Graph-Dictionary signal model, where a finite set of graphs characterizes relationships in data distribution as filters on the weighted sum of their Laplacians. We propose a framework to infer the graph dictionary representation from observed node signals, which allows to include a priori knowledge about signal properties, and about underlying graphs and their coefficients. We introduce a bilinear generalization of the primal-dual splitting algorithm to solve the learning problem. We show the capability of our method to reconstruct graphs from signals in multiple synthetic settings, where our model outperforms popular baselines. Then, we exploit graph-dictionary representations in an illustrative motor imagery decoding task on brain activity data, where we classify imagined motion better than standard methods relying on many more features. Our graph-dictionary model bridges a gap between sparse representations of multivariate data and a structured decomposition of sample-varying relationships into a sparse combination of elementary graph atoms.
Authors: Carlo Alfano, Silvia Sapora, Jakob Nicolaus Foerster, Patrick Rebeschini, Yee Whye Teh
Abstract: Evaluating preference optimization (PO) algorithms on LLM alignment is a challenging task that presents prohibitive costs, noise, and several variables like model size and hyper-parameters. In this work, we show that it is possible to gain insights on the efficacy of PO algorithm on simpler benchmarks. We design a diagnostic suite of MuJoCo tasks and datasets, which we use to systematically evaluate PO algorithms, establishing a more controlled and cheaper benchmark. We then propose a novel family of PO algorithms based on mirror descent, which we call Mirror Preference Optimization (MPO). Through evolutionary strategies, we search this class to discover algorithms specialized to specific properties of preference datasets, such as mixed-quality or noisy data. We demonstrate that our discovered PO algorithms outperform all known algorithms in the targeted MuJoCo settings. Finally, based on the insights gained from our MuJoCo experiments, we design a PO algorithm that significantly outperform existing baselines in an LLM alignment task.
Authors: Ryo Iwaki
Abstract: Regularization is a core component of recent Reinforcement Learning (RL) algorithms. Mirror Descent Value Iteration (MDVI) uses both Kullback-Leibler divergence and entropy as regularizers in its value and policy updates. Despite its empirical success in discrete action domains and strong theoretical guarantees, the performance of KL-entropy-regularized methods does not surpass that of a strong entropy-only-regularized method in continuous action domains. In this study, we propose Mirror Descent Actor Critic (MDAC) as an actor-critic style instantiation of MDVI for continuous action domains, and show that its empirical performance is significantly boosted by bounding the actor's log-density terms in the critic's loss function, compared to a non-bounded naive instantiation. Further, we relate MDAC to Advantage Learning by recalling that the actor's log-probability is equal to the regularized advantage function in tabular cases, and theoretically discuss when and why bounding the advantage terms is validated and beneficial. We also empirically explore effective choices for the bounding functions, and show that MDAC performs better than strong non-regularized and entropy-only-regularized methods with an appropriate choice of the bounding functions.
Authors: Vladimir Frants, Sos Agaian
Abstract: Adverse-weather image restoration (e.g., rain, snow, haze) models remain highly vulnerable to gradient-based white-box adversarial attacks, wherein minimal loss-aligned perturbations cause substantial degradation in the restored output. This paper presents QHNet, a computationally efficient purification-based defense that precedes the restoration network and targets perturbation suppression in the transform and quaternion domains. QHNet incorporates a Quaternion Hadamard Polynomial Denoising Block (QHPDB) and a Quaternion Denoising Residual Block (QDRB) within an encoder-decoder framework to remove high-frequency adversarial noise while preserving fine structural details. Robustness is evaluated using PSNR and SSIM across rain, snow, and haze removal tasks, and further validated under adaptive, defense-aware white-box attacks employing Projected Gradient Descent (PGD), Backward Pass Differentiable Approximation (BPDA), and Expectation Over Transformation (EOT). Experimental results demonstrate that QHNet delivers superior restoration fidelity and significantly improved robustness compared to state-of-the-art purification baselines, confirming its effectiveness for low-level vision pipelines.
Authors: Jiayi Fu, Xuandong Zhao, Chengyuan Yao, Heng Wang, Qi Han, Yanghua Xiao
Abstract: Reinforcement Learning from Human Feedback (RLHF) is essential for aligning large language models (LLMs) with human values. However, RLHF is susceptible to \emph{reward hacking}, where the agent exploits flaws in the reward function rather than learning the intended behavior, thus degrading alignment. Although reward shaping helps stabilize RLHF and partially mitigate reward hacking, a systematic investigation into shaping techniques and their underlying principles remains lacking. To bridge this gap, we present a comprehensive study of the prevalent reward shaping methods. Our analysis suggests two key design principles: (1) the RL reward should be bounded, and (2) the RL reward benefits from rapid initial growth followed by gradual convergence. Guided by these insights, we propose Preference As Reward (PAR), a novel approach that leverages the latent preferences embedded within the reward model as the signal for reinforcement learning. Moreover, PAR exhibits two critical variance-reduction properties that contribute to stabilizing the RLHF training process and effectively extending the tolerance window for early stopping. We evaluated PAR on the base model Gemma2-2B using two datasets, Ultrafeedback-Binarized and HH-RLHF. Experimental results demonstrate PAR's superior performance over other reward shaping methods. On the AlpacaEval 2.0 benchmark, PAR achieves a win rate of at least 5 percentage points higher than competing approaches. Furthermore, PAR exhibits remarkable data efficiency, requiring only a single reference reward for optimal performance, and maintains robustness against reward hacking even after two full epochs of training. The code is available at https://github.com/PorUna-byte/PAR.
Authors: Sahar Admoni, Assaf Hallak, Yftah Ziser, Omer Ben-Porat, Ofra Amir
Abstract: Explaining reinforcement learning agents is challenging because policies emerge from complex reward structures and neural representations that are difficult for humans to interpret. Existing approaches often rely on curated demonstrations that expose local behaviors but provide limited insight into an agent's global strategy, leaving users to infer intent from raw observations. We propose SySLLM (Synthesized Summary using Large Language Models), a framework that reframes policy interpretation as a language-generation problem. Instead of visual demonstrations, SySLLM converts spatiotemporal trajectories into structured text and prompts an LLM to generate coherent summaries describing the agent's goals, exploration style, and decision patterns. SySLLM scales to long-horizon, semantically rich environments without task-specific fine-tuning, leveraging LLM world knowledge and compositional reasoning to capture latent behavioral structure across policies. Expert evaluations show strong alignment with human analyses, and a large-scale user study found that 75.5% of participants preferred SySLLM summaries over state-of-the-art demonstration-based explanations. Together, these results position abstractive textual summarization as a paradigm for interpreting complex RL behavior.
Authors: Jihyun Lim, Junhyuk Jo, Tuo Zhang, Sunwoo Lee
Abstract: Online Knowledge Distillation (KD) is recently highlighted to train large models in Federated Learning (FL) environments. Many existing studies adopt the logit ensemble method to perform KD on the server side. However, they often assume that unlabeled data collected at the edge is centralized on the server. Moreover, the logit ensemble method personalizes local models, which can degrade the quality of soft targets, especially when data is highly non-IID. To address these critical limitations,we propose a novel on-device KD-based heterogeneous FL method. Our approach leverages a small auxiliary model to learn from labeled local data. Subsequently, a subset of clients with strong system resources transfers knowledge to a large model through on-device KD using their unlabeled data. Our extensive experiments demonstrate that our on-device KD-based heterogeneous FL method effectively utilizes the system resources of all edge devices as well as the unlabeled data, resulting in higher accuracy compared to SOTA KD-based FL methods.
Authors: Zinuo You, John Cartlidge, Karen Elliott, Menghan Ge, Daniel Gold
Abstract: Existing black-box portfolio management systems are prevalent in the financial industry due to commercial and safety constraints, though their performance can fluctuate dramatically with changing market regimes. Evaluating these non-transparent systems is computationally expensive, as fixed budgets limit the number of possible observations. Therefore, achieving stable and sample-efficient optimization for these systems has become a critical challenge. This work presents a novel Bayesian optimization framework (TPE-AS) that improves search stability and efficiency for black-box portfolio models under these limited observation budgets. Standard Bayesian optimization, which solely maximizes expected return, can yield erratic search trajectories and misalign the surrogate model with the true objective, thereby wasting the limited evaluation budget. To mitigate these issues, we propose a weighted Lagrangian estimator that leverages an adaptive schedule and importance sampling. This estimator dynamically balances exploration and exploitation by incorporating both the maximization of model performance and the minimization of the variance of model observations. It guides the search from broad, performance-seeking exploration towards stable and desirable regions as the optimization progresses. Extensive experiments and ablation studies, which establish our proposed method as the primary approach and other configurations as baselines, demonstrate its effectiveness across four backtest settings with three distinct black-box portfolio management models.
Authors: Matthew Landers, Taylor W. Killian, Thomas Hartvigsen, Afsaneh Doryab
Abstract: The combinatorial structure of many real-world action spaces leads to exponential growth in the number of possible actions, limiting the effectiveness of conventional reinforcement learning algorithms. Recent approaches for combinatorial action spaces impose factorized or sequential structures over sub-actions, failing to capture complex joint behavior. We introduce the Sub-Action Interaction Network using Transformers (SAINT), a novel policy architecture that represents multi-component actions as unordered sets and models their dependencies via self-attention conditioned on the global state. SAINT is permutation-invariant, sample-efficient, and compatible with standard policy optimization algorithms. In 20 distinct combinatorial environments across three task domains, including environments with nearly 17 million joint actions, SAINT consistently outperforms strong baselines.
Authors: Jizhou Guo, Zhaomin Wu, Hanchen Yang, Philip S. Yu
Abstract: Best-of-N sampling is a powerful method for improving Large Language Model (LLM) performance, but it is often limited by its dependence on massive, text-based reward models. These models are not only computationally expensive but also data-hungry, requiring extensive labeled datasets for training. This creates a significant data challenge, as they overlook a rich, readily available data source: the LLM's own internal hidden states. To address this data and efficiency gap, we introduce SWIFT (Simple Weighted Intrinsic Feedback Technique), a novel and lightweight method that learns a reward function directly from the rich information embedded in LLM hidden states. Operating at the token embedding level, SWIFT employs simple linear layers to effectively distinguish between preferred and dispreferred generations, eliminating the need for computationally intensive text-based modeling. Extensive experiments on standard benchmarks show that SWIFT outperforms existing baselines (12.7% higher accuracy than EurusRM-7B on MATH dataset) while using less than 0.005% of their parameters. Its robust scalability, compatibility with certain closed-source models via logit access, and ability to combine with traditional reward models for additional performance highlight SWIFT's practical value and contribution to more efficient data-driven LLM post-training. Our code is available at https://github.com/aster2024/SWIFT .
Authors: Antoine Moulin, Gergely Neu, Luca Viano
Abstract: We study the problem of offline imitation learning in Markov decision processes (MDPs), where the goal is to learn a well-performing policy given a dataset of state-action pairs generated by an expert policy. Complementing a recent line of work on this topic that assumes the expert belongs to a tractable class of known policies, we approach this problem from a new angle and leverage a different type of structural assumption about the environment. Specifically, for the class of linear $Q^\pi$-realizable MDPs, we introduce a new algorithm called saddle-point offline imitation learning (\SPOIL), which is guaranteed to match the performance of any expert up to an additive error $\varepsilon$ with access to $\mathcal{O}(\varepsilon^{-2})$ samples. Moreover, we extend this result to possibly nonlinear $Q^\pi$-realizable MDPs at the cost of a worse sample complexity of order $\mathcal{O}(\varepsilon^{-4})$. Finally, our analysis suggests a new loss function for training critic networks from expert data in deep imitation learning. Empirical evaluations on standard benchmarks demonstrate that the neural net implementation of \SPOIL is superior to behavior cloning and competitive with state-of-the-art algorithms.
Authors: Gen Li, Changxiao Cai
Abstract: Diffusion models have emerged as a powerful paradigm for modern generative modeling, demonstrating strong potential for large language models (LLMs). Unlike conventional autoregressive (AR) models that generate tokens sequentially, diffusion models allow for parallel sampling, offering a promising path to accelerate generation and eliminate the left-to-right generation constraints. Despite their empirical success, theoretical understandings of diffusion language models remain underdeveloped. In this work, we develop convergence guarantees for diffusion language models from an information-theoretic perspective. Our analysis demonstrates that the sampling error, measured by the Kullback-Leibler (KL) divergence, decays inversely with the number of iterations $T$ and scales linearly with the mutual information between tokens in the target text sequence. Crucially, our theory covers the regime $T
Authors: Anum Nawaz, Muhammad Irfan, Xianjia Yu, Hamad Aldawsari, Rayan Hamza Alsisi, Zhuo Zou, Tomi Westerlund
Abstract: Federated learning (FL) is increasingly recognised for addressing security and privacy concerns in traditional cloud-centric machine learning (ML), particularly within personalised health monitoring such as wearable devices. By enabling global model training through localised policies, FL allows resource-constrained wearables to operate independently. However, conventional first-order FL approaches face several challenges in personalised model training due to the heterogeneous non-independent and identically distributed (non-iid) data by each individual's unique physiology and usage patterns. Recently, second-order FL approaches maintain the stability and consistency of non-iid datasets while improving personalised model training. This study proposes and develops a verifiable and auditable optimised second-order FL framework BFEL (blockchain enhanced federated edge learning) based on optimised FedCurv for personalised healthcare systems. FedCurv incorporates information about the importance of each parameter to each client's task (through fisher information matrix) which helps to preserve client-specific knowledge and reduce model drift during aggregation. Moreover, it minimizes communication rounds required to achieve a target precision convergence for each client device while effectively managing personalised training on non-iid and heterogeneous data. The incorporation of ethereum-based model aggregation ensures trust, verifiability, and auditability while public key encryption enhances privacy and security. Experimental results of federated CNNs and MLPs utilizing mnist, cifar-10, and PathMnist demonstrate framework's high efficiency, scalability, suitability for edge deployment on wearables, and significant reduction in communication cost.
Authors: Antoine Moulin, Emmanuel Esposito, Dirk van der Hoeven
Abstract: We consider the problem setting of prediction with expert advice with possibly heavy-tailed losses, i.e. the only assumption on the losses is an upper bound on their second moments, denoted by $\theta$. We develop adaptive algorithms that do not require any prior knowledge about the range or the second moment of the losses. Existing adaptive algorithms have what is typically considered a lower-order term in their regret guarantees. We show that this lower-order term, which is often the maximum of the losses, can actually dominate the regret bound in our setting. Specifically, we show that even with small constant $\theta$, this lower-order term can scale as $\sqrt{KT}$, where $K$ is the number of experts and $T$ is the time horizon. We propose adaptive algorithms with improved regret bounds that avoid the dependence on such a lower-order term and guarantee $\mathcal{O}(\sqrt{\theta T\log(K)})$ regret in the worst case, and $\mathcal{O}(\theta \log(KT)/\Delta_{\min})$ regret when the losses are sampled i.i.d. from some fixed distribution, where $\Delta_{\min}$ is the difference between the mean losses of the second best expert and the best expert. Additionally, when the loss function is the squared loss, our algorithm also guarantees improved regret bounds over prior results.
Authors: Fei Li, Song Liu, Weiguo Wu, Shiqiang Nie, Jinyu Wang
Abstract: The high memory demands of the Key-Value (KV) Cache during the inference of Large Language Models (LLMs) severely restrict their deployment in resource-constrained platforms. Quantization can effectively alleviate the memory pressure caused by KV Cache. However, existing methods either rely on static one-size-fits-all precision allocation or fail to dynamically prioritize critical KV in long-context tasks, forcing memory-accuracy-throughput tradeoffs. In this work, we propose a novel mixed-precision quantization method for KV Cache named KVmix. KVmix leverages gradient-based importance analysis to evaluate how individual Key and Value projection matrices affect the model loss, enabling layer-specific bit-width allocation for mix-precision quantization. It dynamically prioritizes higher precision for important layers while aggressively quantizing less influential ones, achieving a tunable balance between accuracy and efficiency. KVmix also introduces a dynamic long-context optimization strategy that adaptively keeps full-precision KV pairs for recent pivotal tokens and compresses older ones, achieving high-quality sequence generation with low memory usage. Additionally, KVmix provides efficient low-bit quantization and CUDA kernels to optimize computational overhead. On LLMs such as Llama and Mistral, KVmix achieves near-lossless inference performance with extremely low quantization configuration (Key 2.19bit Value 2.38bit), while delivering a remarkable 4.9x memory compression and a 5.3x speedup in inference throughput.
Authors: Cooper Doyle, Rebecca Chan, Andy Hu, Anna Leontjeva
Abstract: Low-rank adaptation methods enable efficient task-specific updates in large neural networks, but provide no principled mechanism for uncertainty estimation or capacity control. We introduce Low-Rank Variational Dropout (LRVD), a Bayesian framework that operates directly in the space of low-rank adaptation. LRVD employs a scale-invariant, sparsity-inducing prior together with a structured variational family that ties uncertainty at the level of latent rank components, inducing rank-wise noise-to-signal ratios for automatic capacity selection. As a concrete instantiation, we apply LRVD to low-rank adaptation and obtain BayesLoRA, which jointly learns predictive uncertainty and the effective adapter rank with only O(r) additional parameters, where r is the adapter rank. We empirically show that BayesLoRA induces stable, non-arbitrary rank structure aligned with the intrinsic singular directions of the learned updates, and outperforms existing low-rank sparsification methods in accuracy at comparable training cost while delivering substantially improved predictive calibration at negligible additional overhead.
Authors: Harsh Sharma, Juan Diego Draxl Giannoni, Boris Kramer
Abstract: This work presents structure-preserving Lift & Learn, a scientific machine learning method that employs lifting variable transformations to learn structure-preserving reduced-order models for nonlinear partial differential equations (PDEs) with conservation laws. We propose a hybrid learning approach based on a recently developed energy-quadratization strategy that uses knowledge of the nonlinearity at the PDE level to derive an equivalent quadratic lifted system with quadratic system energy. The lifted dynamics obtained via energy quadratization are linear in the old variables, making model learning very effective in the lifted setting. Based on the lifted quadratic PDE model form, the proposed method derives quadratic reduced terms analytically and then uses those derived terms to formulate a constrained optimization problem to learn the remaining linear reduced operators in a structure-preserving way. The proposed hybrid learning approach yields computationally efficient quadratic reduced-order models that respect the underlying physics of the high-dimensional problem. We demonstrate the generalizability of quadratic models learned via the proposed structure-preserving Lift & Learn method through three numerical examples: the one-dimensional wave equation with exponential nonlinearity, the two-dimensional sine-Gordon equation, and the two-dimensional Klein-Gordon-Zakharov equations. The numerical results show that the proposed learning approach is competitive with the state-of-the-art structure-preserving data-driven model reduction method in terms of both accuracy and computational efficiency.
Authors: Wenhao Zeng, Yaoning Wang, Chao Hu, Yuling Shi, Chengcheng Wan, Hongyu Zhang, Xiaodong Gu
Abstract: Large Reasoning Models (LRMs) have demonstrated remarkable capabilities by scaling up the length of Chain-of-Thought (CoT). However, excessively long reasoning traces pose substantial challenges for training cost and inference latency. While various CoT compression approaches have emerged to address this challenge, they face inherent trade-offs: token-level methods often disrupt syntactic and logical coherence, while step-level methods based on perplexity fail to reliably capture the logically critical reasoning steps because of the dilution of logical information. In this paper, we propose ASAP (Anchor-guided, SurprisAl-based Pruning), a novel coarse-to-fine framework for CoT compression. ASAP first performs anchor-guided pruning to preserve the core reasoning structure, which efficiently reduces the search space for subsequent processing. Leveraging the insight that logical branching choices are concentrated at the onset of reasoning steps, it then enables logic-aware pruning by selecting logically essential reasoning steps based on a novel first-token surprisal metric. Finally, ASAP distills the models to autonomously generate and leverage these concise CoTs at inference time, enabling efficient reasoning. Experiments show that ASAP achieves state-of-the-art accuracy across multiple benchmarks while substantially reducing training and inference costs.
Authors: Vincent-Daniel Yun
Abstract: Low-precision training has become crucial for reducing the computational and memory costs of large-scale deep learning. However, quantizing gradients introduces magnitude shrinkage, which can change how stochastic gradient descent (SGD) converges. In this study, we explore SGD convergence under a gradient shrinkage model, where each stochastic gradient is scaled by a factor \( q_k \in (0,1] \). We show that this shrinkage affect the usual stepsize \( \mu_k \) with an effective stepsize \( \mu_k q_k \), slowing convergence when \( q_{\min} < 1 \). With typical smoothness and bounded-variance assumptions, we prove that low-precision SGD still converges, but at a slower pace set by \( q_{\min} \), and with a higher steady error level due to quantization effects. We analyze theoretically how lower numerical precision slows training by treating it as gradient shrinkage within the standard SGD convergence setup.
Authors: Xvyuan Liu, Xiangfei Qiu, Hanyin Cheng, Xingjian Wu, Chenjuan Guo, Bin Yang, Jilin Hu
Abstract: Irregular multivariate time series (IMTS) are prevalent in critical domains like healthcare and finance, where accurate forecasting is vital for proactive decision-making. However, the asynchronous sampling and irregular intervals inherent to IMTS pose two core challenges for existing methods: (1) how to accurately represent the raw information of irregular time series without introducing data distortion, and (2) how to effectively capture the complex dynamic dependencies between observation points. To address these challenges, we propose the Adaptive Spatio-Temporal Graph Interaction (ASTGI) framework. Specifically, the framework first employs a Spatio-Temporal Point Representation module to encode each discrete observation as a point within a learnable spatio-temporal embedding space. Second, a Neighborhood-Adaptive Graph Construction module adaptively builds a causal graph for each point in the embedding space via nearest neighbor search. Subsequently, a Spatio-Temporal Dynamic Propagation module iteratively updates information on these adaptive causal graphs by generating messages and computing interaction weights based on the relative spatio-temporal positions between points. Finally, a Query Point-based Prediction module generates the final forecast by aggregating neighborhood information for a new query point and performing regression. Extensive experiments on multiple benchmark datasets demonstrate that ASTGI outperforms various state-of-the-art methods.
Authors: Albert Vong, Steven Henke, Oliver Hoidn, Hanna Ruth, Junjing Deng, Alexander Hexemer, David Shapiro, Apurva Mehta, Arianna Gleason, Levi Hancock, Nicholas Schwarz
Abstract: X-ray ptychography is a data-intensive imaging technique expected to become ubiquitous at next-generation light sources delivering many-fold increases in coherent flux. The need for real-time feedback under accelerated acquisition rates motivates surrogate reconstruction models like deep neural networks, which offer orders-of-magnitude speedup over conventional methods. However, existing deep learning approaches lack robustness across diverse experimental conditions. We propose an unsupervised training workflow emphasizing probe learning by combining experimentally-measured probes with synthetic, procedurally generated objects. This probe-centric approach enables a single physics-informed neural network to reconstruct unseen experiments across multiple beamlines; among the first demonstrations of multi-probe generalization. We find probe learning is equally important as in-distribution learning; models trained using this synthetic workflow achieve reconstruction fidelity comparable to those trained exclusively on experimental data, even when changing the type of synthetic training object. The proposed approach enables training of experiment-steering models that provide real-time feedback under dynamic experimental conditions.
Authors: Willa Potosnak, Malcolm Wolff, Mengfei Cao, Ruijun Ma, Tatiana Konstantinova, Dmitry Efimov, Michael W. Mahoney, Boris Oreshkin, Kin G. Olivares
Abstract: While accuracy is a critical requirement for time series forecasting, an equally important desideratum is forecast stability across forecast creation dates (FCDs). Even highly accurate models can produce erratic revisions between FCDs, disrupting downstream decision-making. To improve forecast stability of such revisions, several state-of-the-art models including MQCNN, MQT, and SPADE employ a powerful yet underexplored neural network architectural design known as forking-sequences. This architectural design jointly encodes and decodes the entire time series across all FCDs, producing an entire multi-horizon forecast grid in a single forward pass. This approach contrasts with conventional neural forecasting methods that process FCDs independently, generating only a single multi-horizon forecast per forward pass. In this work, we formalize the forking-sequences design and motivate its broader adoption by introducing a metric for quantifying excess volatility in forecast revisions and by providing theoretical and empirical analysis. We theoretically motivate three key benefits of forking-sequences: (i) increased forecast stability through ensembling; (ii) gradient variance reduction, leading to more stable and consistent training steps; and (iii) improved computational efficiency during inference. We validate the benefits of forking-sequences compared to baseline window-sampling on the M-series benchmark, using 16 datasets from the M1, M3, M4, and Tourism competitions. We observe median accuracy improvements across datasets of 29.7%, 46.2%, 49.3%, 28.6%, 24.7%, and 6.4% for MLP, RNN, LSTM, CNN, Transformer, and StateSpace-based architectures, respectively. We then show that forecast ensembling during inference can improve median forecast stability by 10.8%, 13.2%, 13.0%, 10.9%, 10.2%, and 11.2% for these respective models trained with forking-sequences, while maintaining accuracy.
Authors: Shen Dong, Mingxuan Zhang, Pengfei He, Li Ma, Bhavani Thuraisingham, Hui Liu, Yue Xing
Abstract: Large Language Model (LLM)-based Multi-Agent Systems (MAS) have emerged as a powerful paradigm for tackling complex, multi-step tasks across diverse domains. However, despite their impressive capabilities, MAS remain susceptible to adversarial manipulation. Existing studies typically examine isolated attack surfaces or specific scenarios, leaving a lack of holistic understanding of MAS vulnerabilities. To bridge this gap, we introduce PEAR, a benchmark for systematically evaluating both the utility and vulnerability of planner-executor MAS. While compatible with various MAS architectures, our benchmark focuses on the planner-executor structure, which is a practical and widely adopted design. Through extensive experiments, we find that (1) a weak planner degrades overall clean task performance more severely than a weak executor; (2) while a memory module is essential for the planner, having a memory module for the executor does not impact the clean task performance; (3) there exists a trade-off between task performance and robustness; and (4) attacks targeting the planner are particularly effective at misleading the system. These findings offer actionable insights for enhancing the robustness of MAS and lay the groundwork for principled defenses in multi-agent settings.
Authors: Jianqing Zhang, Zhezheng Hao, Wei Xia, Hande Dong, Hong Wang, Chenxing Wei, Yuyan Zhou, Yubin Qi, Qiang Lin, Jian Cao
Abstract: Reinforcement learning (RL) is widely used for post-training large language models (LLMs) in code editing, where group-relative methods, such as GRPO, are popular due to their critic-free and normalized advantage estimation. However, in real-world code-editing scenarios, reward distributions are often skewed with unpredictable noise, leading to distorted advantage computation and increased rollout outliers. To address this issue, we propose Group Adaptive Policy Optimization (GAPO), which adaptively finds an interval with the highest SNR (Signal to Noise Ratio) per prompt and uses the median of that interval as an adaptive Q to replace the group mean in advantage calculation to reduce noise further. This adaptive Q robustly handles rollout noise while remaining plug-and-play and efficient. We evaluate GAPO on nine instruction-tuned LLMs (3B-14B) using a collected large dataset of 51,844 real-world, history-aware code-editing tasks spanning 10 programming languages. GAPO yields up to 4.35 in-domain (ID) and 5.30 out-of-domain (OOD) exact-match improvements over GRPO and its variant DAPO, while achieving lower clipping ratios and higher GPU throughput. Code: https://github.com/TsingZ0/verl-GAPO.
Authors: Shuang Wang, Xuben Wang, Fei Deng, Peifan Jiang, Jian Chen, Gianluca Fiandaca
Abstract: Electromagnetic methods have become one of the most widely used techniques in geological exploration. With the remarkable success of deep learning, applying such techniques to EM methods has emerged as a promising research direction to overcome the limitations of conventional approaches. The effectiveness of deep learning methods depends heavily on the quality of datasets, which directly influences model performance and generalization ability. Existing application studies often construct datasets from random one-dimensional or structurally simple three-dimensional models, which fail to represent the real geological environments. Furthermore, the absence of standardized, publicly 3D geoelectric datasets continues to hinder progress in deep learning based EM exploration. To address these limitations, we present OpenEM, a large-scale, multi-structural three-dimensional geoelectric dataset that encompasses a broad range of geologically plausible subsurface structures. OpenEM consists of nine categories of geoelectric models, spanning from simple configurations with anomalous bodies in half-space to more complex structures such as flat layers, folded layers, flat faults, curved faults, and their corresponding variants with anomalous bodies. Since three-dimensional forward modeling in electromagnetics is extremely time-consuming, we further developed a deep learning based fast forward modeling approach for OpenEM, enabling efficient and reliable forward modeling across the entire dataset. This capability allows OpenEM to be rapidly deployed for a wide range of tasks. OpenEM provides a unified, comprehensive, and large-scale dataset for common EM exploration systems to accelerate the application of deep learning in electromagnetic methods.The complete dataset is publicly available at https://doi.org/10.5281/zenodo.17141981.
Authors: Han Wu, Jie Yin
Abstract: Few-shot knowledge graph relational learning seeks to perform reasoning over relations given only a limited number of training examples. While existing approaches largely adopt a meta-learning framework for enabling fast adaptation to new relations, they suffer from two key pitfalls. First, they learn relation meta-knowledge in isolation, failing to capture common relational patterns shared across tasks. Second, they struggle to effectively incorporate local, task-specific contexts crucial for rapid adaptation. To address these limitations, we propose MoEMeta, a novel meta-learning framework that disentangles globally shared knowledge from task-specific contexts to enable both effective model generalization and rapid adaptation. MoEMeta introduces two key innovations: (i) a mixture-of-experts (MoE) model that learns globally shared relational prototypes to enhance generalization, and (ii) a task-tailored adaptation mechanism that captures local contexts for fast task-specific adaptation. By balancing global generalization with local adaptability, MoEMeta significantly advances few-shot relational learning. Extensive experiments and analyses on three KG benchmarks show that MoEMeta consistently outperforms existing baselines, achieving state-of-the-art performance.
Authors: Chaymae Yahyati, Ismail Lamaakal, Khalid El Makkaoui, Ibrahim Ouahbi, Yassine Maleh
Abstract: We introduce Simplex-FEM Networks (SiFEN), a learned piecewise-polynomial predictor that represents f: R^d -> R^k as a globally C^r finite-element field on a learned simplicial mesh in an optionally warped input space. Each query activates exactly one simplex and at most d+1 basis functions via barycentric coordinates, yielding explicit locality, controllable smoothness, and cache-friendly sparsity. SiFEN pairs degree-m Bernstein-Bezier polynomials with a light invertible warp and trains end-to-end with shape regularization, semi-discrete OT coverage, and differentiable edge flips. Under standard shape-regularity and bi-Lipschitz warp assumptions, SiFEN achieves the classic FEM approximation rate M^(-m/d) with M mesh vertices. Empirically, on synthetic approximation tasks, tabular regression/classification, and as a drop-in head on compact CNNs, SiFEN matches or surpasses MLPs and KANs at matched parameter budgets, improves calibration (lower ECE/Brier), and reduces inference latency due to geometric locality. These properties make SiFEN a compact, interpretable, and theoretically grounded alternative to dense MLPs and edge-spline networks.
Authors: Muhammad Kashif, Shaf Khalid, Alberto Marchisio, Nouhaila Innan, Muhammad Shafique
Abstract: Hybrid Quantum Neural Networks (HQNNs), which combine parameterized quantum circuits with classical neural layers, are emerging as promising models in the noisy intermediate-scale quantum (NISQ) era. While quantum circuits are not naturally measured in floating point operations (FLOPs), most HQNNs (in NISQ era) are still trained on classical simulators where FLOPs directly dictate runtime and scalability. Hence, FLOPs represent a practical and viable metric to measure the computational complexity of HQNNs. In this work, we introduce FAQNAS, a FLOPs-aware neural architecture search (NAS) framework that formulates HQNN design as a multi-objective optimization problem balancing accuracy and FLOPs. Unlike traditional approaches, FAQNAS explicitly incorporates FLOPs into the optimization objective, enabling the discovery of architectures that achieve strong performance while minimizing computational cost. Experiments on five benchmark datasets (MNIST, Digits, Wine, Breast Cancer, and Iris) show that quantum FLOPs dominate accuracy improvements, while classical FLOPs remain largely fixed. Pareto-optimal solutions reveal that competitive accuracy can often be achieved with significantly reduced computational cost compared to FLOPs-agnostic baselines. Our results establish FLOPs-awareness as a practical criterion for HQNN design in the NISQ era and as a scalable principle for future HQNN systems.
Authors: Brian Godwin Lim, Galvin Brice Lim, Renzo Roel Tan, Irwin King, Kazushi Ikeda
Abstract: Graph neural networks (GNNs) have become an indispensable tool for analyzing relational data. Classical GNNs are broadly classified into three variants: convolutional, attentional, and message-passing. While the standard message-passing variant is expressive, its typical pair-wise messages only consider the features of the center node and each neighboring node individually. This design fails to incorporate contextual information contained within the broader local neighborhood, potentially hindering its ability to learn complex relationships within the entire set of neighboring nodes. To address this limitation, this work first formalizes the concept of neighborhood-contextualization, rooted in a key property of the attentional variant. This then serves as the foundation for generalizing the message-passing variant to the proposed neighborhood-contextualized message-passing (NCMP) framework. To demonstrate its utility, a simple, practical, and efficient method to parametrize and operationalize NCMP is presented, leading to the development of the proposed Soft-Isomorphic Neighborhood-Contextualized Graph Convolution Network (SINC-GCN). Across a diverse set of synthetic and benchmark GNN datasets, SINC-GCN demonstrates competitive performance against baseline GNN models, highlighting its expressivity and efficiency. Notably, it also delivers substantial and statistically significant performance gains in graph property prediction tasks, further underscoring the distinctive utility of neighborhood-contextualization. Overall, the paper lays the foundation for the NCMP framework as a practical path toward enhancing the graph representational power of classical GNNs.
Authors: Zheyu Lin, Jirui Yang, Yukui Qiu, Hengqi Guo, Yubing Bao, Yao Guan
Abstract: Evaluating the safety robustness of LLMs is critical for their deployment. However, mainstream Red Teaming methods rely on online generation and black-box output analysis. These approaches are not only costly but also suffer from feedback latency, making them unsuitable for agile diagnostics after training a new model. To address this, we propose N-GLARE (A Non-Generative, Latent Representation-Efficient LLM Safety Evaluator). N-GLARE operates entirely on the model's latent representations, bypassing the need for full text generation. It characterizes hidden layer dynamics by analyzing the APT (Angular-Probabilistic Trajectory) of latent representations and introducing the JSS (Jensen-Shannon Separability) metric. Experiments on over 40 models and 20 red teaming strategies demonstrate that the JSS metric exhibits high consistency with the safety rankings derived from Red Teaming. N-GLARE reproduces the discriminative trends of large-scale red-teaming tests at less than 1\% of the token cost and the runtime cost, providing an efficient output-free evaluation proxy for real-time diagnostics.
Authors: Yansong Liu, Ronnie Stafford, Pramit Khetrapal, Huriye Kocadag, Gra\c{c}a Carvalho, Patricia de Winter, Maryam Imran, Amelia Snook, Adamos Hadjivasiliou, D. Vijay Anand, Weining Lin, John Kelly, Yukun Zhou, Ivana Drobnjak
Abstract: For patients undergoing systemic cancer therapy, the time between clinic visits is full of uncertainties and risks of unmonitored side effects. To bridge this gap in care, we developed and prospectively trialed a multi-modal AI framework for remote patient monitoring (RPM). This system integrates multi-modal data from the HALO-X platform, such as demographics, wearable sensors, daily surveys, and clinical events. Our observational trial is one of the largest of its kind and has collected over 2.1 million data points (6,080 patient-days) of monitoring from 84 patients. We developed and adapted a multi-modal AI model to handle the asynchronous and incomplete nature of real-world RPM data, forecasting a continuous risk of future adverse events. The model achieved an accuracy of 83.9% (AUROC=0.70). Notably, the model identified previous treatments, wellness check-ins, and daily maximum heart rate as key predictive features. A case study demonstrated the model's ability to provide early warnings by outputting escalating risk profiles prior to the event. This work establishes the feasibility of multi-modal AI RPM for cancer care and offers a path toward more proactive patient support.(Accepted at Europe NeurIPS 2025 Multimodal Representation Learning for Healthcare Workshop. Best Paper Poster Award.)
Authors: Yizhou Zhang
Abstract: Neural scaling laws and double-descent phenomena suggest that deep-network training obeys a simple macroscopic structure despite highly nonlinear optimization dynamics. We derive such structure directly from gradient descent in function space. For mean-squared error loss, the training error evolves as $\dot e_t=-M(t)e_t$ with $M(t)=J_{\theta(t)}J_{\theta(t)}^{\!*}$, a time-dependent self-adjoint operator induced by the network Jacobian. Using Kato perturbation theory, we obtain an exact system of coupled modewise ODEs in the instantaneous eigenbasis of $M(t)$. To extract macroscopic behavior, we introduce a logarithmic spectral-shell coarse-graining and track quadratic error energy across shells. Microscopic interactions within each shell cancel identically at the energy level, so shell energies evolve only through dissipation and external inter-shell interactions. We formalize this via a \emph{renormalizable shell-dynamics} assumption, under which cumulative microscopic effects reduce to a controlled net flux across shell boundaries. Assuming an effective power-law spectral transport in a relevant resolution range, the shell dynamics admits a self-similar solution with a moving resolution frontier and explicit scaling exponents. This framework explains neural scaling laws and double descent, and unifies lazy (NTK-like) training and feature learning as two limits of the same spectral-shell dynamics.
Authors: Antonio Roye-Azar, Santiago Vargas-Naranjo, Dhruv Ghai, Nithin Balamurugan, Rayan Amir
Abstract: Tiny Recursive Models (TRM) were proposed as a parameter-efficient alternative to large language models for solving Abstraction and Reasoning Corpus (ARC) style tasks. The original work reports strong performance and suggests that recursive latent updates enable non-trivial reasoning, but it remains unclear how much of this performance stems from architecture, test-time compute, or task-specific priors. In this technical note, we empirically analyze the ARC Prize TRM checkpoint on ARC-AGI-1 and report four behavioral findings and an efficiency comparison. First, we show that test-time augmentation and majority-vote ensembling account for a substantial fraction of reported performance: the 1000-sample voting pipeline improves Pass@1 by about 11 percentage points over single-pass canonical inference. Second, a puzzle-identity ablation reveals strict dependence on task identifiers: replacing the correct puzzle ID with a blank or random token yields zero accuracy. Third, a recursion trajectory analysis shows that most of the final accuracy is achieved at the first recursion step and that performance saturates after few latent updates, indicating shallow effective recursion. Fourth, early-stage training experiments under canonical versus heavy augmentation regimes suggest that heavy augmentation broadens the distribution of candidate solutions and improves multi-sample success. Finally, we compare TRM with a naive QLoRA fine-tune of Llama 3 8B on canonical ARC-AGI-1, finding that TRM's non-autoregressive design achieves much higher throughput and substantially lower memory usage in this setting. Overall, TRM's ARC-AGI-1 performance appears to arise from an interaction between efficiency, task-specific conditioning, and aggressive test-time compute rather than deep internal reasoning.
Authors: Jinhao Zhang, Yunquan Zhang, Daning Chen, JunSun, Zicheng Yan
Abstract: Current mainstream post-training quantization methods for large language models typically apply a uniform quantization strategy across all network layers, overlooking the substantial differences in algorithmic suitability among layers. To address this limitation, we propose CALM (A CKA-guided Adaptive Layer-wise Modularization)a fine-tuning-free, plug-and-play framework for algorithmic heterogeneous quantization. CALM independently evaluates multiple PTQ algorithms on each layer and employs Linear Centered Kernel Alignment (CKA) as a metric to automatically select the optimal quantization strategy per layer. The individually optimized strategies are then integrated to construct a hybrid quantized model. Experiments demonstrate that our approach consistently outperforms both uniform quantization baselines and state-of-the-art mixed-precision methods across mainstream LLMsincluding LLaMA and Qwenin terms of perplexity (PPL) and downstream task performance.
Authors: Yuma Ichikawa, Naoya Takagi, Takumi Nakagawa, Yuzi Kanazawa, Akira Sakai
Abstract: Transformers operate as horizontal token-by-token scanners; at each generation step, attending to an ever-growing sequence of token-level states. This access pattern increases prefill latency and makes long-context decoding more memory-bound, as KV-cache reads and writes dominate inference time over arithmetic operations. We propose Parallel Hierarchical Operation for TOp-down Networks (PHOTON), a hierarchical autoregressive model that replaces horizontal scanning with vertical, multi-resolution context scanning. PHOTON maintains a hierarchy of latent streams: a bottom-up encoder compresses tokens into low-rate contextual states, while lightweight top-down decoders reconstruct fine-grained token representations in parallel. We further introduce recursive generation that updates only the coarsest latent stream and eliminates bottom-up re-encoding. Experimental results show that PHOTON is superior to competitive Transformer-based language models regarding the throughput-quality trade-off, providing advantages in long-context and multi-query tasks. In particular, this reduces decode-time KV-cache traffic, yielding up to $10^{3}\times$ higher throughput per unit memory.
Authors: Giacomo Turri, Gr\'egoire Pacreau, Giacomo Meanti, Timoth\'ee Devergne, Daniel Ordonez, Erfan Mirzaei, Bruno Belucci, Karim Lounici, Vladimir Kostic, Massimiliano Pontil, Pietro Novelli
Abstract: kooplearn is a machine-learning library that implements linear, kernel, and deep-learning estimators of dynamical operators and their spectral decompositions. kooplearn can model both discrete-time evolution operators (Koopman/Transfer) and continuous-time infinitesimal generators. By learning these operators, users can analyze dynamical systems via spectral methods, derive data-driven reduced-order models, and forecast future states and observables. kooplearn's interface is compliant with the scikit-learn API, facilitating its integration into existing machine learning and data science workflows. Additionally, kooplearn includes curated benchmark datasets to support experimentation, reproducibility, and the fair comparison of learning algorithms. The software is available at https://github.com/Machine-Learning-Dynamical-Systems/kooplearn.
URLs: https://github.com/Machine-Learning-Dynamical-Systems/kooplearn.
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: Yilun Luo, Huaqing Zheng, Haoqian Meng, Wenyuan Liu, Peng Zhang
Abstract: Huawei's openPangu-Embedded-1B and openPangu-Embedded-7B are variants of the openPangu large language model, designed for efficient deployment on Ascend NPUs. The 7B variant supports three distinct Chain-of-Thought (CoT) reasoning paradigms, namely slow_think, auto_think, and no_think, while the 1B variant operates exclusively in the no_think mode, which employs condensed reasoning for higher efficiency. Although CoT reasoning enhances capability, the generation of extended reasoning traces introduces substantial memory and latency overheads, posing challenges for practical deployment on Ascend NPUs. This paper addresses these computational constraints by leveraging low-bit quantization, which transforms FP16 computations into more efficient integer arithmetic. We introduce a unified low-bit inference framework, supporting INT8 (W8A8) and W4A8 quantization, specifically optimized for openPangu-Embedded models on the Atlas A2. Our comprehensive evaluation on code generation benchmarks (HumanEval and MBPP) demonstrates the efficacy of this approach. INT8 quantization consistently preserves over 90\% of the FP16 baseline accuracy and achieves a 1.5x prefill speedup on the Atlas A2. Furthermore, W4A8 quantization significantly reduces memory consumption, albeit with a moderate trade-off in accuracy. These findings collectively indicate that low-bit quantization effectively facilitates efficient CoT reasoning on Ascend NPUs, maintaining high model fidelity.
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: Nicolas Caron, Christophe Guyeux, Hassan Noura, Benjamin Aynes
Abstract: Wildfires are highly imbalanced natural hazards in both space and severity, making the prediction of extreme events particularly challenging. In this work, we introduce the first ordinal classification framework for forecasting wildfire severity levels directly aligned with operational decision-making in France. Our study investigates the influence of loss-function design on the ability of neural models to predict rare yet critical high-severity fire occurrences. We compare standard cross-entropy with several ordinal-aware objectives, including the proposed probabilistic TDeGPD loss derived from a truncated discrete exponentiated Generalized Pareto Distribution. Through extensive benchmarking over multiple architectures and real operational data, we show that ordinal supervision substantially improves model performance over conventional approaches. In particular, the Weighted Kappa Loss (WKLoss) achieves the best overall results, with more than +0.1 IoU (Intersection Over Union) gain on the most extreme severity classes while maintaining competitive calibration quality. However, performance remains limited for the rarest events due to their extremely low representation in the dataset. These findings highlight the importance of integrating both severity ordering, data imbalance considerations, and seasonality risk into wildfire forecasting systems. Future work will focus on incorporating seasonal dynamics and uncertainty information into training to further improve the reliability of extreme-event prediction.
Authors: Hao Tang, Hao Chen, Hao Li, Chao Li
Abstract: Spherical deep learning has been widely applied to a broad range of real-world problems. Existing approaches often face challenges in balancing strong spherical geometric inductive biases with the need to model real-world heterogeneity. To solve this while retaining spherical geometry, we first introduce a designable Green's function framework (DGF) to provide new spherical operator solution strategy: Design systematic Green's functions under rotational group. Based on DGF, to model biological heterogeneity, we propose Green's-Function Spherical Neural Operator (GSNO) fusing 3 operator solutions: (1) Equivariant Solution derived from Equivariant Green's Function for symmetry-consistent modeling; (2) Invariant Solution derived from Invariant Green's Function to eliminate nuisance heterogeneity, e.g., consistent background field; (3) Anisotropic Solution derived from Anisotropic Green's Function to model anisotropic systems, especially fibers with preferred direction. Therefore, the resulting model, GSNO can adapt to real-world heterogeneous systems with nuisance variability and anisotropy while retaining spectral efficiency. Evaluations on spherical MNIST, Shallow Water Equation, diffusion MRI fiber prediction, cortical parcellation and molecule structure modeling demonstrate the superiority of GSNO.
Authors: Zhengyi Kwan, Wei Zhang, Aik Beng Ng, Zhengkui Wang, Simon See
Abstract: Job scheduling is widely used in real-world manufacturing systems to assign ordered job operations to machines under various constraints. Existing solutions remain limited by long running time or insufficient schedule quality, especially when problem scale increases. In this paper, we propose ReLA, a reinforcement-learning (RL) scheduler built on structured representation learning and aggregation. ReLA first learns diverse representations from scheduling entities, including job operations and machines, using two intra-entity learning modules with self-attention and convolution and one inter-entity learning module with cross-attention. These modules are applied in a multi-scale architecture, and their outputs are aggregated to support RL decision-making. Across experiments on small, medium, and large job instances, ReLA achieves the best makespan in most tested settings over the latest solutions. On non-large instances, ReLA reduces the optimality gap of the SOTA baseline by 13.0%, while on large-scale instances it reduces the gap by 78.6%, with the average optimality gaps lowered to 7.3% and 2.1%, respectively. These results confirm that ReLA's learned representations and aggregation provide strong decision support for RL scheduling, and enable fast job completion and decision-making for real-world applications.
Authors: Gil Shabat
Abstract: Low-rank approximations of large kernel matrices are ubiquitous in machine learning, particularly for scaling Gaussian Processes to massive datasets. The Pivoted Cholesky decomposition is a standard tool for this task, offering a computationally efficient, greedy low-rank approximation. While its algebraic properties are well-documented in numerical linear algebra, its geometric intuition within the context of kernel methods often remains obscure. In this note, we elucidate the geometric interpretation of the algorithm within the Reproducing Kernel Hilbert Space (RKHS). We demonstrate that the pivotal selection step is mathematically equivalent to Farthest Point Sampling (FPS) using the kernel metric, and that the Cholesky factor construction is an implicit Gram-Schmidt orthogonalization. We provide a concise derivation and a minimalist Python implementation to bridge the gap between theory and practice.
Authors: Akash Kumar
Abstract: We study when geometric simplicity of decision boundaries, used here as a notion of interpretability, can conflict with accurate approximation of axis-aligned decision trees by shallow neural networks. Decision trees induce rule-based, axis-aligned decision regions (finite unions of boxes), whereas shallow ReLU networks are typically trained as score models whose predictions are obtained by thresholding. We analyze the infinite-width, bounded-norm, single-hidden-layer ReLU class through the Radon total variation ($\mathrm{R}\mathrm{TV}$) seminorm, which controls the geometric complexity of level sets. We first show that the hard tree indicator $1_A$ has infinite $\mathrm{R}\mathrm{TV}$. Moreover, two natural split-wise continuous surrogates--piecewise-linear ramp smoothing and sigmoidal (logistic) smoothing--also have infinite $\mathrm{R}\mathrm{TV}$ in dimensions $d>1$, while Gaussian convolution yields finite $\mathrm{R}\mathrm{TV}$ but with an explicit exponential dependence on $d$. We then separate two goals that are often conflated: classification after thresholding (recovering the decision set) versus score learning (learning a calibrated score close to $1_A$). For classification, we construct a smooth barrier score $S_A$ with finite $\mathrm{R}\mathrm{TV}$ whose fixed threshold $\tau=1$ exactly recovers the box. Under a mild tube-mass condition near $\partial A$, we prove an $L_1(P)$ calibration bound that decays polynomially in a sharpness parameter, along with an explicit $\mathrm{R}\mathrm{TV}$ upper bound in terms of face measures. Experiments on synthetic unions of rectangles illustrate the resulting accuracy--complexity tradeoff and how threshold selection shifts where training lands along it.
Authors: Eghbal Rahimikia, Stefan Zohren, Ser-Huang Poon
Abstract: We examine whether news can improve realised volatility forecasting using a modern yet operationally simple NLP framework. News text is transformed into embedding-based representations, and forecasts are evaluated both as a standalone, news-only model and as a complement to standard realised volatility benchmarks. In out-of-sample tests on a cross-section of stocks, news contains useful predictive information, with stronger effects for stock-related content and during high volatility days. Combining the news-based signal with a leading benchmark yields consistent improvements in statistical performance and economically meaningful gains, while explainability analysis highlights the news themes most relevant for volatility.
Authors: Juyoung Yun, Jungmin Shin
Abstract: Solar flares, especially C, M, and X class, pose significant risks to satellite operations, communication systems, and power grids. We present a novel approach for predicting extreme solar flares using HMI intensitygrams and magnetograms. By detecting sunspots from intensitygrams and extracting magnetic field patches from magnetograms, we train a Residual Network (ResNet) to classify extreme class flares. Our model demonstrates high accuracy, offering a robust tool for predicting extreme solar flares and improving space weather forecasting. Additionally, we show that HMI magnetograms provide more useful data for deep learning compared to other SDO AIA images by better capturing features critical for predicting flare magnitudes. This study underscores the importance of identifying magnetic fields in solar flare prediction, marking a significant advancement in solar activity prediction with practical implications for mitigating space weather impacts.
Authors: Freya Shah, Taylor L. Patti, Julius Berner, Bahareh Tolooshams, Jean Kossaifi, Anima Anandkumar
Abstract: Fourier Neural Operators (FNOs) excel on tasks using functional data, such as those originating from partial differential equations. Such characteristics render them an effective approach for simulating the time evolution of quantum wavefunctions, which is a computationally challenging, yet coveted task for studying quantum systems. In this manuscript, we use FNOs to model the evolution of quantum spin systems, so chosen due to their representative quantum dynamics. We explore two distinct FNO architectures, examining their performance for learning and predicting time evolution on both random and low-energy input states. We find that standard neural networks in fixed dimensions, such as U-Net, exhibit limited ability to extrapolate beyond the training time interval, whereas FNOs reliably capture the underlying time-evolution operator, generalizing effectively to unseen times. Additionally, we apply FNOs to a compact set of Hamiltonian observables ($\sim\text{poly}(n)$) instead of the entire $2^n$ quantum wavefunction, which greatly reduces the size of our FNO inputs, outputs and model dimensions. Moreover, this Hamiltonian observable-based method demonstrates that FNOs can effectively distill information from high-dimensional spaces into lower-dimensional spaces. Using this approach, we perform numerical experiments on a 20-qubit system and extrapolate Hamiltonian observables to twice the training time with a relative error of $5.8\%$. Relative to numerical time-evolution methods, FNO achieves an inference speedup of approximately $10^{4}\times$ for 20-qubit systems. The extrapolation of Hamiltonian observables to times later than those used in training is of particular interest, as this stands to fundamentally increase the simulatability of quantum systems past both the coherence times of contemporary quantum architectures and the circuit-depths of tractable tensor networks.
Authors: Yifan Zhang, Yang Yuan, Andrew Chi-Chih Yao
Abstract: Large Language Models (LLMs) excel at many tasks but often falter on complex problems that require structured, multi-step reasoning. We introduce the Diagram of Thought (DoT), a new framework that enables a single LLM to build and navigate a mental map of its reasoning. Instead of thinking in a straight line, the model constructs a dynamic diagram of ideas, where it can propose different lines of thought, critique its own steps, and synthesize validated insights into a final conclusion. This entire process is self-contained within the model, making it highly efficient by avoiding the complex external controllers or search algorithms required by other methods. To ensure the reliability of this process, we ground DoT in a rigorous mathematical framework from category theory. This foundation guarantees that the way the model combines information is logical, consistent, and robust, regardless of the order in which ideas were explored. The result is a more powerful and transparent reasoning process that produces a fully auditable, step-by-step trace of the LLM's thinking, bridging the gap between fluent language and formal reasoning.
Authors: K. R. Schuurman, A. Meyer
Abstract: Accurate estimates of surface solar irradiance (SSI) are essential for solar resource assessments and solar energy forecasts in grid integration and building control applications. SSI estimates for spatially extended regions can be retrieved from geostationary satellites such as Meteosat. Traditional SSI satellite retrievals like Heliosat rely on physical radiative transfer modelling. We introduce the first machine-learning-based satellite retrieval for instantaneous SSI and demonstrate its capability to provide accurate and generalizable SSI estimates across Europe. Our deep learning retrieval provides near real-time SSI estimates based on data-driven emulation of Heliosat and fine-tuning on pyranometer networks. By including SSI from ground stations, our SSI retrieval model can outperform Heliosat accuracy and generalize well to regions with other climates and surface albedos in cloudy conditions (clear-sky index < 0.8). We also show that the SSI retrieved from Heliosat exhibits large biases in mountain regions, and that training and fine-tuning our retrieval models on SSI data from ground stations strongly reduces these biases, outperforming Heliosat. Furthermore, we quantify the relative importance of the Meteosat channels and other predictor variables like solar zenith angle for the accuracy of our deep learning SSI retrieval model in different cloud conditions. We find that in cloudy conditions multiple near-infrared and infrared channels enhance the performance. Our results can facilitate the development of more accurate satellite retrieval models of surface solar irradiance.
Authors: Emanuele Iannone, Quang-Cuong Bui, Riccardo Scandariato
Abstract: Software vulnerabilities are often detected via taint analysis, penetration testing, or fuzzing. They are also found via unit tests that exercise security-sensitive behavior with specific inputs, called vulnerability-witnessing tests. Generative AI models could help developers in writing them, but they require many examples to learn from, which are currently scarce. This paper introduces VuTeCo, an AI-driven framework for collecting examples of vulnerability-witnessing tests from Java repositories. VuTeCo carries out two tasks: (1) The "Finding" task to determine whether a unit test case is security-related, and (2) the "Matching" task to relate a test case to the vulnerability it witnesses. VuTeCo addresses the Finding task with UniXcoder, achieving an F0.5 score of 0.73 and a precision of 0.83 on a test set of unit tests from Vul4J. The Matching task is addressed using DeepSeek Coder, achieving an F0.5 score of 0.65 and a precision of 0.75 on a test set of pairs of unit tests and vulnerabilities from Vul4J. VuTeCo has been used in the wild on 427 Java projects and 1,238 vulnerabilities, obtaining 224 test cases confirmed to be security-related and 35 tests correctly matched to 29 vulnerabilities. The validated tests were collected in a new dataset called Test4Vul. VuTeCo lays the foundation for large-scale retrieval of vulnerability-witnessing tests, enabling future AI models to better understand and generate security unit tests.
Authors: Vladimir Vargas-Calder\'on, Santiago Acevedo-Mancera, Herbert Vinck-Posada
Abstract: Decision diagrams (DDs) have emerged as an efficient tool for simulating quantum circuits due to their capacity to exploit data redundancies in quantum states and quantum operations, enabling the efficient computation of probability amplitudes. However, their application in quantum machine learning (QML) has remained unexplored. This paper introduces variational decision diagrams (VDDs), a novel graph structure that combines the structural benefits of DDs with the adaptability of variational methods for efficiently representing quantum states. We investigate the trainability of VDDs by applying them to the ground state estimation problem for transverse-field Ising and Heisenberg Hamiltonians. Analysis of gradient variance suggests that training VDDs is possible, as no signs of vanishing gradients--also known as barren plateaus--are observed. This work provides new insights into the use of decision diagrams in QML as an alternative to design and train variational ans\"atze.
Authors: Andr\'es Herrera-Poyatos, Javier Del Ser, Marcos L\'opez de Prado, Fei-Yue Wang, Enrique Herrera-Viedma, Francisco Herrera
Abstract: Responsible Artificial Intelligence (RAI) addresses the ethical and regulatory challenges of deploying AI systems in high-risk scenarios. This paper proposes a comprehensive framework for the design of an RAI system (RAIS) that integrates five key dimensions: domain definition, trustworthy AI design, auditability, accountability, and governance. Unlike prior work that treats these components in isolation, our proposal emphasizes their inter-dependencies and iterative feedback loops, enabling proactive and reactive accountability throughout the AI lifecycle. Beyond presenting the framework, we synthesize recent developments in global AI governance and analyze limitations in existing principles-based approaches, highlighting fragmentation, implementation gaps, and the need for participatory governance. The paper also identifies critical challenges and research directions for the RAIS framework, including sector-specific adaptation and operationalization, to support certification, post-deployment monitoring, and risk-based auditing. By bridging technical design and institutional responsibility, this work offers a practical blueprint for embedding responsibility throughout the AI lifecycle, enabling transparent, ethically aligned, and legally compliant AI-based systems.
Authors: Amjad Ali, Saeed Aldahmani, Hailiang Du, Zardad Khan
Abstract: This paper introduces the centroid decision forest (CDF), a novel ensemble learning framework that redefines the splitting strategy and tree building in the ordinary decision trees for high-dimensional classification. The splitting approach in CDF differs from the traditional decision trees in theat the class separability score (CSS) determines the selection of the most discriminative features at each node to construct centroids of the partitions (daughter nodes). The splitting criterion uses the Euclidean distance measurements from each class centroid to achieve a splitting mechanism that is more flexible and robust. Centroids are constructed by computing the mean feature values of the selected features for each class, ensuring a class-representative division of the feature space. This centroid-driven approach enables CDF to capture complex class structures while maintaining interpretability and scalability. To evaluate CDF, 23 high-dimensional datasets are used to assess its performance against different state-of-the-art classifiers through classification accuracy and Cohen's kappa statistic. The experimental results show that CDF outperforms the conventional methods establishing its effectiveness and flexibility for high-dimensional classification problems.
Authors: Baolei Zhang, Yuxi Chen, Zhuqing Liu, Lihai Nie, Tong Li, Zheli Liu, Minghong Fang
Abstract: Large language models (LLMs) have demonstrated impressive natural language processing abilities but face challenges such as hallucination and outdated knowledge. Retrieval-Augmented Generation (RAG) has emerged as a state-of-the-art approach to mitigate these issues. While RAG enhances LLM outputs, it remains vulnerable to poisoning attacks. Recent studies show that injecting poisoned text into the knowledge database can compromise RAG systems, but most existing attacks assume that the attacker can insert a sufficient number of poisoned texts per query to outnumber correct-answer texts in retrieval, an assumption that is often unrealistic. To address this limitation, we propose CorruptRAG, a practical poisoning attack against RAG systems in which the attacker injects only a single poisoned text, enhancing both feasibility and stealth. Extensive experiments conducted on multiple large-scale datasets demonstrate that CorruptRAG achieves higher attack success rates than existing baselines.
Authors: Chenhao Li, Andreas Krause, Marco Hutter
Abstract: Reinforcement Learning (RL) has achieved impressive results in robotics, yet high-performing pipelines remain highly task-specific, with little reuse of prior data. Offline Model-based RL (MBRL) offers greater data efficiency by training policies entirely from existing datasets, but suffers from compounding errors and distribution shift in long-horizon rollouts. Although existing methods have shown success in controlled simulation benchmarks, robustly applying them to the noisy, biased, and partially observed datasets typical of real-world robotics remains challenging. We present a principled pipeline for making offline MBRL effective on physical robots. Our RWM-U extends autoregressive world models with epistemic uncertainty estimation, enabling temporally consistent multi-step rollouts with uncertainty effectively propagated over long horizons. We combine RWM-U with MOPO-PPO, which adapts uncertainty-penalized policy optimization to the stable, on-policy PPO framework for real-world control. We evaluate our approach on diverse manipulation and locomotion tasks in simulation and on real quadruped and humanoid, training policies entirely from offline datasets. The resulting policies consistently outperform model-free and uncertainty-unaware model-based baselines, and fusing real-world data in model learning further yields robust policies that surpass online model-free baselines trained solely in simulation.
Authors: Jiaqi Leng, Zhiyan Ding, Zherui Chen, Lin Lin
Abstract: Sampling from probability distributions of the form $\sigma \propto e^{-\beta V}$, where $V$ is a continuous potential, is a fundamental task across physics, chemistry, biology, computer science, and statistics. However, when $V$ is non-convex, the resulting distribution becomes non-logconcave, and classical methods such as Langevin dynamics often exhibit poor performance. We introduce the first quantum algorithm that provably accelerates a broad class of continuous-time sampling dynamics. For Langevin dynamics, our method encodes the target Gibbs measure into the amplitudes of a quantum state, identified as the kernel of a block matrix derived from a factorization of the Witten Laplacian operator. This connection enables Gibbs sampling via singular value thresholding and yields up to a quartic quantum speedup over best-known classical Langevin-based methods in the non-logconcave setting. Building on this framework, we further develop the first quantum algorithm that accelerates replica exchange Langevin diffusion, a widely used method for sampling from complex, rugged energy landscapes.
Authors: Jamie Hayes, Ilia Shumailov, Christopher A. Choquette-Choo, Matthew Jagielski, George Kaissis, Milad Nasr, Sahra Ghalebikesabi, Meenatchi Sundaram Mutu Selva Annamalai, Niloofar Mireshghallah, Igor Shilov, Matthieu Meeus, Yves-Alexandre de Montjoye, Katherine Lee, Franziska Boenisch, Adam Dziedzic, A. Feder Cooper
Abstract: State-of-the-art membership inference attacks (MIAs) typically require training many reference models, making it difficult to scale these attacks to large pre-trained language models (LLMs). As a result, prior research has either relied on weaker attacks that avoid training references (e.g., fine-tuning attacks), or on stronger attacks applied to small models and datasets. However, weaker attacks have been shown to be brittle and insights from strong attacks in simplified settings do not translate to today's LLMs. These challenges prompt an important question: are the limitations observed in prior work due to attack design choices, or are MIAs fundamentally ineffective on LLMs? We address this question by scaling LiRA--one of the strongest MIAs--to GPT-2 architectures ranging from 10M to 1B parameters, training references on over 20B tokens from the C4 dataset. Our results advance the understanding of MIAs on LLMs in four key ways. While (1) strong MIAs can succeed on pre-trained LLMs, (2) their effectiveness, remains limited (e.g., AUC<0.7) in practical settings. (3) Even when strong MIAs achieve better-than-random AUC, aggregate metrics can conceal substantial per-sample MIA decision instability: due to training randomness, many decisions are so unstable that they are statistically indistinguishable from a coin flip. Finally, (4) the relationship between MIA success and related LLM privacy metrics is not as straightforward as prior work has suggested.
Authors: Feng Luo, Yu-Neng Chuang, Guanchu Wang, Hoang Anh Duy Le, Shaochen Zhong, Hongyi Liu, Jiayi Yuan, Yang Sui, Vladimir Braverman, Vipin Chaudhary, Xia Hu
Abstract: Reasoning-capable large language models (LLMs) achieve strong performance on complex tasks but often exhibit overthinking after distillation, generating unnecessarily long chain-of-thought (CoT) reasoning even for simple inputs and incurring high inference cost. However, naively shortening reasoning length can degrade reasoning accuracy, as concise reasoning may be insufficient for certain inputs and lacks explicit supervision. We propose Auto Long-Short Reasoning (AutoL2S), a distillation framework that empowers non-reasoning LLMs to think thoroughly but only when necessary. AutoL2S first learns a lightweight switching token with verified long-short CoTs to enable instance-wise long-short reasoning selection. Then it leverages long-short reasoning rollouts induced by a switching token in a GRPO-style loss to improve reasoning efficiency while maintaining accuracy. Experiments demonstrate that AutoL2S effectively reduces reasoning length up to 71% with minimal accuracy loss, yielding markedly better trade-off in token length and inference time while preserving accuracy.
Authors: Zhuohan Xie, Daniil Orel, Rushil Thareja, Dhruv Sahnan, Hachem Madmoun, Fan Zhang, Debopriyo Banerjee, Georgi Georgiev, Xueqing Peng, Lingfei Qian, Jimin Huang, Jinyan Su, Aaryamonvikram Singh, Rui Xing, Rania Elbadry, Chen Xu, Haonan Li, Fajri Koto, Ivan Koychev, Tanmoy Chakraborty, Yuxia Wang, Salem Lahlou, Veselin Stoyanov, Sophia Ananiadou, Preslav Nakov
Abstract: Multi-step symbolic reasoning is essential for robust financial analysis; yet, current benchmarks largely overlook this capability. Existing datasets such as FinQA and ConvFinQA emphasize final numerical answers while neglecting the intermediate reasoning required for transparency and verification. To address this gap, we introduce FINCHAIN, the first benchmark specifically designed for verifiable Chain-of-Thought (CoT) evaluation in finance. FINCHAIN spans 58 topics across 12 financial domains, each represented by parameterized symbolic templates with executable Python traces that enable fully machine-verifiable reasoning and scalable, contamination-free data generation. To assess reasoning capacity, we propose CHAINEVAL, a dynamic alignment measure that jointly evaluates both the final-answer correctness and the step-level reasoning consistency. Our evaluation of 26 leading LLMs reveals that even frontier proprietary LLMs exhibit clear limitations in symbolic financial reasoning, while domain-adapted and math-enhanced fine-tuned models can substantially narrow this gap. Overall, FINCHAIN exposes persistent weaknesses in multi-step financial reasoning and provides a foundation for developing trustworthy, interpretable, and verifiable financial AI.
Authors: Hongyu Yao, Zijin Hong, Hao Chen, Zhiqing Li, Qijie Shen, Zuobin Ying, Qihua Feng, Huan Gong, Feiran Huang
Abstract: Graph-based multi-task learning at billion-scale presents a significant challenge, as different tasks correspond to distinct billion-scale graphs. Traditional multi-task learning methods often neglect these graph structures, relying solely on individual user and item embeddings. However, disregarding graph structures overlooks substantial potential for improving performance. In this paper, we introduce the Macro Graph of Experts (MGOE) framework, the first approach capable of leveraging macro graph embeddings to capture task-specific macro features while modeling the correlations between task-specific experts. Specifically, we propose the concept of a Macro Graph Bottom, which, for the first time, enables multi-task learning models to incorporate graph information effectively. We design the Macro Prediction Tower to dynamically integrate macro knowledge across tasks. MGOE has been deployed at scale, powering multi-task learning for a leading billion-scale recommender system, Alibaba. Extensive offline experiments conducted on three public benchmark datasets demonstrate its superiority over state-of-the-art multi-task learning methods, establishing MGOE as a breakthrough in multi-task graph-based recommendation. Furthermore, online A/B tests confirm the superiority of MGOE in billion-scale recommender systems.
Authors: Xiang Cheng, Chengyan Pan, Minjun Zhao, Deyang Li, Fangchao Liu, Xinyu Zhang, Xiao Zhang, Yong Liu
Abstract: In-Context Learning (ICL) is an essential emergent ability of Large Language Models (LLMs), and recent studies introduce Chain-of-Thought (CoT) to exemplars of ICL to enhance the reasoning capability, especially in mathematics tasks. However, given the continuous advancement of model capabilities, it remains unclear whether CoT exemplars still benefit recent, stronger models in such tasks. Through systematic experiments, we find that for recent strong models such as the Qwen2.5 series, adding traditional CoT exemplars does not improve reasoning performance compared to Zero-Shot CoT. Instead, their primary function is to align the output format with human expectations. We further investigate the effectiveness of enhanced CoT exemplars, constructed using answers from advanced models such as \texttt{Qwen2.5-Max} and \texttt{DeepSeek-R1}. Experimental results indicate that these enhanced exemplars still fail to improve the model's reasoning performance. Further analysis reveals that models tend to ignore the exemplars and focus primarily on the instructions, leading to no observable gain in reasoning ability. Overall, our findings highlight the limitations of the current ICL+CoT framework in mathematical reasoning, calling for a re-examination of the ICL paradigm and the definition of exemplars.
Authors: Sebastian Nagies, Emiliano Tolotti, Davide Pastorello, Enrico Blanzieri
Abstract: Quantum neural networks (QNNs) based on parametrized quantum circuits are promising candidates for machine learning applications, yet many architectures lack clear connections to classical models, potentially limiting their ability to leverage established classical neural network techniques. We examine QNNs built from SWAP test circuits and discuss their equivalence to classical two-layer feedforward networks with quadratic activations under amplitude encoding. Evaluation on real-world and synthetic datasets shows that while this architecture learns many practical binary classification tasks, it has fundamental expressivity limitations: polynomial activation functions do not satisfy the universal approximation theorem, and we show analytically that the architecture cannot learn the parity check function beyond two dimensions, regardless of network size. To address this, we introduce generalized SWAP test circuits with multiple Fredkin gates sharing an ancilla, implementing product layers with polynomial activations of arbitrary even degree. This modification enables successful learning of parity check functions in arbitrary dimensions as well as binary n-spiral tasks, and we provide numerical evidence that the expressivity enhancement extends to alternative encoding schemes such as angle (Z) and ZZ feature maps. We validate the practical feasibility of our proposed architecture by implementing a classically pretrained instance on the IBM Torino quantum processor, achieving 84% classification accuracy on the three-dimensional parity check despite hardware noise. Our work establishes a framework for analyzing and enhancing QNN expressivity through correspondence with classical architectures, and demonstrates that SWAP test-based QNNs possess broad representational capacity relevant to both classical and potentially quantum learning tasks.
Authors: Callum Duffy, Marcin Jastrzebski
Abstract: In this work, we investigate the phenomenon of spectral bias in quantum machine learning, where, in classical settings, models tend to fit low-frequency components of a target function earlier during training than high-frequency ones, demonstrating a frequency-dependent rate of convergence. We study this effect specifically in parameterised quantum circuits (PQCs). Leveraging the established formulation of PQCs as Fourier series, we prove that spectral bias in this setting arises from the ``redundancy'' of the Fourier coefficients, which denotes the number of terms in the analytical form of the model contributing to the same frequency component. The choice of data encoding scheme dictates the degree of redundancy for a Fourier coefficient. We find that the magnitude of the Fourier coefficients' gradients during training strongly correlates with the coefficients' redundancy. We then further demonstrate this empirically with three different encoding schemes. Additionally, we demonstrate that PQCs with greater redundancy exhibit increased robustness to random perturbations in their parameters at the corresponding frequencies. We investigate how design choices affect the ability of PQCs to learn Fourier sums, focusing on parameter initialization scale and entanglement structure, finding large initializations and low-entanglement schemes tend to slow convergence.
Authors: Anqi Li, Wenwei Jin, Jintao Tong, Pengda Qin, Weijia Li, Guo Lu
Abstract: Social platforms have revolutionized information sharing, but also accelerated the dissemination of harmful and policy-violating content. To ensure safety and compliance at scale, moderation systems must go beyond efficiency and offer accuracy and interpretability. However, current approaches largely rely on noisy, label-driven learning, lacking alignment with moderation rules and producing opaque decisions that hinder human review. Therefore, we propose Hierarchical Guard (Hi-Guard), a multimodal moderation framework that introduces a new policy-aligned decision paradigm. The term "Hierarchical" reflects two key aspects of our system design: (1) a hierarchical moderation pipeline, where a lightweight binary model first filters safe content and a stronger model handles fine-grained risk classification; and (2) a hierarchical taxonomy in the second stage, where the model performs path-based classification over a hierarchical taxonomy ranging from coarse to fine-grained levels. To ensure alignment with evolving moderation policies, Hi-Guard directly incorporates rule definitions into the model prompt. To further enhance structured prediction and reasoning, we introduce a multi-level soft-margin reward and optimize with Group Relative Policy Optimization (GRPO), penalizing semantically adjacent misclassifications and improving explanation quality. Extensive experiments and real-world deployment demonstrate that Hi-Guard achieves superior classification accuracy, generalization, and interpretability, paving the way toward scalable, transparent, and trustworthy content safety systems. Code is available at: https://github.com/lianqi1008/Hi-Guard.
Authors: Chengyandan Shen, Christoffer Sloth
Abstract: This paper proposes an exploration-efficient Deep Reinforcement Learning with Reference policy (DRLR) framework for learning robotics tasks that incorporates demonstrations. The DRLR framework is developed based on an algorithm called Imitation Bootstrapped Reinforcement Learning (IBRL). We propose to improve IBRL by modifying the action selection module. The proposed action selection module provides a calibrated Q-value, which mitigates the bootstrapping error that otherwise leads to inefficient exploration. Furthermore, to prevent the RL policy from converging to a sub-optimal policy, SAC is used as the RL policy instead of TD3. The effectiveness of our method in mitigating bootstrapping error and preventing overfitting is empirically validated by learning two robotics tasks: bucket loading and open drawer, which require extensive interactions with the environment. Simulation results also demonstrate the robustness of the DRLR framework across tasks with both low and high state-action dimensions, and varying demonstration qualities. To evaluate the developed framework on a real-world industrial robotics task, the bucket loading task is deployed on a real wheel loader. The sim2real results validate the successful deployment of the DRLR framework.
Authors: Avinash Patil
Abstract: Efficient bug resolution is critical for maintaining software quality and user satisfaction. However, specific bug reports experience unusually long resolution times, which may indicate underlying process inefficiencies or complex issues. This study presents a comprehensive analysis of bug resolution anomalies across seven prominent open-source repositories: Cassandra, Firefox, Hadoop, HBase, SeaMonkey, Spark, and Thunderbird. Utilizing statistical methods such as Z-score and Interquartile Range (IQR), we identify anomalies in bug resolution durations. To understand the thematic nature of these anomalies, we apply Term Frequency-Inverse Document Frequency (TF-IDF) for textual feature extraction and KMeans clustering to group similar bug summaries. Our findings reveal consistent patterns across projects, with anomalies often clustering around test failures, enhancement requests, and user interface issues. This approach provides actionable insights for project maintainers to prioritize and effectively address long-standing bugs.
Authors: Bosong Zhang, Timothy M. Merlis
Abstract: Machine learning models for the global atmosphere that are capable of producing stable, multi-year simulations of Earth's climate have recently been developed. However, the ability of these ML models to generalize beyond the training distribution remains an open question. In this study, we evaluate the climate response of several state-of-the-art ML models (ACE2-ERA5, NeuralGCM, and cBottle) to a uniform sea surface temperature warming, a widely used benchmark for evaluating climate change. We assess each ML model's performance relative to a physics-based general circulation model (NOAA's Geophysical Fluid Dynamics Laboratory AM4) across key diagnostics, including surface air temperature, precipitation, temperature and wind profiles, and top-of-atmosphere radiation. While the ML models reproduce key aspects of the physical model response, particularly the response of precipitation, some exhibit notable departures from robust physical responses, including radiative responses and land region warming. Our results highlight the promise and current limitations of ML models for climate change applications and suggest that further improvements are needed for robust out-of-sample generalization.
Authors: Marc Amor\'os-Trepat, Luis Medrano-Navarro, Qiang Liu, Luca Guastoni, Nils Thuerey
Abstract: The reconstruction of unsteady flow fields from limited measurements is a challenging and crucial task for many engineering applications. Machine learning models are gaining popularity for solving this problem due to their ability to learn complex patterns from data and to generalize across diverse conditions. Among these, diffusion models have emerged as being particularly powerful for generative tasks, producing high-quality samples by iteratively refining noisy inputs. In contrast to other methods, these generative models are capable of reconstructing the smallest scales of the fluid spectrum. In this work, we introduce a novel sampling method for diffusion models that enables the reconstruction of high-fidelity samples by guiding the reverse process using the available sparse data. Moreover, we enhance the reconstructions with available physics knowledge using a conflict-free update method during training. To evaluate the effectiveness of our method, we conduct experiments on 2 and 3-dimensional turbulent flow data. Our method consistently outperforms other diffusion-based methods in predicting the fluid's structure and in pixel-wise accuracy. This study underscores the remarkable potential of diffusion models in reconstructing flow field data, paving the way for leveraging them in fluid dynamics research and applications ranging from super-resolution to reconstructions of experiments.
Authors: Ido Nachum, R\"udiger Urbanke, Thomas Weinberger
Abstract: Despite ongoing theoretical research on cross-validation (CV), many theoretical questions remain widely open. This motivates our investigation into how properties of algorithm-distribution pairs can affect the choice for the number of folds in $k$-fold CV. Our results consist of a novel decomposition of the mean-squared error of cross-validation for risk estimation, which explicitly captures the correlations of error estimates across overlapping folds and includes a novel algorithmic stability notion, squared loss stability, that is considerably weaker than the typically required hypothesis stability in other comparable works. Furthermore, we prove: 1. For any learning algorithm that minimizes empirical risk, the mean-squared error of the $k$-fold cross-validation estimator $\widehat{L}_{\mathrm{CV}}^{(k)}$ of the population risk $L_{D}$ satisfies the following minimax lower bound: \[ \min_{k \mid n} \max_{D} \mathbb{E}\left[\big(\widehat{L}_{\mathrm{CV}}^{(k)} - L_{D}\big)^{2}\right]=\Omega\big(\sqrt{k^*}/n\big), \] where $n$ is the sample size, $k$ the number of folds, and $k^*$ denotes the number of folds attaining the minimax optimum. This shows that even under idealized conditions, for large values of $k$, CV cannot attain the optimum of order $1/n$ achievable by a validation set of size $n$, reflecting an inherent penalty caused by dependence between folds. 2. Complementing this, we exhibit learning rules for which \[ \max_{D}\mathbb{E}\!\left[\big(\widehat{L}_{\mathrm{CV}}^{(k)} - L_{D}\big)^{2}\right]=\Omega(k/n), \] matching (up to constants) the accuracy of a hold-out estimator of a single fold of size $n/k$. Together these results delineate the fundamental trade-off in resampling-based risk estimation: CV cannot fully exploit all $n$ samples for unbiased risk evaluation, and its minimax performance is pinned between the $k/n$ and $\sqrt{k}/n$ regimes.
Authors: Zhenyu Chen, Yuguo Shao, Zhengwei Liu, Zhaohui Wei
Abstract: Quantum algorithms based on parameterized quantum circuits (PQCs) have enabled a wide range of applications on near-term quantum devices. However, existing PQC architectures face several challenges, among which the ``barren plateaus" phenomenon is particularly prominent. In such cases, the loss function concentrates exponentially with increasing system size, thereby hindering effective parameter optimization. To address this challenge, we propose a general and hardware-efficient method for eliminating barren plateaus in an arbitrary PQC. Specifically, our approach achieves this by inserting a layer of easily implementable quantum channels into the original PQC, each channel requiring only one ancilla qubit and four additional gates, yielding a modified PQC (MPQC) that is provably at least as expressive as the original PQC and, under mild assumptions, is guaranteed to be free from barren plateaus. Furthermore, by appropriately adjusting the structure of MPQCs, we rigorously prove that any parameter in the original PQC can be made trainable. Importantly, the absence of barren plateaus in MPQCs is robust against realistic noise, making our approach directly applicable to near-term quantum hardware. Numerical simulations demonstrate that MPQC effectively eliminates barren plateaus in PQCs for preparing thermal states of systems with up to 100 qubits and 2400 layers. Furthermore, in end-to-end simulations, MPQC significantly outperforms PQC in finding the ground-state energy of a complex Hamiltonian.
Authors: Emanuele Pardini, Katerina Papagiannouli
Abstract: Unbalanced optimal transport (UOT) provides a flexible way to match or compare nonnegative finite Radon measures. However, UOT requires a predefined ground transport cost, which may misrepresent the data's underlying geometry. Choosing such a cost is particularly challenging when datasets live in heterogeneous spaces, often motivating practitioners to adopt Gromov-Wasserstein formulations. To address this challenge, we introduce cost-regularized unbalanced optimal transport (CR-UOT), a framework that allows the ground cost to vary while allowing mass creation and removal. We show that CR-UOT incorporates unbalanced Gromov-Wasserstein type problems through families of inner-product costs parameterized by linear transformations, enabling the matching of measures or point clouds across Euclidean spaces. We develop algorithms for such CR-UOT problems using entropic regularization and demonstrate that this approach improves the alignment of heterogeneous single-cell omics profiles, especially when many cells lack direct matches.
Authors: Yang-Wen Sun, Katerina Papagiannouli, Vladimir Spokoiny
Abstract: Inspired by graph-based methodologies, we introduce a novel graph-spanning algorithm designed to identify changes in both offline and online data across low to high dimensions. This versatile approach is applicable to Euclidean and graph-structured data with unknown distributions, while maintaining control over error probabilities. Theoretically, we demonstrate that the algorithm achieves high detection power when the magnitude of the change surpasses the lower bound of the minimax separation rate, which scales on the order of $\sqrt{nd}$. Our method outperforms other techniques in terms of accuracy for both Gaussian and non-Gaussian data. Notably, it maintains strong detection power even with small observation windows, making it particularly effective for online environments where timely and precise change detection is critical.
Authors: Yu-Pin Hsu, Yi-Hsuan Tseng
Abstract: This paper investigates an information update system in which a mobile device monitors a physical process and sends status updates to an access point (AP). A fundamental trade-off arises between the timeliness of the information maintained at the AP and the update cost incurred at the device. To address this trade-off, we propose an online algorithm that determines when to transmit updates using only available observations. The proposed algorithm asymptotically achieves the optimal competitive ratio against an adversary that can simultaneously manipulate multiple sources of uncertainty, including the operation duration, information staleness, update cost, and update opportunities. Furthermore, by incorporating machine learning (ML) advice of unknown reliability into the design, we develop an ML-augmented algorithm that asymptotically attains the optimal consistency-robustness trade-off, even when the adversary can additionally corrupt the ML advice. The optimal competitive ratio scales linearly with the range of update costs, but is unaffected by other sources of uncertainty. Moreover, an optimal competitive online algorithm exhibits a threshold-like response to the ML advice: it either fully trusts or completely ignores the ML advice, as partially trusting the advice cannot improve the consistency without severely degrading the robustness. Extensive simulations in stochastic settings further validate the theoretical findings in the adversarial environment.
Authors: Benyamin Hosseiny, Kamirul Kamirul, Odysseas Pappas, Alin Achim
Abstract: Synthetic Aperture Radar (SAR) offers a unique capability for all-weather, space-based maritime activity monitoring by capturing and imaging strong reflections from ships at sea. A well-defined challenge in this domain is ship type classification. Due to the high diversity and complexity of ship types, accurate recognition is difficult and typically requires specialized deep learning models. These models, however, depend on large, high-quality ground-truth datasets to achieve robust performance and generalization. Furthermore, the growing variety of SAR satellites operating at different frequencies and spatial resolutions has amplified the need for more annotated datasets to enhance model accuracy. To address this, we present the NovaSAR Automated Ship Target Recognition (NASTaR) dataset. This dataset comprises of 3415 ship patches extracted from NovaSAR S-band imagery, with labels matched to AIS data. It includes distinctive features such as 23 unique classes, inshore/offshore separation, and an auxiliary wake dataset for patches where ship wakes are visible. We validated the dataset applicability across prominent ship-type classification scenarios using benchmark deep learning models. Results demonstrate over 60% accuracy for classifying four major ship types, over 70% for a three-class scenario, more than 75% for distinguishing cargo from tanker ships, and over 87% for identifying fishing vessels. The NASTaR dataset is available at https://doi.org/10.5523/bris.2tfa6x37oerz2lyiw6hp47058, while relevant codes for benchmarking and analysis are available at https://github.com/benyaminhosseiny/nastar.
URLs: https://doi.org/10.5523/bris.2tfa6x37oerz2lyiw6hp47058,, https://github.com/benyaminhosseiny/nastar.
Authors: Samya Praharaj, Koulik Khamaru
Abstract: Statistical inference in contextual bandits is challenging due to the adaptive, non-i.i.d. nature of the data. A growing body of work shows that classical least-squares inference can fail under adaptive sampling, and that valid confidence intervals for linear functionals typically require an inflation of order $\sqrt{d \log T}$. This phenomenon -- often termed the price of adaptivity -- reflects the intrinsic difficulty of reliable inference under general contextual bandit policies. A key structural condition that overcomes this limitation is the stability condition of Lai and Wei, which requires the empirical feature covariance to converge to a deterministic limit. When stability holds, the ordinary least-squares estimator satisfies a central limit theorem, and classical Wald-type confidence intervals remain asymptotically valid under adaptation, without incurring the $\sqrt{d \log T}$ price of adaptivity. In this paper, we propose and analyze a regularized EXP4 algorithm for linear contextual bandits. Our first main result shows that this procedure satisfies the Lai--Wei stability condition and therefore admits valid Wald-type confidence intervals for linear functionals. We additionally provide quantitative rates of convergence in the associated central limit theorem. Our second result establishes that the same algorithm achieves regret guarantees that are minimax optimal up to logarithmic factors, demonstrating that stability and statistical efficiency can coexist within a single contextual bandit method. As an application of our theory, we show how it can be used to construct confidence intervals for the conditional average treatment effect (CATE) under adaptively collected data. Finally, we complement our theory with simulations illustrating the empirical normality of the resulting estimators and the sharpness of the corresponding confidence intervals.
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: Fady Rezk, Yuangang Pan, Chuan-Sheng Foo, Xun Xu, Nancy Chen, Henry Gouk, Timothy Hospedales
Abstract: Personalized alignment from preference data has focused primarily on improving personal reward model (RM) accuracy, with the implicit assumption that better preference ranking translates to better personalized behavior. However, in deployment, computational constraints necessitate inference-time adaptation such as reward-guided decoding (RGD) rather than per-user policy fine-tuning. This creates a critical but overlooked requirement: reward models must not only rank preferences accurately but also effectively guide generation. We demonstrate that standard RM accuracy fails catastrophically as a selection criterion for deployment-ready personalized rewards. We introduce policy accuracy; a metric quantifying whether RGD-adapted LLMs correctly discriminate between preferred and dispreferred responses and show that upstream RM accuracy correlates only weakly with downstream policy accuracy (Kendall's tau = 0.08--0.31). More critically, we introduce Pref-LaMP the first personalized alignment benchmark with ground-truth user completions, enabling direct behavioural evaluation. On Pref-LaMP, we expose a complete decoupling between discriminative ranking and generation metrics: methods with 20-point RM accuracy differences produce almost identical output quality, and methods with high ranking accuracy can fail to generate behaviorally aligned responses. These findings reveal that the field has been optimizing for proxy metrics that do not predict deployment performance, and that current personalized alignment methods fail to operationalize preferences into behavioral adaptation under realistic deployment constraints. In contrast, we find simple in-context learning (ICL) to be highly effective - dominating all reward-guided methods for models $\geq$3B parameters, achieving $\sim$3 point ROUGE-1 gains over the best reward method at 7B scale.
Authors: Anbang Zhang, Chenyuan Feng, Wai Ho Mow, Jia Ye, Shuaishuai Guo, Geyong Min, Tony Q. S. Quek
Abstract: The rapid deployment of mega-constellations is driving the long-term vision of space data centers (SDCs), where interconnected satellites form in-orbit distributed computing and learning infrastructures. Enabling distributed federated learning in such systems is challenging because iterative training requires frequent aggregation over inter-satellite links that are bandwidth- and energy-constrained, and the link conditions can be highly dynamic. In this work, we exploit over-the-air computation (AirComp) as an in-network aggregation primitive. However, conventional coherent AirComp relies on stringent phase alignment, which is difficult to maintain in space environments due to satellite jitter and Doppler effects. To overcome this limitation, we propose OptiVote, a robust and communication-efficient non-coherent free-space optical (FSO) AirComp framework for federated learning toward Space Data Centers. OptiVote integrates sign stochastic gradient descent (signSGD) with a majority-vote (MV) aggregation principle and pulse-position modulation (PPM), where each satellite conveys local gradient signs by activating orthogonal PPM time slots. The aggregation node performs MV detection via non-coherent energy accumulation, transforming phase-sensitive field superposition into phase-agnostic optical intensity combining, thereby eliminating the need for precise phase synchronization and improving resilience under dynamic impairments. To mitigate aggregation bias induced by heterogeneous FSO channels, we further develop an importance-aware, channel state information (CSI)-free dynamic power control scheme that balances received energies without additional signaling. We provide theoretical analysis by characterizing the aggregate error probability under statistical FSO channels and establishing convergence guarantees for non-convex objectives.
Authors: Osvaldo Simeone
Abstract: The rapid growth of artificial intelligence (AI) has brought novel data processing and generative capabilities but also escalating energy requirements. This challenge motivates renewed interest in neuromorphic computing principles, which promise brain-like efficiency through discrete and sparse activations, recurrent dynamics, and non-linear feedback. In fact, modern AI architectures increasingly embody neuromorphic principles through heavily quantized activations, state-space dynamics, and sparse attention mechanisms. This paper elaborates on the connections between neuromorphic models, state-space models, and transformer architectures through the lens of the distinction between intra-token processing and inter-token processing. Most early work on neuromorphic AI was based on spiking neural networks (SNNs) for intra-token processing, i.e., for transformations involving multiple channels, or features, of the same vector input, such as the pixels of an image. In contrast, more recent research has explored how neuromorphic principles can be leveraged to design efficient inter-token processing methods, which selectively combine different information elements depending on their contextual relevance. Implementing associative memorization mechanisms, these approaches leverage state-space dynamics or sparse self-attention. Along with a systematic presentation of modern neuromorphic AI models through the lens of intra-token and inter-token processing, training methodologies for neuromorphic AI models are also reviewed. These range from surrogate gradients leveraging parallel convolutional processing to local learning rules based on reinforcement learning mechanisms.
Authors: Hongkun Yang, Lionel Z. Wang, Wei Fan, Yiran Hu, Lixu Wang, Chenyu Liu, Shenghong Fu, Haoyang Li, Xin Xu, Jiexin Zheng, Wei Dong
Abstract: Legal judgment generation is a critical task in legal intelligence. However, existing research in legal judgment generation has predominantly focused on first-instance trials, relying on static fact-to-verdict mappings while neglecting the dialectical nature of appellate (second-instance) review. To address this, we introduce AppellateGen, a benchmark for second-instance legal judgment generation comprising 7,351 case pairs. The task requires models to draft legally binding judgments by reasoning over the initial verdict and evidentiary updates, thereby modeling the causal dependency between trial stages. We further propose a judicial Standard Operating Procedure (SOP)-based Legal Multi-Agent System (SLMAS) to simulate judicial workflows, which decomposes the generation process into discrete stages of issue identification, retrieval, and drafting. Experimental results indicate that while SLMAS improves logical consistency, the complexity of appellate reasoning remains a substantial challenge for current LLMs. The dataset and code are publicly available at: https://anonymous.4open.science/r/AppellateGen-5763.
Authors: Jisoo Lee, Sunki Hong
Abstract: Accurate grid load forecasting is safety-critical: under-predictions risk supply shortfalls, while symmetric error metrics mask this operational asymmetry. We introduce a grid-specific evaluation framework (Asymmetric MAPE, Under-Prediction Rate, and Reserve Margin) that directly measures operational risk rather than statistical accuracy alone. Using this framework, we conduct a systematic evaluation of Mamba-based State Space Models for California grid forecasting on a weather-aligned CA ISO-TAC dataset spanning Nov 2023 to Nov 2025 (84,498 hourly records across 5 transmission areas). Our analysis reveals that standard accuracy metrics are poor proxies for operational safety: models with identical MAPE can require vastly different reserve margins. We demonstrate that forecast errors are weakly but statistically significantly associated with temperature (r = 0.16), motivating weather-aware modeling rather than loss function modification alone. The S-Mamba model achieves the lowest 99.5th-percentile reserve margin (14.12 percent) compared to 16.66 percent for iTransformer, demonstrating superior forecast reliability under a 99.5th-percentile tail-risk reserve proxy.
Authors: Jiwei Guan, Haibo Jin, Haohan Wang
Abstract: Recent advancements in Large Vision-Language Models (LVLMs) have shown groundbreaking capabilities across diverse multimodal tasks. However, these models remain vulnerable to adversarial jailbreak attacks, where adversaries craft subtle perturbations to bypass safety mechanisms and trigger harmful outputs. Existing white-box attacks methods require full model accessibility, suffer from computing costs and exhibit insufficient adversarial transferability, making them impractical for real-world, black-box settings. To address these limitations, we propose a black-box jailbreak attack on LVLMs via Zeroth-Order optimization using Simultaneous Perturbation Stochastic Approximation (ZO-SPSA). ZO-SPSA provides three key advantages: (i) gradient-free approximation by input-output interactions without requiring model knowledge, (ii) model-agnostic optimization without the surrogate model and (iii) lower resource requirements with reduced GPU memory consumption. We evaluate ZO-SPSA on three LVLMs, including InstructBLIP, LLaVA and MiniGPT-4, achieving the highest jailbreak success rate of 83.0% on InstructBLIP, while maintaining imperceptible perturbations comparable to white-box methods. Moreover, adversarial examples generated from MiniGPT-4 exhibit strong transferability to other LVLMs, with ASR reaching 64.18%. These findings underscore the real-world feasibility of black-box jailbreaks and expose critical weaknesses in the safety mechanisms of current LVLMs
Authors: Michael C. Darling, Alan H. Hesu, Michael A. Mardikes, Brian C. McGuigan, Reed M. Milewicz
Abstract: We propose a maturity-based framework for certifying embodied AI systems through explicit measurement mechanisms. We argue that certifiable embodied AI requires structured assessment frameworks, quantitative scoring mechanisms, and methods for navigating multi-objective trade-offs inherent in trustworthiness evaluation. We demonstrate this approach using uncertainty quantification as an exemplar measurement mechanism and illustrate feasibility through an Uncrewed Aircraft System (UAS) detection case study.
Authors: Dennis Holzmann, Sven Wachsmuth
Abstract: We present a novel approach for egocentric action recognition that leverages 2D point tracks as an additional motion cue. While most existing methods rely on RGB appearance, human pose estimation, or their combination, our work demonstrates that tracking randomly sampled image points across video frames can substantially improve recognition accuracy. Unlike prior approaches, we do not detect hands, objects, or interaction regions. Instead, we employ CoTracker to follow a set of randomly initialized points through each video and use the resulting trajectories, together with the corresponding image frames, as input to a Transformer-based recognition model. Surprisingly, our method achieves notable gains even when only the initial frame and its associated point tracks are provided, without incorporating the full video sequence. Experimental results confirm that integrating 2D point tracks consistently enhances performance compared to the same model trained without motion information, highlighting their potential as a lightweight yet effective representation for egocentric action understanding.
Authors: Xiang Zhang, Huan Yan, Jinyang Huang, Bin Liu, Yuanhao Feng, Jianchun Liu, Meng Li, Fusang Zhang, Zhi Liu
Abstract: In this paper, we propose GesFi, a novel WiFi-based gesture recognition system that introduces WiFi latent domain mining to redefine domains directly from the data itself. GesFi first processes raw sensing data collected from WiFi receivers using CSI-ratio denoising, Short-Time Fast Fourier Transform, and visualization techniques to generate standardized input representations. It then employs class-wise adversarial learning to suppress gesture semantic and leverages unsupervised clustering to automatically uncover latent domain factors responsible for distributional shifts. These latent domains are then aligned through adversarial learning to support robust cross-domain generalization. Finally, the system is applied to the target environment for robust gesture inference. We deployed GesFi under both single-pair and multi-pair settings using commodity WiFi transceivers, and evaluated it across multiple public datasets and real-world environments. Compared to state-of-the-art baselines, GesFi achieves up to 78% and 50% performance improvements over existing adversarial methods, and consistently outperforms prior generalization approaches across most cross-domain tasks.
Authors: Cheng Qian, Emre Can Acikgoz, Bingxuan Li, Xiusi Chen, Yuji Zhang, Bingxiang He, Qinyu Luo, Dilek Hakkani-T\"ur, Gokhan Tur, Yunzhu Li, Heng Ji
Abstract: Agents built on vision-language models increasingly face tasks that demand anticipating future states rather than relying on short-horizon reasoning. Generative world models offer a promising remedy: agents could use them as external simulators to foresee outcomes before acting. This paper empirically examines whether current agents can leverage such world models as tools to enhance their cognition. Across diverse agentic and visual question answering tasks, we observe that some agents rarely invoke simulation (fewer than 1%), frequently misuse predicted rollouts (approximately 15%), and often exhibit inconsistent or even degraded performance (up to 5%) when simulation is available or enforced. Attribution analysis further indicates that the primary bottleneck lies in the agents' capacity to decide when to simulate, how to interpret predicted outcomes, and how to integrate foresight into downstream reasoning. These findings underscore the need for mechanisms that foster calibrated, strategic interaction with world models, paving the way toward more reliable anticipatory cognition in future agent systems.
Authors: Marvin Illian, Ramin Khalili, Antonio A. de A. Rocha, Lin Wang
Abstract: The widespread deployment of 5G networks, together with the coexistence of 4G/LTE networks, provides mobile devices a diverse set of candidate cells to connect to. However, associating mobile devices to cells to maximize overall network performance, a.k.a. cell (re)selection, remains a key challenge for mobile operators. Today, cell (re)selection parameters are typically configured manually based on operator experience and rarely adapted to dynamic network conditions. In this work, we ask: Can an agent automatically learn and adapt cell (re)selection parameters to consistently improve network performance? We present a reinforcement learning (RL)-based framework called CellPilot that adaptively tunes cell (re)selection parameters by learning spatiotemporal patterns of mobile network dynamics. Our study with real-world data demonstrates that even a lightweight RL agent can outperform conventional heuristic reconfigurations by up to 167%, while generalizing effectively across different network scenarios. These results indicate that data-driven approaches can significantly improve cell (re)selection configurations and enhance mobile network performance.