Authors: Ruixiang Wang, Yuyang Hong, Shiming Xiang, Chunhong Pan
Abstract: Transformers have demonstrated success in graph learning, particularly for node-level tasks. However, existing methods encounter an information bottleneck when generating graph-level representations. The prevalent single token paradigm fails to fully leverage the inherent strength of self-attention in encoding token sequences, and degenerates into a weighted sum of node signals. To address this issue, we design a novel serialized token paradigm to encapsulate global signals more effectively. Specifically, a graph serialization method is proposed to aggregate node signals into serialized graph tokens, with positional encoding being automatically involved. Then, stacked self-attention layers are applied to encode this token sequence and capture its internal dependencies. Our method can yield more expressive graph representations by modeling complex interactions among multiple graph tokens. Experimental results show that our method achieves state-of-the-art results on several graph-level benchmarks. Ablation studies verify the effectiveness of the proposed modules.
Authors: Jinjin Guo, Yexin Li, Zhichao Huang, Jun Fang, Zhiyuan Liu, Chao Liu, Pengzhang Liu, Qixia Jiang
Abstract: Large-scale multimodal contrastive learning has recently achieved impressive success in learning rich and transferable representations, yet it remains fundamentally limited by the uniform treatment of feature dimensions and the neglect of the intrinsic spectral structure of the learned features. Empirical evidence indicates that high-dimensional embeddings tend to collapse into narrow cones, concentrating task-relevant semantics in a small subspace, while the majority of dimensions remain occupied by noise and spurious correlations. Such spectral imbalance and entanglement undermine model generalization. We propose Spectral Disentanglement and Enhancement (SDE), a novel framework that bridges the gap between the geometry of the embedded spaces and their spectral properties. Our approach leverages singular value decomposition to adaptively partition feature dimensions into strong signals that capture task-critical semantics, weak signals that reflect ancillary correlations, and noise representing irrelevant perturbations. A curriculum-based spectral enhancement strategy is then applied, selectively amplifying informative components with theoretical guarantees on training stability. Building upon the enhanced features, we further introduce a dual-domain contrastive loss that jointly optimizes alignment in both the feature and spectral spaces, effectively integrating spectral regularization into the training process and encouraging richer, more robust representations. Extensive experiments on large-scale multimodal benchmarks demonstrate that SDE consistently improves representation robustness and generalization, outperforming state-of-the-art methods. SDE integrates seamlessly with existing contrastive pipelines, offering an effective solution for multimodal representation learning.
Authors: Djohan Bonnet, Jamie Lohoff, Jan Finkbeiner, Elidona Skhikerujah, Emre Neftci
Abstract: In-Context Learning (ICL) in transformers acts as an online associative memory and is believed to underpin their high performance on complex sequence processing tasks. However, in gated linear attention models, this memory has a fixed capacity and is prone to interference, especially for long sequences. We propose Palimpsa, a self-attention model that views ICL as a continual learning problem that must address a stability-plasticity dilemma. Palimpsa uses Bayesian metaplasticity, where the plasticity of each attention state is tied to an importance state grounded by a prior distribution that captures accumulated knowledge. We demonstrate that various gated linear attention models emerge as specific architecture choices and posterior approximations, and that Mamba2 is a special case of Palimpsa where forgetting dominates. This theoretical link enables the transformation of any non-metaplastic model into a metaplastic one, significantly expanding its memory capacity. Our experiments show that Palimpsa consistently outperforms baselines on the Multi-Query Associative Recall (MQAR) benchmark and on Commonsense Reasoning tasks.
Authors: Zachary N. Flamholz, Dillon Tracy, Ripple Khera, Jordan Wolinsky, Nicholas Lee, Nathaniel Tann, Xiao Yin Zhu, Harry Phillips, Jeffrey Sherman
Abstract: Accurate risk stratification in patients with overweight or obesity is critical for guiding preventive care and allocating high-cost therapies such as GLP-1 receptor agonists. We present PatientTPP, a neural temporal point process (TPP) model trained on over 500,000 real-world clinical trajectories to learn patient representations from sequences of diagnoses, labs, and medications. We extend existing TPP modeling approaches to include static and numeric features and incorporate clinical knowledge for event encoding. PatientTPP representations support downstream prediction tasks, including classification of obesity-associated outcomes in low-risk individuals, even for events not explicitly modeled during training. In health economic evaluation, PatientTPP outperformed body mass index in stratifying patients by future cardiovascular-related healthcare costs, identifying higher-risk patients more efficiently. By modeling both the type and timing of clinical events, PatientTPP offers an interpretable, general-purpose foundation for patient risk modeling with direct applications to obesity-related care and cost targeting.
Authors: Ruihan Xu, Yuting Gao, Lan Wang, Jianing Li, Weihao Chen, Qingpei Guo, Ming Yang, Shiliang Zhang
Abstract: Large Multimodal Models (LMMs) have achieved remarkable success in vision-language tasks, yet their vast parameter counts are often underutilized during both training and inference. In this work, we embrace the idea of looping back to move forward: reusing model parameters through recursive refinement to extract stronger multimodal representations without increasing model size. We propose RecursiveVLM, a recursive Transformer architecture tailored for LMMs. Two key innovations enable effective looping: (i) a Recursive Connector that aligns features across recursion steps by fusing intermediate-layer hidden states and applying modality-specific projections, respecting the distinct statistical structures of vision and language tokens; (ii) a Monotonic Recursion Loss that supervises every step and guarantees performance improves monotonically with recursion depth. This design transforms recursion into an on-demand refinement mechanism: delivering strong results with few loops on resource-constrained devices and progressively improving outputs when more computation resources are available. Experiments show consistent gains of +3% over standard Transformers and +7% over vanilla recursive baselines, demonstrating that strategic looping is a powerful path toward efficient, deployment-adaptive LMMs.
Authors: Ruxuan Chen, Fang Sun
Abstract: State Space Models (SSMs), particularly Mamba, have shown potential in long-term time series forecasting. However, existing Mamba-based architectures often struggle with datasets characterized by non-stationary patterns. A key observation from time series theory is that the statistical nature of inter-variable relationships differs fundamentally between the trend and seasonal components of a decomposed series. Trend relationships are often driven by a few common stochastic factors or long-run equilibria, suggesting that they reside on a lower-dimensional manifold. In contrast, seasonal relationships involve dynamic, high-dimensional interactions like phase shifts and amplitude co-movements, requiring more expressive modeling. In this paper, we propose DMamba, a novel forecasting model that explicitly aligns architectural complexity with this component-specific characteristic. DMamba employs seasonal-trend decomposition and processes the components with specialized, differentially complex modules: a variable-direction Mamba encoder captures the rich, cross-variable dynamics within the seasonal component, while a simple Multi-Layer Perceptron (MLP) suffices to learn from the lower-dimensional inter-variable relationships in the trend component. Extensive experiments on diverse datasets demonstrate that DMamba sets a new state-of-the-art (SOTA), consistently outperforming both recent Mamba-based architectures and leading decomposition-based models.
Authors: Carlos Heredia
Abstract: In this paper, we derive an accelerated continuous-time formulation of Adam by modeling it as a second-order integro-differential dynamical system. We relate this inertial nonlocal model to an existing first-order nonlocal Adam flow through an $\alpha$-refinement limit, and we provide Lyapunov-based stability and convergence analyses. We also introduce an Adam-inspired nonlocal Lagrangian formulation, offering a variational viewpoint. Numerical simulations on Rosenbrock-type examples show agreement between the proposed dynamics and discrete Adam.
Authors: Hossam Amer, Rezaul Karim, Ali Pourranjbar, Weiwei Zhang, Walid Ahmed, Boxing Chen
Abstract: With the rapid growth of large language models (LLMs), a wide range of methods have been developed to distribute computation and memory across hardware devices for efficient training and inference. While existing surveys provide descriptive overviews of these techniques, systematic analysis of their benefits and trade offs and how such insights can inform principled methodology for designing optimal distributed systems remain limited. This paper offers a comprehensive review of collective operations and distributed parallel strategies, complemented by mathematical formulations to deepen theoretical understanding. We further examine hybrid parallelization designs, emphasizing communication computation overlap across different stages of model deployment, including both training and inference. Recent advances in automated search for optimal hybrid parallelization strategies using cost models are also discussed. Moreover, we present case studies with mainstream architecture categories to reveal empirical insights to guide researchers and practitioners in parallelism strategy selection. Finally, we highlight open challenges and limitations of current LLM training paradigms and outline promising directions for the next generation of large scale model development.
Authors: Rohit Dutta, Paramita Koley, Soham Poddar, Janardan Misra, Sanjay Podder, Naveen Balani, Saptarshi Ghosh, Niloy Ganguly
Abstract: Speculative decoding has emerged as an effective method to reduce latency and inference cost of LLM inferences. However, there has been inadequate attention towards the energy requirements of these models. To address this gap, this paper presents a comprehensive survey of energy requirements of speculative decoding strategies, with detailed analysis on how various factors -- model size and family, speculative decoding strategies, and dataset characteristics -- influence the energy optimizations.
Authors: Daniele Caligiore
Abstract: The capacity to transfer knowledge across scientific domains relies on shared organizational principles. However, existing transfer-learning methodologies often fail to bridge radically heterogeneous systems, particularly under severe data scarcity or stochastic noise. This study formalizes Explainable Cross-Domain Transfer Learning (X-CDTL), a framework unifying network science and explainable artificial intelligence to identify structural invariants that generalize across biological, linguistic, molecular, and social networks. By introducing the Importance Inversion Transfer (IIT) mechanism, the framework prioritizes domain-invariant structural anchors over idiosyncratic, highly discriminative features. In anomaly detection tasks, models guided by these principles achieve significant performance gains - exhibiting a 56\% relative improvement in decision stability under extreme noise - over traditional baselines. These results provide evidence for a shared organizational signature across heterogeneous domains, establishing a principled paradigm for cross-disciplinary knowledge propagation. By shifting from opaque latent representations to explicit structural laws, this work advances machine learning as a robust engine for scientific discovery.
Authors: Elisa Roldan, Tasneem Sabir
Abstract: Electrospinning is a powerful technique for producing micro to nanoscale fibers with application specific architectures. Small variations in solution or operating conditions can shift the jet regime, generating non Gaussian fiber diameter distributions. Despite substantial progress, no existing framework enables inverse design toward desired fiber outcomes while integrating polymer solvent chemical constraints or predicting full distributions. SpinCastML is an open source, distribution aware, chemically informed machine learning and Inverse Monte Carlo (IMC) software for inverse electrospinning design. Built on a rigorously curated dataset of 68,480 fiber diameters from 1,778 datasets across 16 polymers, SpinCastML integrates three structured sampling methods, a suite of 11 high-performance learners, and chemistry aware constraints to predict not only mean diameter but the entire distribution. Cubist model with a polymer balanced Sobol D optimal sampling provides the highest global performance (R2 > 0.92). IMC accurately captures the fiber distributions, achieving R2 > 0.90 and <1% error between predicted and experimental success rates. The IMC engine supports both retrospective analysis and forward-looking inverse design, generating physically and chemically feasible polymer solvent parameter combinations with quantified success probabilities for user-defined targets. SpinCastML reframes electrospinning from trial and error to a reproducible, data driven design process. As an open source executable, it enables laboratories to analyze their own datasets and co create an expanding community software. SpinCastML reduces experimental waste, accelerates discovery, and democratizes access to advanced modeling, establishing distribution aware inverse design as a new standard for sustainable nanofiber manufacturing across biomedical, filtration, and energy applications.
Authors: Lei You
Abstract: Recent generative and tool-using AI systems can surface a large volume of candidates at low marginal cost, yet only a small fraction can be checked carefully. This creates a decoder-side bottleneck: downstream decision-makers must form reliable posteriors from many public records under scarce attention. We formalize this regime via Attention-Constrained Inference (ACI), in which a cheap screening stage processes $K$ records and an expensive verification stage can follow up on at most $B$ of them. Under Bayes log-loss, we study the maximum achievable reduction in posterior uncertainty per window, which we call \emph{epistemic throughput}. Our main result is a ``JaKoB'' scaling law showing that epistemic throughput has a baseline term that grows linearly with verification and prevalence, and an additional \emph{information-leverage} term that scales as $\sqrt{JKB}$, where $J$ summarizes screening quality. Thus, expanding cheap screening can nonlinearly amplify scarce verification, even when informative records are rare. We further show that this scaling is tight in a weak-screening limit, and that in the sparse-verification regime ($B \ll K$), substantial leverage requires heavy-tailed score distributions; for light-tailed scores the amplification is only logarithmic.
Authors: Awa Khouna, Julien Ferry, Thibaut Vidal
Abstract: Counterfactual explanations are a central tool in interpretable machine learning, yet computing them exactly for complex models remains challenging. For tree ensembles, predictions are piecewise constant over a large collection of axis-aligned hyperrectangles, implying that an optimal counterfactual for a point corresponds to its projection onto the nearest rectangle with an alternative label under a chosen metric. Existing methods largely overlook this geometric structure, relying either on heuristics with no optimality guarantees or on mixed-integer programming formulations that do not scale to interactive use. In this work, we revisit counterfactual generation through the lens of nearest-region search and introduce counterfactual maps, a global representation of recourse for tree ensembles. Leveraging the fact that any tree ensemble can be compressed into an equivalent partition of labeled hyperrectangles, we cast counterfactual search as the problem of identifying the generalized Voronoi cell associated with the nearest rectangle of an alternative label. This leads to an exact, amortized algorithm based on volumetric k-dimensional (KD) trees, which performs branch-and-bound nearest-region queries with explicit optimality certificates and sublinear average query time after a one-time preprocessing phase. Our experimental analyses on several real datasets drawn from high-stakes application domains show that this approach delivers globally optimal counterfactual explanations with millisecond-level latency, achieving query times that are orders of magnitude faster than existing exact, cold-start optimization methods.
Authors: Jonathan von Rad, Yong Cao, Andreas Geiger
Abstract: Model compression is increasingly essential for deploying large language models (LLMs), yet existing evaluations are limited in method coverage and focus primarily on knowledge-centric benchmarks. Thus, we introduce UniComp, a unified evaluation framework for comparing pruning, quantization, and knowledge distillation. UniComp evaluates compressed models along three dimensions: performance, reliability, and efficiency, using a diverse set of capability- and safety-oriented benchmarks together with a hardware-aware efficiency analysis. Through extensive evaluation of six compression techniques on modern LLMs across more than 40 datasets, we find that (i) compression exhibits a consistent knowledge bias, where knowledge-intensive tasks are relatively preserved while reasoning, multilingual, and instruction-following capabilities degrade substantially; (ii) quantization provides the best overall trade-off between retained performance and efficiency, whereas distillation yields strong runtime acceleration gains at high computational cost; and (iii) task-specific calibration can significantly improve the reasoning ability of pruned models by up to 50%.
Authors: Eric Yeats, John Buckheit, Sarah Scullen, Brendan Kennedy, Loc Truong, Davis Brown, Bill Kay, Cliff Joslyn, Tegan Emerson, Michael J. Henry, John Emanuello, Henry Kvinge
Abstract: Hallucination remains a barrier to deploying generative models in high-consequence applications. This is especially true in cases where external ground truth is not readily available to validate model outputs. This situation has motivated the study of geometric signals in the internal state of an LLM that are predictive of hallucination and require limited external knowledge. Given that there are a range of factors that can lead model output to be called a hallucination (e.g., irrelevance vs incoherence), in this paper we ask what specific properties of a hallucination these geometric statistics actually capture. To assess this, we generate a synthetic dataset which varies distinct properties of output associated with hallucination. This includes output correctness, confidence, relevance, coherence, and completeness. We find that different geometric statistics capture different types of hallucinations. Along the way we show that many existing geometric detection methods have substantial sensitivity to shifts in task domain (e.g., math questions vs. history questions). Motivated by this, we introduce a simple normalization method to mitigate the effect of domain shift on geometric statistics, leading to AUROC gains of +34 points in multi-domain settings.
Authors: Aditya Choudhary, Saaketh Desai, Prasad Iyer
Abstract: Analog Ising machines (AIMs) have emerged as a promising paradigm for combinatorial optimization, utilizing physical dynamics to solve Ising problems with high energy efficiency. However, the performance of traditional optimization and sampling algorithms on these platforms is often limited by inherent measurement noise. We introduce BRAIN (Boltzmann Reinforcement for Analog Ising Networks), a distribution learning framework that utilizes variational reinforcement learning to approximate the Boltzmann distribution. By shifting from state-by-state sampling to aggregating information across multiple noisy measurements, BRAIN is resilient to Gaussian noise characteristic of AIMs. We evaluate BRAIN across diverse combinatorial topologies, including the Curie-Weiss and 2D nearest-neighbor Ising systems. We find that under realistic 3\% Gaussian measurement noise, BRAIN maintains 98\% ground state fidelity, whereas Markov Chain Monte Carlo (MCMC) methods degrade to 51\% fidelity. Furthermore, BRAIN reaches the MCMC-equivalent solution up to 192x faster under these conditions. BRAIN exhibits $\mathcal{O}(N^{1.55})$ scaling up to 65,536 spins and maintains robustness against severe measurement uncertainty up to 40\%. Beyond ground state optimization, BRAIN accurately captures thermodynamic phase transitions and metastable states, providing a scalable and noise-resilient method for utilizing analog computing architectures in complex optimizations.
Authors: Guanghui Wang, Satyen Kale
Abstract: In this paper, we study federated optimization for solving stochastic variational inequalities (VIs), a problem that has attracted growing attention in recent years. Despite substantial progress, a significant gap remains between existing convergence rates and the state-of-the-art bounds known for federated convex optimization. In this work, we address this limitation by establishing a series of improved convergence rates. First, we show that, for general smooth and monotone variational inequalities, the classical Local Extra SGD algorithm admits tighter guarantees under a refined analysis. Next, we identify an inherent limitation of Local Extra SGD, which can lead to excessive client drift. Motivated by this observation, we propose a new algorithm, the Local Inexact Proximal Point Algorithm with Extra Step (LIPPAX), and show that it mitigates client drift and achieves improved guarantees in several regimes, including bounded Hessian, bounded operator, and low-variance settings. Finally, we extend our results to federated composite variational inequalities and establish improved convergence guarantees.
Authors: Jonathan Svirsky, Yehonathan Refael, Ofir Lindenbaum
Abstract: Fully finetuning foundation language models (LMs) with billions of parameters is often impractical due to high computational costs, memory requirements, and the risk of overfitting. Although methods like low-rank adapters help address these challenges by adding small trainable modules to the frozen LM, they also increase memory usage and do not reduce inference latency. We uncover an intriguing phenomenon: sparsifying specific model rows and columns enables efficient task adaptation without requiring weight tuning. We propose a scheme for effective finetuning via sparsification using training stochastic gates, which requires minimal trainable parameters, reduces inference time, and removes 20--40\% of model parameters without significant accuracy loss. Empirical results show it outperforms recent finetuning baselines in efficiency and performance. Additionally, we provide theoretical guarantees for the convergence of this stochastic gating process, and show that our method admits a simpler and better-conditioned optimization landscape compared to LoRA. Our results highlight sparsity as a compelling mechanism for task-specific adaptation in LMs.
Authors: Ryozo Masukawa, Sanggeon Yun, Hyunwoo Oh, SuhgHeon Jeong, Raheeb Hassa, Hanning Chen, Wenjun Huang, Mahdi Imani, Pietro Mercati, Nathaniel D. Bastian, Mohsen Imani
Abstract: Recent progress in reinforcement learning with verifiable rewards (RLVR) shows that small, specialized language models (SLMs) can exhibit structured reasoning without relying on large monolithic LLMs. We introduce soft hidden-state collaboration, where multiple heterogeneous frozen SLM experts are integrated through their internal representations via a trainable attention interface. Experiments on Reasoning Gym and GSM8K show that this latent integration is competitive with strong single-model RLVR baselines. Ablations further reveal a dual mechanism of expert utilization: for simpler arithmetic domains, performance gains can largely be explained by static expert preferences, whereas more challenging settings induce increasingly concentrated and structured expert attention over training, indicating emergent specialization in how the router connects to relevant experts. Overall, hidden-state collaboration provides a compact mechanism for leveraging frozen experts, while offering an observational window into expert utilization patterns and their evolution under RLVR.
Authors: Antonio Candelieri, Francesco Archetti
Abstract: Exploiting the analogy between Gaussian Distributions and Gaussian Processes' posterior, we present how the weighted Wasserstein Barycenter of Gaussian Processes (W2BGP) can be used to unify, under a common framework, different exotic Bayesian Optimization (BO) tasks. Specifically, collaborative/federated BO, (synchronous) batch BO, and multi-fidelity BO are considered in this paper. Our empirical analysis proves that each one of these tasks requires just an appropriate weighting schema for the W2BGP, while the entire framework remains untouched. Moreover, we demonstrate that the most well-known BO acquisition functions can be easily re-interpreted under the proposed framework and also enable a more computationally efficient way to deal with the computation of the Wasserstein Barycenter, compared with state-of-the-art methods from the Machine Learning literature. Finally, research perspectives branching from the proposed approach are presented.
Authors: Yangchen Pan, Qizhen Ying, Philip Torr, Bo Liu
Abstract: Existing work has linked properties of a function's gradient to the difficulty of function approximation. Motivated by these insights, we study how gradient information can be leveraged to improve neural network's ability to approximate high-frequency functions, and we propose a gradient-based residual connection as a complement to the standard identity skip connection used in residual networks. We provide simple theoretical intuition for why gradient information can help distinguish inputs and improve the approximation of functions with rapidly varying behaviour. On a synthetic regression task with a high-frequency sinusoidal ground truth, we show that conventional residual connections struggle to capture high-frequency patterns. In contrast, our gradient residual substantially improves approximation quality. We then introduce a convex combination of the standard and gradient residuals, allowing the network to flexibly control how strongly it relies on gradient information. After validating the design choices of our proposed method through an ablation study, we further validate our approach's utility on the single-image super-resolution task, where the underlying function may be high-frequency. Finally, on standard tasks such as image classification and segmentation, our method achieves performance comparable to standard residual networks, suggesting its broad utility.
Authors: Jiacheng Li, Yixiong Meng, Yi wu, Yun Zhao, Sharare Zehtabian, Jiayin Jin, Degao Peng, Jinfeng Zhuang, Qifei Shen, Kungang Li
Abstract: Deep learning recommendation systems rely on feature interaction modules to model complex user-item relationships across sparse categorical and dense features. In large-scale ad ranking, increasing model capacity is a promising path to improving both predictive performance and business outcomes, yet production serving budgets impose strict constraints on latency and FLOPs. This creates a central tension: we want interaction modules that both scale effectively with additional compute and remain compute-efficient at serving time. In this work, we study how to scale feature interaction modules under a fixed serving budget. We find that naively scaling DCNv2 and MaskNet, despite their widespread adoption in industry, yields rapidly diminishing offline gains in the Pinterest ads ranking system. To overcome aforementioned limitations, we propose ML-DCN, an interaction module that integrates an instance-conditioned mask into a low-rank crossing layer, enabling per-example selection and amplification of salient interaction directions while maintaining efficient computation. This novel architecture combines the strengths of DCNv2 and MaskNet, scales efficiently with increased compute, and achieves state-of-the-art performance. Experiments on a large internal Pinterest ads dataset show that ML-DCN achieves higher AUC than DCNv2, MaskNet, and recent scaling-oriented alternatives at matched FLOPs, and it scales more favorably overall as compute increases, exhibiting a stronger AUC-FLOPs trade-off. Finally, online A/B tests demonstrate statistically significant improvements in key ads metrics (including CTR and click-quality measures) and ML-DCN has been deployed in the production system with neutral serving cost.
Authors: Camille Little, Madeline Navarro, Santiago Segarra, Genevera Allen
Abstract: As machine learning models increasingly impact society, their opaque nature poses challenges to trust and accountability, particularly in fairness contexts. Understanding how individual features influence model outcomes is crucial for building interpretable and equitable models. While feature importance metrics for accuracy are well-established, methods for assessing feature contributions to fairness remain underexplored. We propose two model-agnostic approaches to measure fair feature importance. First, we propose to compare model fairness before and after permuting feature values. This simple intervention-based approach decouples a feature and model predictions to measure its contribution to training. Second, we evaluate the fairness of models trained with and without a given feature. This occlusion-based score enjoys dramatic computational simplification via minipatch learning. Our empirical results reflect the simplicity and effectiveness of our proposed metrics for multiple predictive tasks. Both methods offer simple, scalable, and interpretable solutions to quantify the influence of features on fairness, providing new tools for responsible machine learning development.
Authors: Xiaofeng Xiao, Xiao Hu, Yang Ye, Xubo Yue
Abstract: Reinforcement learning (RL) has achieved remarkable success in a wide range of sequential decision-making problems. Recent diffusion-based policies further improve RL by modeling complex, high-dimensional action distributions. However, existing diffusion policies primarily rely on statistical associations and fail to explicitly account for causal relationships among states, actions, and rewards, limiting their ability to identify which action components truly cause high returns. In this paper, we propose Causality-guided Diffusion Policy (CausalGDP), a unified framework that integrates causal reasoning into diffusion-based RL. CausalGDP first learns a base diffusion policy and an initial causal dynamical model from offline data, capturing causal dependencies among states, actions, and rewards. During real-time interaction, the causal information is continuously updated and incorporated as a guidance signal to steer the diffusion process toward actions that causally influence future states and rewards. By explicitly considering causality beyond association, CausalGDP focuses policy optimization on action components that genuinely drive performance improvements. Experimental results demonstrate that CausalGDP consistently achieves competitive or superior performance over state-of-the-art diffusion-based and offline RL methods, especially in complex, high-dimensional control tasks.
Authors: Julien Guit\'e-Vinet, Alexandre Blondin Mass\'e, \'Eric Beaudry
Abstract: Time series forecasting is a critical task across domains such as energy, finance, and meteorology, where accurate predictions enable informed decision-making. While transformer-based and large-parameter models have recently achieved state-of-the-art results, their complexity can lead to overfitting and unstable forecasts, especially when older data points become less relevant. In this paper, we propose a lightweight multi-view approach to short-term load forecasting that leverages single-value embeddings and a scaled time-range input to capture temporally relevant features efficiently. We introduce an embedding dropout mechanism to prevent over-reliance on specific features and enhance interpretability. Our method achieves competitive performance with significantly fewer parameters, demonstrating robustness across multiple datasets, including scenarios with noisy or sparse data, and provides insights into the contributions of individual features to the forecast.
Authors: Shreya Saha, Zoe Wanying He, Meenakshi Khosla
Abstract: Comparing representations across neural networks is challenging because representations admit symmetries, such as arbitrary reordering of units or rotations of activation space, that obscure underlying equivalence between models. We introduce a barycentric alignment framework that quotients out these nuisance symmetries to construct a universal embedding space across many models. Unlike existing similarity measures, which summarize relationships over entire stimulus sets, this framework enables similarity to be defined at the level of individual stimuli, revealing inputs that elicit convergent versus divergent representations across models. Using this instance-level notion of similarity, we identify systematic input properties that predict representational convergence versus divergence across vision and language model families. We also construct universal embedding spaces for brain representations across individuals and cortical regions, enabling instance-level comparison of representational agreement across stages of the human visual hierarchy. Finally, we apply the same barycentric alignment framework to purely unimodal vision and language models and find that post-hoc alignment into a shared space yields image text similarity scores that closely track human cross-modal judgments and approach the performance of contrastively trained vision-language models. This strikingly suggests that independently learned representations already share sufficient geometric structure for human-aligned cross-modal comparison. Together, these results show that resolving representational similarity at the level of individual stimuli reveals phenomena that cannot be detected by set-level comparison metrics.
Authors: Xincan Feng, Taro Watanabe
Abstract: Cosine similarity is prevalent in contrastive learning, yet it makes an implicit assumption: embedding magnitude is noise. Prior work occasionally found dot product and cosine similarity comparable, but left unanswered WHAT information magnitude carries, WHEN it helps, and HOW to leverage it. We conduct a systematic study through a $2 \times 2$ ablation that independently controls input-side and output-side normalization across text and vision models. Our findings reveal three key insights. First, in text retrieval, output (document) magnitude strongly correlates with relevance (Cohen's $d$ up to 1.80), yielding the largest gains on reasoning-intensive tasks. Second, input and output magnitudes serve asymmetric roles: output magnitude directly scales similarity scores while input magnitude modulates training dynamics. Third, magnitude learning benefits asymmetric tasks (text retrieval, RAG) but harms symmetric tasks (STS, text-image alignment). These findings establish a task symmetry principle: the choice between cosine and dot product depends on whether the task has distinct input roles, enabling cost-free improvements by simply removing an unnecessary constraint.
Authors: Tianhui Liu, Lili Mou
Abstract: Continual learning has become a trending topic in machine learning. Recent studies have discovered an interesting phenomenon called loss of plasticity, referring to neural networks gradually losing the ability to learn new tasks. However, existing plasticity research largely relies on contrived settings with abrupt task transitions, which often do not reflect real-world environments. In this paper, we propose to investigate a gradually changing environment, and we simulate this by input/output interpolation and task sampling. We perform theoretical and empirical analysis, showing that the loss of plasticity is an artifact of abrupt tasks changes in the environment and can be largely mitigated if the world changes gradually.
Authors: Matthias Templ, Oscar Thees, Roman M\"uller
Abstract: Statistical data anonymization increasingly relies on fully synthetic microdata, for which classical identity disclosure measures are less informative than an adversary's ability to infer sensitive attributes from released data. We introduce RAPID (Risk of Attribute Prediction--Induced Disclosure), a disclosure risk measure that directly quantifies inferential vulnerability under a realistic attack model. An adversary trains a predictive model solely on the released synthetic data and applies it to real individuals' quasi-identifiers. For continuous sensitive attributes, RAPID reports the proportion of records whose predicted values fall within a specified relative error tolerance. For categorical attributes, we propose a baseline-normalized confidence score that measures how much more confident the attacker is about the true class than would be expected from class prevalence alone, and we summarize risk as the fraction of records exceeding a policy-defined threshold. This construction yields an interpretable, bounded risk metric that is robust to class imbalance, independent of any specific synthesizer, and applicable with arbitrary learning algorithms. We illustrate threshold calibration, uncertainty quantification, and comparative evaluation of synthetic data generators using simulations and real data. Our results show that RAPID provides a practical, attacker-realistic upper bound on attribute-inference disclosure risk that complements existing utility diagnostics and disclosure control frameworks.
Authors: Benedict Clark, Marta Oliveira, Rick Wilming, Stefan Haufe
Abstract: Explainable AI (XAI) promises to provide insight into machine learning models' decision processes, where one goal is to identify failures such as shortcut learning. This promise relies on the field's assumption that input features marked as important by an XAI must contain information about the target variable. However, it is unclear whether informativeness is indeed the main driver of importance attribution in practice, or if other data properties such as statistical suppression, novelty at test-time, or high feature salience substantially contribute. To clarify this, we trained deep learning models on three variants of a binary image classification task, in which translucent watermarks are either absent, act as class-dependent confounds, or represent class-independent noise. Results for five popular attribution methods show substantially elevated relative importance in watermarked areas (RIW) for all models regardless of the training setting ($R^2 \geq .45$). By contrast, whether the presence of watermarks is class-dependent or not only has a marginal effect on RIW ($R^2 \leq .03$), despite a clear impact impact on model performance and generalisation ability. XAI methods show similar behaviour to model-agnostic edge detection filters and attribute substantially less importance to watermarks when bright image intensities are encoded by smaller instead of larger feature values. These results indicate that importance attribution is most strongly driven by the salience of image structures at test time rather than statistical associations learned by machine learning models. Previous studies demonstrating successful XAI application should be reevaluated with respect to a possibly spurious concurrency of feature salience and informativeness, and workflows using feature attribution methods as building blocks should be scrutinised.
Authors: Xiaoguang Guo, Zehong Wang, Jiazheng Li, Shawn Spitzel, Qi Yang, Kaize Ding, Jundong Li, Chuxu Zhang
Abstract: Deployed graph neural networks (GNNs) are frozen at deployment yet must fit clean data, generalize under distribution shifts, and remain stable to perturbations. We show that static inference induces a fundamental tradeoff: improving stability requires reducing reliance on shift-sensitive features, leaving an irreducible worst-case generalization floor. Instance-conditional routing can break this ceiling, but is fragile because shifts can mislead routing and perturbations can make routing fluctuate. We capture these effects via two decompositions separating coverage vs selection, and base sensitivity vs fluctuation amplification. Based on these insights, we propose STEM-GNN, a pretrain-then-finetune framework with a mixture-of-experts encoder for diverse computation paths, a vector-quantized token interface to stabilize encoder-to-head signals, and a Lipschitz-regularized head to bound output amplification. Across nine node, link, and graph benchmarks, STEM-GNN achieves a stronger three-way balance, improving robustness to degree/homophily shifts and to feature/edge corruptions while remaining competitive on clean graphs.
Authors: Benedict Clark, Stoyan Karastoyanov, Rick Wilming, Stefan Haufe
Abstract: Explainable Artificial Intelligence (XAI) aims to provide transparent insights into machine learning models, yet the reliability of many feature attribution methods remains a critical challenge. Prior research (Haufe et al., 2014; Wilming et al., 2022, 2023) has demonstrated that these methods often erroneously assign significant importance to non-informative variables, such as suppressor variables, leading to fundamental misinterpretations. Since statistical suppression is induced by feature dependencies, this study investigates whether data whitening, a common preprocessing technique for decorrelation, can mitigate such errors. Using the established XAI-TRIS benchmark (Clark et al., 2024b), which offers synthetic ground-truth data and quantitative measures of explanation correctness, we empirically evaluate 16 popular feature attribution methods applied in combination with 5 distinct whitening transforms. Additionally, we analyze a minimal linear two-dimensional classification problem (Wilming et al., 2023) to theoretically assess whether whitening can remove the impact of suppressor features from Bayes-optimal models. Our results indicate that, while specific whitening techniques can improve explanation performance, the degree of improvement varies substantially across XAI methods and model architectures. These findings highlight the complex relationship between data non-linearities, preprocessing quality, and attribution fidelity, underscoring the vital role of pre-processing techniques in enhancing model interpretability.
Authors: Michael Zuo, Inwon Kang, Stacy Patterson, Oshani Seneviratne
Abstract: We explore the privacy-utility tradeoff of synthetic data generation schemes on tabular financial datasets, a domain characterized by high regulatory risk and severe class imbalance. We consider representative tabular data generators, including autoencoders, generative adversarial networks, diffusion, and copula synthesizers. To address the challenges of the financial domain, we provide novel privacy-preserving implementations of GAN and autoencoder synthesizers. We evaluate whether and how well the generators simultaneously achieve data quality, downstream utility, and privacy, with comparison across balanced and imbalanced input datasets. Our results offer insight into the distinct challenges of generating synthetic data from datasets that exhibit severe class imbalance and mixed-type attributes.
Authors: Bret Nestor, Bohan Yao, Jasmine Moore, Jasper Kanes
Abstract: This work presents the largest curation of Southern Resident Killer Whale (SRKW) acoustic data to date, also containing other marine mammals in their environment. We systematically search all available public archival hydrophone data within the SRKW habitat (over 30 years of audio data). The search consists of a weakly-supervised, positive-unlabelled, active learning strategy to identify all instances of marine mammals. The resulting transformer-based detectors outperform state-of-the-art detectors on the DEEPAL, DCLDE-2026, and two newly introduced expert-annotated datasets in terms of accuracy, energy efficiency, and speed. The detection model has a specificity of 0-28.8% at 95% sensitivity. Our multiclass species classifier obtains a top-1 accuracy of 42.1% (11 train classes, 4 test classes) and our ecotype classifier obtains a top-1 accuracy of 43.0% (4 train classes, 5 test classes) on the DCLDE-2026 dataset. We yield 919 hours of SRKW data, 230 hours of Bigg's orca data, 1374 hours of orca data from unlabelled ecotypes, 1501 hours of humpback data, 88 hours of sea lion data, 246 hours of pacific white-sided dolphin data, and over 784 hours of unspecified marine mammal data. This SRKW dataset is larger than DCLDE-2026, Ocean Networks Canada, and OrcaSound combined. The curated species labels are available under CC-BY 4.0 license, and the corresponding audio data are available under the licenses of the original owners. The comprehensive nature of this dataset makes it suitable for unsupervised machine translation, habitat usage surveys, and conservation endeavours for this critically endangered ecotype.
Authors: Yuchong Zhang, Vardan Papyan
Abstract: Transformers leverage attention, the residual connection, and layer normalization to control the variance of token representations. We propose to modify attention into a Laplacian mechanism that gives the model more direct control over token variance. We conjecture that this helps transformers achieve the ideal token geometry. To investigate our conjecture, we first show that incorporating the Laplacian mechanism into transformers induces consistent improvements across benchmarks in computer vision and language. Next, we study how the Laplacian mechanism impacts the geometry of token representations using various tools: 1) principal component analysis, 2) cosine similarity metric, 3) analysis of variance, and 4) Neural Collapse metrics. Our investigation shows that the Laplacian mechanism reshapes token embeddings toward a geometry of maximal separability: tokens collapse according to their classes, and the class means exhibit Neural Collapse.
Authors: Sumedh Gupte, Shrey Rakeshkumar Patel, Soumen Pachal, Prashanth L. A., Sanjay P. Bhat
Abstract: We propose risk-sensitive reinforcement learning algorithms catering to three families of risk measures, namely expectiles, utility-based shortfall risk and optimized certainty equivalent risk. For each risk measure, in the context of a finite horizon Markov decision process, we first derive a policy gradient theorem. Second, we propose estimators of the risk-sensitive policy gradient for each of the aforementioned risk measures, and establish $\mathcal{O}\left(1/m\right)$ mean-squared error bounds for our estimators, where $m$ is the number of trajectories. Further, under standard assumptions for policy gradient-type algorithms, we establish smoothness of the risk-sensitive objective, in turn leading to stationary convergence rate bounds for the overall risk-sensitive policy gradient algorithm that we propose. Finally, we conduct numerical experiments to validate the theoretical findings on popular RL benchmarks.
Authors: Che-Chia Chang, Chen-Yang Dai, Te-Sheng Lin, Ming-Chih Lai, Chieh-Hsin Lai
Abstract: We propose a physics-informed consistency modeling framework for solving partial differential equations (PDEs) via fast, few-step generative inference. We identify a key stability challenge in physics-constrained consistency training, where PDE residuals can drive the model toward trivial or degenerate solutions, degrading the learned data distribution. To address this, we introduce a structure-preserving two-stage training strategy that decouples distribution learning from physics enforcement by freezing the coefficient decoder during physics-informed fine-tuning. We further propose a two-step residual objective that enforces physical consistency on refined, structurally valid generative trajectories rather than noisy single-step predictions. The resulting framework enables stable, high-fidelity inference for both unconditional generation and forward problems. We demonstrate that forward solutions can be obtained via a projection-based zero-shot inpainting procedure, achieving consistent accuracy of diffusion baselines with orders of magnitude reduction in computational cost.
Authors: Mohammad Partohaghighi, Roummel Marcia, Bruce J. West, YangQuan Chen
Abstract: Machine unlearning aims to remove the influence of a designated forget set from a trained model while preserving utility on the retained data. In modern deep networks, approximate unlearning frequently fails under large or adversarial deletions due to pronounced layer-wise heterogeneity: some layers exhibit stable, well-regularized representations while others are brittle, undertrained, or overfit, so naive update allocation can trigger catastrophic forgetting or unstable dynamics. We propose Statistical-Roughness Adaptive Gradient Unlearning (SRAGU), a mechanism-first unlearning algorithm that reallocates unlearning updates using layer-wise statistical roughness operationalized via heavy-tailed spectral diagnostics of layer weight matrices. Starting from an Adaptive Gradient Unlearning (AGU) sensitivity signal computed on the forget set, SRAGU estimates a WeightWatcher-style heavy-tailed exponent for each layer, maps it to a bounded spectral stability weight, and uses this stability signal to spectrally reweight the AGU sensitivities before applying the same minibatch update form. This concentrates unlearning motion in spectrally stable layers while damping updates in unstable or overfit layers, improving stability under hard deletions. We evaluate unlearning via behavioral alignment to a gold retrained reference model trained from scratch on the retained data, using empirical prediction-divergence and KL-to-gold proxies on a forget-focused query set; we additionally report membership inference auditing as a complementary leakage signal, treating forget-set points as should-be-forgotten members during evaluation.
Authors: Pei-Chi Pan, Yingbin Liang, Sen Lin
Abstract: Large Language Models (LLMs) demonstrate transformative potential, yet their reasoning remains inconsistent and unreliable. Reinforcement learning (RL)-based fine-tuning is a key mechanism for improvement, but its effectiveness is fundamentally governed by reward design. Despite its importance, the relationship between reward modeling and core LLM challenges--such as evaluation bias, hallucination, distribution shift, and efficient learning--remains poorly understood. This work argues that reward modeling is not merely an implementation detail but a central architect of reasoning alignment, shaping what models learn, how they generalize, and whether their outputs can be trusted. We introduce Reasoning-Aligned Reinforcement Learning (RARL), a unifying framework that systematizes diverse reward paradigms for multi-step reasoning. Within this framework, we present a taxonomy of reward mechanisms, analyze reward hacking as a pervasive failure mode, and examine how reward signals unify challenges ranging from inference-time scaling to hallucination mitigation. We further critically evaluate existing benchmarks, highlighting vulnerabilities such as data contamination and reward misalignment, and outline directions for more robust evaluation. By integrating fragmented research threads and clarifying the interplay between reward design and fundamental reasoning capabilities, this work provides a foundational roadmap for building reasoning models that are robust, verifiable, and trustworthy.
Authors: Thi Minh Chau Nguyen, Minh Hieu Nguyen, Duc Anh Nguyen, Xuan Huong Tran, Thanh Trung Huynh, Quoc Viet Hung Nguyen
Abstract: Federated sequential recommendation (FedSeqRec) aims to perform next-item prediction while keeping user data decentralised, yet model quality is frequently constrained by fragmented, noisy, and homogeneous interaction logs stored on individual devices. Many existing approaches attempt to compensate through manual data augmentation or additional server-side constraints, but these strategies either introduce limited semantic diversity or increase system overhead. To overcome these challenges, we propose \textbf{LUMOS}, a parameter-isolated FedSeqRec architecture that integrates large language models (LLMs) as \emph{local semantic generators}. Instead of sharing gradients or auxiliary parameters, LUMOS privately invokes an on-device LLM to construct three complementary sequence variants from each user history: (i) \emph{future-oriented} trajectories that infer plausible behavioural continuations, (ii) \emph{semantically equivalent rephrasings} that retain user intent while diversifying interaction patterns, and (iii) \emph{preference-inconsistent counterfactuals} that serve as informative negatives. These synthesized sequences are jointly encoded within the federated backbone through a tri-view contrastive optimisation scheme, enabling richer representation learning without exposing sensitive information. Experimental results across three public benchmarks show that LUMOS achieves consistent gains over competitive centralised and federated baselines on HR@20 and NDCG@20. In addition, the use of semantically grounded positive signals and counterfactual negatives improves robustness under noisy and adversarial environments, even without dedicated server-side protection modules. Overall, this work demonstrates the potential of LLM-driven semantic generation as a new paradigm for advancing privacy-preserving federated recommendation.
Authors: Runa Eschenhagen, Anna Cai, Tsung-Hsien Lee, Hao-Jun Michael Shi
Abstract: Optimizers leveraging the matrix structure in neural networks, such as Shampoo and Muon, are more data-efficient than element-wise algorithms like Adam and Signum. While in specific settings, Shampoo and Muon reduce to spectral descent analogous to how Adam and Signum reduce to sign descent, their general relationship and relative data efficiency under controlled settings remain unclear. Through extensive experiments on language models, we demonstrate that Shampoo achieves higher token efficiency than Muon, mirroring Adam's advantage over Signum. We show that Shampoo's update applied to weight matrices can be decomposed into an adapted Muon update. Consistent with this, Shampoo's benefits can be exclusively attributed to its application to weight matrices, challenging interpretations agnostic to parameter shapes. This admits a new perspective that also avoids shortcomings of related interpretations based on variance adaptation and whitening: rather than enforcing semi-orthogonality as in spectral descent, Shampoo's updates are time-averaged semi-orthogonal in expectation.
Authors: Zhendong Mi, Yixiao Chen, Pu Zhao, Xiaodong Yu, Hao Wang, Yanzhi Wang, Shaoyi Huang
Abstract: Mixture-of-Experts (MoE) based Large Language Models (LLMs) have achieved superior performance, yet the massive memory overhead caused by storing multiple expert networks severely hinders their practical deployment. Singular Value Decomposition (SVD)-based compression has emerged as a promising post-training technique; however, most existing methods apply uniform rank allocation or rely solely on static weight properties. This overlooks the substantial heterogeneity in expert utilization observed in MoE models, where frequent routing patterns and intrinsic information density vary significantly across experts. In this work, we propose RFID-MoE, an effective framework for MoE compression by exploiting heterogeneous Routing Frequency and Information Density. We first introduce a fused metric that combines expert activation frequency with effective rank to measure expert importance, adaptively allocating higher ranks to critical expert groups under a fixed budget. Moreover, instead of discarding compression residuals, we reconstruct them via a parameter-efficient sparse projection mechanism to recover lost information with minimal parameter overhead. Extensive experiments on representative MoE LLMs (e.g., Qwen3, DeepSeekMoE) across multiple compression ratios demonstrate that RFID-MoE consistently outperforms state-of-the-art methods like MoBE and D2-MoE. Notably, RFID-MoE achieves a perplexity of 16.92 on PTB with the Qwen3-30B model at a 60% compression ratio, reducing perplexity by over 8.0 compared to baselines, and improves zero-shot accuracy on HellaSwag by approximately 8%.
Authors: Ya-Chi Chu, Alkiviades Boukas, Madeleine Udell
Abstract: Neural networks are increasingly used as surrogate solvers and control policies, but unconstrained predictions can violate physical, operational, or safety requirements. We propose SnareNet, a feasibility-controlled architecture for learning mappings whose outputs must satisfy input-dependent nonlinear constraints. SnareNet appends a differentiable repair layer that navigates in the constraint map's range space, steering iterates toward feasibility and producing a repaired output that satisfies constraints to a user-specified tolerance. To stabilize end-to-end training, we introduce adaptive relaxation, which designs a relaxed feasible set that snares the neural network at initialization and shrinks it into the feasible set, enabling early exploration and strict feasibility later in training. On optimization-learning and trajectory planning benchmarks, SnareNet consistently attains improved objective quality while satisfying constraints more reliably than prior work.
Authors: Kiljae Lee, Ziqi Liu, Weijing Tang, Yuan Zhang
Abstract: Shapley values are widely used for model-agnostic data valuation and feature attribution, yet they implicitly assume contributors are interchangeable. This can be problematic when contributors are dependent (e.g., reused/augmented data or causal feature orderings) or when contributions should be adjusted by factors such as trust or risk. We propose Priority-Aware Shapley Value (PASV), which incorporates both hard precedence constraints and soft, contributor-specific priority weights. PASV is applicable to general precedence structures, recovers precedence-only and weight-only Shapley variants as special cases, and is uniquely characterized by natural axioms. We develop an efficient adjacent-swap Metropolis-Hastings sampler for scalable Monte Carlo estimation and analyze limiting regimes induced by extreme priority weights. Experiments on data valuation (MNIST/CIFAR10) and feature attribution (Census Income) demonstrate more structure-faithful allocations and a practical sensitivity analysis via our proposed "priority sweeping".
Authors: Jiaming Liu, Cheng Ding, Daoqiang Zhang
Abstract: The absence of pre-hospital physiological data in standard clinical datasets fundamentally constrains the early prediction of stroke, as patients typically present only after stroke has occurred, leaving the predictive value of continuous monitoring signals such as photoplethysmography (PPG) unvalidated. In this work, we overcome this limitation by focusing on a rare but clinically critical cohort - patients who suffered stroke during hospitalization while already under continuous monitoring - thereby enabling the first large-scale analysis of pre-stroke PPG waveforms aligned to verified onset times. Using MIMIC-III and MC-MED, we develop an LLM-assisted data mining pipeline to extract precise in-hospital stroke onset timestamps from unstructured clinical notes, followed by physician validation, identifying 176 patients (MIMIC) and 158 patients (MC-MED) with high-quality synchronized pre-onset PPG data, respectively. We then extract hemodynamic features from PPG and employ a ResNet-1D model to predict impending stroke across multiple early-warning horizons. The model achieves F1-scores of 0.7956, 0.8759, and 0.9406 at 4, 5, and 6 hours prior to onset on MIMIC-III, and, without re-tuning, reaches 0.9256, 0.9595, and 0.9888 on MC-MED for the same horizons. These results provide the first empirical evidence from real-world clinical data that PPG contains predictive signatures of stroke several hours before onset, demonstrating that passively acquired physiological signals can support reliable early warning, supporting a shift from post-event stroke recognition to proactive, physiology-based surveillance that may materially improve patient outcomes in routine clinical care.
Authors: Xueying Ding, Simon Kl\"uttermann, Haomin Wen, Yilong Chen, Leman Akoglu
Abstract: Quality benchmarks are essential for fairly and accurately tracking scientific progress and enabling practitioners to make informed methodological choices. Outlier detection (OD) on tabular data underpins numerous real-world applications, yet existing OD benchmarks remain limited. The prominent OD benchmark AdBench is the de facto standard in the literature, yet comprises only 57 datasets. In addition to other shortcomings discussed in this work, its small scale severely restricts diversity and statistical power. We introduce MacrOData, a large-scale benchmark suite for tabular OD comprising three carefully curated components: OddBench, with 790 datasets containing real-world semantic anomalies; OvrBench, with 856 datasets featuring real-world statistical outliers; and SynBench, with 800 synthetically generated datasets spanning diverse data priors and outlier archetypes. Owing to its scale and diversity, MacrOData enables comprehensive and statistically robust evaluation of tabular OD methods. Our benchmarks further satisfy several key desiderata: We provide standardized train/test splits for all datasets, public/private benchmark partitions with held-out test labels for the latter reserved toward an online leaderboard, and annotate our datasets with semantic metadata. We conduct extensive experiments across all benchmarks, evaluating a broad range of OD methods comprising classical, deep, and foundation models, over diverse hyperparameter configurations. We report detailed empirical findings, practical guidelines, as well as individual performances as references for future research. All benchmarks containing 2,446 datasets combined are open-sourced, along with a publicly accessible leaderboard hosted at https://huggingface.co/MacrOData-CMU.
Authors: Nguyen Thach, Xingchen Sha, Hau Chan
Abstract: Participatory budgeting (PB) is a democratic paradigm for deciding the funding of public projects given the residents' preferences, which has been adopted in numerous cities across the world. The main focus of PB is designing rules, functions that return feasible budget allocations for a set of projects subject to some budget constraint. Designing PB rules that optimize both utility and fairness objectives based on agent preferences had been challenging due to the extensive domain knowledge required and the proven trade-off between the two notions. Recently, large language models (LLMs) have been increasingly employed for automated algorithmic design. Given the resemblance of PB rules to algorithms for classical knapsack problems, in this paper, we introduce a novel framework, named LLMRule, that addresses the limitations of existing works by incorporating LLMs into an evolutionary search procedure for automating the design of PB rules. Our experimental results, evaluated on more than 600 real-world PB instances obtained from the U.S., Canada, Poland, and the Netherlands with different representations of agent preferences, demonstrate that the LLM-generated rules generally outperform existing handcrafted rules in terms of overall utility while still maintaining a similar degree of fairness.
Authors: Hanchen Xia, Baoyou Chen, Zelin Zang, Yutang Ge, Guojiang Zhao, Siyu Zhu
Abstract: We propose LaPha, a method for training AlphaZero-like LLM agents in a Poincar\'e latent space. Under LaPha, the search process can be visualized as a tree rooted at the prompt and growing outward from the origin toward the boundary of the Poincar\'e ball, where negative curvature provides exponentially increasing capacity with radius. Using hyperbolic geodesic distance to rule-verified correctness, we define a node potential and assign dense process rewards by potential differences. We further attach a lightweight value head on the same shared latent space, enabling self-guided test-time scaling with almost no additional overhead. On MATH-500, LaPha improves Qwen2.5-Math-1.5B from 66.0% to 88.2%. With value-head-guided search, LaPha-1.5B reaches 56.7% accuracy on AIME'24, and LaPha-7B further achieves 60.0% on AIME'24 and 53.3% on AIME'25.
Authors: Yifei Cheng, Xianglin Yang, Guoxia Wang, Chao Huang, Fei Ma, Dianhai Yu, Xiaochun Cao, Li Shen
Abstract: Sharpness-aware minimization (SAM) seeks the minima with a flat loss landscape to improve the generalization performance in machine learning tasks, including fine-tuning. However, its extra parameter perturbation step doubles the computation cost, which becomes the bottleneck of SAM in the practical implementation. In this work, we propose an approach SL-SAM to break this bottleneck by introducing the sparse technique to layers. Our key innovation is to frame the dynamic selection of layers for both the gradient ascent (perturbation) and descent (update) steps as a multi-armed bandit problem. At the beginning of each iteration, SL-SAM samples a part of the layers of the model according to the gradient norm to participate in the backpropagation of the following parameter perturbation and update steps, thereby reducing the computation complexity. We then provide the analysis to guarantee the convergence of SL-SAM. In the experiments of fine-tuning models in several tasks, SL-SAM achieves the performances comparable to the state-of-the-art baselines, including a \#1 rank on LLM fine-tuning. Meanwhile, SL-SAM significantly reduces the ratio of active parameters in backpropagation compared to vanilla SAM (SL-SAM activates 47\%, 22\% and 21\% parameters on the vision, moderate and large language model respectively while vanilla SAM always activates 100\%), verifying the efficiency of our proposed algorithm.
Authors: Nilaksh, Antoine Clavaud, Mathieu Reymond, Fran\c{c}ois Rivest, Sarath Chandar
Abstract: In streaming Reinforcement Learning (RL), transitions are observed and discarded immediately after a single update. While this minimizes resource usage for on-device applications, it makes agents notoriously sample-inefficient, since value-based losses alone struggle to extract meaningful representations from transient data. We propose extending Self-Predictive Representations (SPR) to the streaming pipeline to maximize the utility of every observed frame. However, due to the highly correlated samples induced by the streaming regime, naively applying this auxiliary loss results in training instabilities. Thus, we introduce orthogonal gradient updates relative to the momentum target and resolve gradient conflicts arising from streaming-specific optimizers. Validated across the Atari, MinAtar, and Octax suites, our approach systematically outperforms existing streaming baselines. Latent-space analysis, including t-SNE visualizations and effective-rank measurements, confirms that our method learns significantly richer representations, bridging the performance gap caused by the absence of a replay buffer, while remaining efficient enough to train on just a few CPU cores.
Authors: Alireza F. Pour, Farnam Mansouri, Shai Ben-David
Abstract: We study an online learning problem with multiple correct answers, where each instance admits a set of valid labels, and in each round the learner must output a valid label for the queried example. This setting is motivated by language generation tasks, in which a prompt may admit many acceptable completions, but not every completion is acceptable. We study this problem under three feedback models. For each model, we characterize the optimal mistake bound in the realizable setting using an appropriate combinatorial dimension. We then establish a trichotomy of regret bounds across the three models in the agnostic setting. Our results also imply sample complexity bounds for the batch setup that depend on the respective combinatorial dimensions.
Authors: Prin Phunyaphibarn, Minhyuk Sung
Abstract: Discrete diffusion models have recently emerged as a powerful class of generative models for chemistry and biology data. In these fields, the goal is to generate various samples with high rewards (e.g., drug-likeness in molecules), making reward-based guidance crucial. Most existing methods are based on guiding the diffusion model using intermediate rewards but tend to underperform since intermediate rewards are noisy due to the non-smooth nature of reward functions used in scientific domains. To address this, we propose Clean-Sample Markov Chain (CSMC) Sampler, a method that performs effective test-time reward-guided sampling for discrete diffusion models, enabling local search without relying on intermediate rewards. CSMC constructs a Markov chain of clean samples using the Metropolis-Hastings algorithm such that its stationary distribution is the target distribution. We design a proposal distribution by sequentially applying the forward and backward diffusion processes, making the acceptance probability tractable. Experiments on molecule and biological sequence generation with various reward functions demonstrate that our method consistently outperforms prior approaches that rely on intermediate rewards.
Authors: Xinxu Wei, Rong Zhou, Lifang He, Yu Zhang
Abstract: With the growing interest in foundation models for brain signals, graph-based pretraining has emerged as a promising paradigm for learning transferable representations from connectome data. However, existing contrastive and masked autoencoder methods typically rely on naive random dropping or masking for augmentation, which is ill-suited for brain graphs and hypergraphs as it disrupts semantically meaningful connectivity patterns. Moreover, commonly used graph-level readout and reconstruction schemes fail to capture global structural information, limiting the robustness of learned representations. In this work, we propose a unified diffusion-based pretraining framework that addresses both limitations. First, diffusion is designed to guide structure-aware dropping and masking strategies, preserving brain graph semantics while maintaining effective pretraining diversity. Second, diffusion enables topology-aware graph-level readout and node-level global reconstruction by allowing graph embeddings and masked nodes to aggregate information from globally related regions. Extensive experiments across multiple neuroimaging datasets with over 25,000 subjects and 60,000 scans involving various mental disorders and brain atlases demonstrate consistent performance improvements.
Authors: Hao Qin, Chicheng Zhang
Abstract: We propose an algorithmic framework, Offline Estimation to Decisions (OE2D), that reduces contextual bandit learning with general reward function approximation to offline regression. The framework allows near-optimal regret for contextual bandits with large action spaces with $O(log(T))$ calls to an offline regression oracle over $T$ rounds, and makes $O(loglog(T))$ calls when $T$ is known. The design of OE2D algorithm generalizes Falcon~\citep{simchi2022bypassing} and its linear reward version~\citep[][Section 4]{xu2020upper} in that it chooses an action distribution that we term ``exploitative F-design'' that simultaneously guarantees low regret and good coverage that trades off exploration and exploitation. Central to our regret analysis is a new complexity measure, the Decision-Offline Estimation Coefficient (DOEC), which we show is bounded in bounded Eluder dimension per-context and smoothed regret settings. We also establish a relationship between DOEC and Decision Estimation Coefficient (DEC)~\citep{foster2021statistical}, bridging the design principles of offline- and online-oracle efficient contextual bandit algorithms for the first time.
Authors: Lihao Mai, Chenhan Xiao, Yang Weng
Abstract: Increasing penetration of inverter-based resources, flexible loads, and rapidly changing operating conditions make higher-order $N\!-\!k$ contingency assessment increasingly important but computationally prohibitive. Exhaustive evaluation of all outage combinations using AC power-flow or ACOPF is infeasible in routine operation. This fact forces operators to rely on heuristic screening methods whose ability to consistently retain all critical contingencies is not formally established. This paper proposes a scalable, state-aware contingency inference framework designed to directly generate high-impact $N\!-\!k$ outage scenarios without enumerating the combinatorial contingency space. The framework employs a conditional diffusion model to produce candidate contingencies tailored to the current operating state, while a topology-aware graph neural network trained only on base and $N\!-\!1$ cases efficiently constructs high-risk training samples offline. Finally, the framework is developed to provide controllable coverage guarantees for severe contingencies, allowing operators to explicitly manage the risk of missing critical events under limited AC power-flow evaluation budgets. Experiments on IEEE benchmark systems show that, for a given evaluation budget, the proposed approach consistently evaluates higher-severity contingencies than uniform sampling. This allows critical outages to be identified more reliably with reduced computational effort.
Authors: Ofir Schlisselberg, Tal Lancewicki, Yishay Mansour
Abstract: We study reinforcement learning in MDPs whose transition function is stochastic at most steps but may behave adversarially at a fixed subset of $\Lambda$ steps per episode. This model captures environments that are stable except at a few vulnerable points. We introduce \emph{conditioned occupancy measures}, which remain stable across episodes even with adversarial transitions, and use them to design two algorithms. The first handles arbitrary adversarial steps and achieves regret $\tilde{O}(H S^{\Lambda}\sqrt{K S A^{\Lambda+1}})$, where $K$ is the number of episodes, $S$ is the number of state, $A$ is the number of actions and $H$ is the episode's horizon. The second, assuming the adversarial steps are consecutive, improves the dependence on $S$ to $\tilde{O}(H\sqrt{K S^{3} A^{\Lambda+1}})$. We further give a $K^{2/3}$-regret reduction that removes the need to know which steps are the $\Lambda$ adversarial steps. We also characterize the regret of adversarial MDPs in the \emph{fully adversarial} setting ($\Lambda=H-1$) both for full-information and bandit feedback, and provide almost matching upper and lower bounds (slightly strengthen existing lower bounds, and clarify how different feedback structures affect the hardness of learning).
Authors: Huseyin Tunc
Abstract: Reaction-diffusion (RD) equations underpin pattern formation across chemistry, biology, and physics, yet learning stable operators that forecast their long-term dynamics from data remains challenging. Neural-operator surrogates provide resolution-robust prediction, but autoregressive rollouts can drift due to the accumulation of error, and out-of-distribution (OOD) initial conditions often degrade accuracy. Physics-based numerical residual objectives can regularize operator learning, although they introduce additional assumptions, sensitivity to discretization and loss design, and higher training cost. Here we develop a purely data-driven operator learner with adaptive recurrent training (DDOL-ART) using a robust recurrent strategy with lightweight validation milestones that early-exit unproductive rollout segments and redirect optimization. Trained only on a single in-distribution toroidal Gaussian family over short horizons, DDOL-ART learns one-step operators that remain stable under long rollouts and generalize zero-shot to strong morphology shifts across FitzHugh-Nagumo (FN), Gray-Scott (GS), and Lambda-Omega (LO) systems. Across these benchmarks, DDOL-ART delivers a strong accuracy and cost trade-off. It is several-fold faster than a physics-based numerical-loss operator learner (NLOL) under matched settings, and it remains competitive on both in-distribution stability and OOD robustness. Training-dynamics diagnostics show that adaptivity strengthens the correlation between validation error and OOD test error performance, acting as a feedback controller that limits optimization drift. Our results indicate that feedback-controlled recurrent training of DDOL-ART generates robust flow-map surrogates without PDE residuals, while simultaneously maintaining competitiveness with NLOL at significantly reduced training costs.
Authors: Sangyoon Lee, Jaeho Lee
Abstract: Low-rank adaptation (LoRA) is a standard approach for fine-tuning large language models, yet its many variants report conflicting empirical gains, often on the same benchmarks. We show that these contradictions arise from a single overlooked factor: the batch size. When properly tuned, vanilla LoRA often matches the performance of more complex variants. We further propose a proxy-based, cost-efficient strategy for batch size tuning, revealing the impact of rank, dataset size, and model capacity on the optimal batch size. Our findings elevate batch size from a minor implementation detail to a first-order design parameter, reconciling prior inconsistencies and enabling more reliable evaluations of LoRA variants.
Authors: Moshe Noivirt, Jessica Sorrell, Eliad Tsfadia
Abstract: We study the computational relationship between replicability (Impagliazzo et al. [STOC `22], Ghazi et al. [NeurIPS `21]) and other stability notions. Specifically, we focus on replicable PAC learning and its connections to differential privacy (Dwork et al. [TCC 2006]) and to the statistical query (SQ) model (Kearns [JACM `98]). Statistically, it was known that differentially private learning and replicable learning are equivalent and strictly more powerful than SQ-learning. Yet, computationally, all previously known efficient (i.e., polynomial-time) replicable learning algorithms were confined to SQ-learnable tasks or restricted distributions, in contrast to differentially private learning. Our main contribution is the first computationally efficient replicable algorithm for realizable learning of parities over arbitrary distributions, a task that is known to be hard in the SQ-model, but possible under differential privacy. This result provides the first evidence that efficient replicable learning over general distributions strictly extends efficient SQ-learning, and is closer in power to efficient differentially private learning, despite computational separations between replicability and privacy. Our main building block is a new, efficient, and replicable algorithm that, given a set of vectors, outputs a subspace of their linear span that covers most of them.
Authors: Yuanyu Wan, Yu Shen, Dingzhi Yu, Bo Xue, Mingli Song
Abstract: To expand the applicability of decentralized online learning, previous studies have proposed several algorithms for decentralized online continuous submodular maximization (D-OCSM) -- a non-convex/non-concave setting with continuous DR-submodular reward functions. However, there exist large gaps between their approximate regret bounds and the regret bounds achieved in the convex setting. Moreover, if focusing on projection-free algorithms, which can efficiently handle complex decision sets, they cannot even recover the approximate regret bounds achieved in the centralized setting. In this paper, we first demonstrate that for D-OCSM over general convex decision sets, these two issues can be addressed simultaneously. Furthermore, for D-OCSM over downward-closed decision sets, we show that the second issue can be addressed while significantly alleviating the first issue. Our key techniques are two reductions from D-OCSM to decentralized online convex optimization (D-OCO), which can exploit D-OCO algorithms to improve the approximate regret of D-OCSM in these two cases, respectively.
Authors: Wenzhe Yin, Pan Zhou, Zehao Xiao, Jie Liu, Shujian Yu, Jan-Jakob Sonke, Efstratios Gavves
Abstract: Multimodal representation learning aims to construct a shared embedding space in which heterogeneous modalities are semantically aligned. Despite strong empirical results, InfoNCE-based objectives introduce inherent conflicts that yield distribution gaps across modalities. In this work, we identify two conflicts in the multimodal regime, both exacerbated as the number of modalities increases: (i) an alignment-uniformity conflict, whereby the repulsion of uniformity undermines pairwise alignment, and (ii) an intra-alignment conflict, where aligning multiple modalities induces competing alignment directions. To address these issues, we propose a principled decoupling of alignment and uniformity for multimodal representations, providing a conflict-free recipe for multimodal learning that simultaneously supports discriminative and generative use cases without task-specific modules. We then provide a theoretical guarantee that our method acts as an efficient proxy for a global H\"older divergence over multiple modality distributions, and thus reduces the distribution gap among modalities. Extensive experiments on retrieval and UnCLIP-style generation demonstrate consistent gains.
Authors: Yiyun Zhou, Jingwei Shi, Mingjing Xu, Zhonghua Jiang, Jingyuan Chen
Abstract: Knowledge Distillation (KD) has emerged as a powerful technique for model compression, enabling lightweight student networks to benefit from the performance of redundant teacher networks. However, the inherent capacity gap often limits the performance of student networks. Inspired by the expressiveness of pretrained teacher networks, a compelling research question arises: is there a type of network that can not only inherit the teacher's structure but also maximize the inheritance of its knowledge? Furthermore, how does the performance of such an inheriting network compare to that of student networks, all benefiting from the same teacher network? To further explore this question, we propose InherNet, a neural network inheritance method that performs asymmetric low-rank decomposition on the teacher's weights and reconstructs a lightweight yet expressive network without significant architectural disruption. By leveraging Singular Value Decomposition (SVD) for initialization to ensure the inheritance of principal knowledge, InherNet effectively balances depth, width, and compression efficiency. Experimental results across unimodal and multimodal tasks demonstrate that InherNet achieves higher performance compared to student networks of similar parameter sizes. Our findings reveal a promising direction for future research in efficient model compression beyond traditional distillation.
Authors: Xenia Heilmann, Luca Corbucci, Mattia Cerrato
Abstract: The Rashomon set captures the collection of models that achieve near-identical empirical performance yet may differ substantially in their decision boundaries. Understanding the differences among these models, i.e., their multiplicity, is recognized as a crucial step toward model transparency, fairness, and robustness, as it reveals decision boundaries instabilities that standard metrics obscure. However, the existing definitions of Rashomon set and multiplicity metrics assume centralized learning and do not extend naturally to decentralized, multi-party settings like Federated Learning (FL). In FL, multiple clients collaboratively train models under a central server's coordination without sharing raw data, which preserves privacy but introduces challenges from heterogeneous client data distribution and communication constraints. In this setting, the choice of a single best model may homogenize predictive behavior across diverse clients, amplify biases, or undermine fairness guarantees. In this work, we provide the first formalization of Rashomon sets in FL.First, we adapt the Rashomon set definition to FL, distinguishing among three perspectives: (I) a global Rashomon set defined over aggregated statistics across all clients, (II) a t-agreement Rashomon set representing the intersection of local Rashomon sets across a fraction t of clients, and (III) individual Rashomon sets specific to each client's local distribution.Second, we show how standard multiplicity metrics can be estimated under FL's privacy constraints. Finally, we introduce a multiplicity-aware FL pipeline and conduct an empirical study on standard FL benchmark datasets. Our results demonstrate that all three proposed federated Rashomon set definitions offer valuable insights, enabling clients to deploy models that better align with their local data, fairness considerations, and practical requirements.
Authors: Zihang Liu, Oleg Balabanov, Yaoqing Yang, Michael W. Mahoney
Abstract: We introduce AutoSpec, a neural network framework for discovering iterative spectral algorithms for large-scale numerical linear algebra and numerical optimization. Our self-supervised models adapt to input operators using coarse spectral information (e.g., eigenvalue estimates and residual norms), and they predict recurrence coefficients for computing or applying a matrix polynomial tailored to a downstream task. The effectiveness of AutoSpec relies on three ingredients: an architecture whose inference pass implements short, executable numerical linear algebra recurrences; efficient training on small synthetic problems with transfer to large-scale real-world operators; and task-defined objectives that enforce the desired approximation or preconditioning behavior across the range of spectral profiles represented in the training set. We apply AutoSpec to discovering algorithms for representative numerical linear algebra tasks: accelerating matrix-function approximation; accelerating sparse linear solvers; and spectral filtering/preconditioning for eigenvalue computations. On real-world matrices, the learned procedures deliver orders-of-magnitude improvements in accuracy and/or reductions in iteration count, relative to basic baselines. We also find clear connections to classical theory: the induced polynomials often exhibit near-equiripple, near-minimax behavior characteristic of Chebyshev polynomials.
Authors: Vajira Thambawita, Jonas L. Isaksen, J{\o}rgen K. Kanters, Hugo L. Hammer, P{\aa}l Halvorsen
Abstract: Deep learning has achieved expert-level performance in automated electrocardiogram (ECG) diagnosis, yet the "black-box" nature of these models hinders their clinical deployment. Trust in medical AI requires not just high accuracy but also transparency regarding the specific physiological features driving predictions. Existing explainability methods for ECGs typically rely on post-hoc approximations (e.g., Grad-CAM and SHAP), which can be unstable, computationally expensive, and unfaithful to the model's actual decision-making process. In this work, we propose the ECG-IMN, an Interpretable Mesomorphic Neural Network tailored for high-resolution 12-lead ECG classification. Unlike standard classifiers, the ECG-IMN functions as a hypernetwork: a deep convolutional backbone generates the parameters of a strictly linear model specific to each input sample. This architecture enforces intrinsic interpretability, as the decision logic is mathematically transparent and the generated weights (W) serve as exact, high-resolution feature attribution maps. We introduce a transition decoder that effectively maps latent features to sample-wise weights, enabling precise localization of pathological evidence (e.g., ST-elevation, T-wave inversion) in both time and lead dimensions. We evaluate our approach on the PTB-XL dataset for classification tasks, demonstrating that the ECG-IMN achieves competitive predictive performance (AUROC comparable to black-box baselines) while providing faithful, instance-specific explanations. By explicitly decoupling parameter generation from prediction execution, our framework bridges the gap between deep learning capability and clinical trustworthiness, offering a principled path toward "white-box" cardiac diagnostics.
Authors: Hao Wang, Ziao Wang, Xiangpeng Liang, Han Zhao, Jianqi Hu, Junjie Jiang, Xing Fu, Jianshi Tang, Huaqiang Wu, Sylvain Gigan, Qiang Liu
Abstract: Deep learning has revolutionized modern society but faces growing energy and latency constraints. Deep physical neural networks (PNNs) are interconnected computing systems that directly exploit analog dynamics for energy-efficient, ultrafast AI execution. Realizing this potential, however, requires universal training methods tailored to physical intricacies. Here, we present the Physical Information Bottleneck (PIB), a general and efficient framework that integrates information theory and local learning, enabling deep PNNs to learn under arbitrary physical dynamics. By allocating matrix-based information bottlenecks to each unit, we demonstrate supervised, unsupervised, and reinforcement learning across electronic memristive chips and optical computing platforms. PIB also adapts to severe hardware faults and allows for parallel training via geographically distributed resources. Bypassing auxiliary digital models and contrastive measurements, PIB recasts PNN training as an intrinsic, scalable information-theoretic process compatible with diverse physical substrates.
Authors: Zhida Jiang, Zhaolong Xing, Jiawei Lu, Yipei Niu, Qingyuan Sang, Liangxu Zhang, Wenquan Dai, Junhua Shu, Jiaxing Wang, Qiangyu Pei, Qiong Chen, Xinyu Liu, Fangming Liu, Ai Han, Zhen Chen, Ke Zhang
Abstract: Despite algorithm-level innovations for multi-agent reinforcement learning (MARL), the underlying networked infrastructure for large-scale MARL training remains underexplored. Existing training frameworks primarily optimize for single-agent scenarios and fail to address the unique system-level challenges of MARL, including rollout-training synchronization barriers, rollout load imbalance, and training resource underutilization. To bridge this gap, we propose FlexMARL, the first end-to-end training framework that holistically optimizes rollout, training, and their orchestration for large-scale LLM-based MARL. Specifically, FlexMARL introduces the joint orchestrator to manage data flow under the rollout-training disaggregated architecture. Building upon the experience store, a novel micro-batch driven asynchronous pipeline eliminates the synchronization barriers while providing strong consistency guarantees. Rollout engine adopts a parallel sampling scheme combined with hierarchical load balancing, which adapts to skewed inter/intra-agent request patterns. Training engine achieves on-demand hardware binding through agent-centric resource allocation. The training states of different agents are swapped via unified and location-agnostic communication. Empirical results on a large-scale production cluster demonstrate that FlexMARL achieves up to 7.3x speedup and improves hardware utilization by up to 5.6x compared to existing frameworks.
Authors: Donghwan Kim, Hyunsoo Yoon
Abstract: Deep generative models that can tractably compute input likelihoods, including normalizing flows, often assign unexpectedly high likelihoods to out-of-distribution (OOD) inputs. We mitigate this likelihood paradox by manipulating input entropy based on semantic similarity, applying stronger perturbations to inputs that are less similar to an in-distribution memory bank. We provide a theoretical analysis showing that entropy control increases the expected log-likelihood gap between in-distribution and OOD samples in favor of the in-distribution, and we explain why the procedure works without any additional training of the density model. We then evaluate our method against likelihood-based OOD detectors on standard benchmarks and find consistent AUROC improvements over baselines, supporting our explanation.
Authors: Donghwan Kim, Junghun Phee, Hyunsoo Yoon
Abstract: Deep generative models with tractable and analytically computable likelihoods, exemplified by normalizing flows, offer an effective basis for anomaly detection through likelihood-based scoring. We demonstrate that, unlike in the image domain where deep generative models frequently assign higher likelihoods to anomalous data, such counterintuitive behavior occurs far less often in tabular settings. We first introduce a domain-agnostic formulation that enables consistent detection and evaluation of the counterintuitive phenomenon, addressing the absence of precise definition. Through extensive experiments on 47 tabular datasets and 10 CV/NLP embedding datasets in ADBench, benchmarked against 13 baseline models, we demonstrate that the phenomenon, as defined, is consistently rare in general tabular data. We further investigate this phenomenon from both theoretical and empirical perspectives, focusing on the roles of data dimensionality and difference in feature correlation. Our results suggest that likelihood-only detection with normalizing flows offers a practical and reliable approach for anomaly detection in tabular domains.
Authors: Naveen Gill, Ajvad Haneef K, Madhu Kumar S D
Abstract: Feature selection (FS) remains essential for building accurate and interpretable detection models, particularly in high-dimensional malware datasets. Conventional FS methods such as Extra Trees, Variance Threshold, Tree-based models, Chi-Squared tests, ANOVA, Random Selection, and Sequential Attention rely primarily on statistical heuristics or model-driven importance scores, often overlooking the semantic context of features. Motivated by recent progress in LLM-driven FS, we investigate whether large language models (LLMs) can guide feature selection in a zero-shot setting, using only feature names and task descriptions, as a viable alternative to traditional approaches. We evaluate multiple LLMs (GPT-5.0, GPT-4.0, Gemini-2.5 etc.) on the EMBOD dataset (a fusion of EMBER and BODMAS benchmark datasets), comparing them against established FS methods across several classifiers, including Random Forest, Extra Trees, MLP, and KNN. Performance is assessed using accuracy, precision, recall, F1, AUC, MCC, and runtime. Our results demonstrate that LLM-guided zero-shot feature selection achieves competitive performance with traditional FS methods while offering additional advantages in interpretability, stability, and reduced dependence on labeled data. These findings position zero-shot LLM-based FS as a promising alternative strategy for effective and interpretable malware detection, paving the way for knowledge-guided feature selection in security-critical applications
Authors: Zahra Kadkhodaie, Aram-Alexandre Pooladian, Sinho Chewi, Eero Simoncelli
Abstract: We analyze, theoretically and empirically, the performance of generative diffusion models based on \emph{blind denoisers}, in which the denoiser is not given the noise amplitude in either the training or sampling processes. Assuming that the data distribution has low intrinsic dimensionality, we prove that blind denoising diffusion models (BDDMs), despite not having access to the noise amplitude, \emph{automatically} track a particular \emph{implicit} noise schedule along the reverse process. Our analysis shows that BDDMs can accurately sample from the data distribution in polynomially many steps as a function of the intrinsic dimension. Empirical results corroborate these mathematical findings on both synthetic and image data, demonstrating that the noise variance is accurately estimated from the noisy image. Remarkably, we observe that schedule-free BDDMs produce samples of higher quality compared to their non-blind counterparts. We provide evidence that this performance gain arises because BDDMs correct the mismatch between the true residual noise (of the image) and the noise assumed by the schedule used in non-blind diffusion models.
Authors: Shinhoo Kang, Sangwook Kim, Sehyun Yun
Abstract: The transition toward low-inertia power systems demands modeling frameworks that provide not only accurate state predictions but also physically consistent sensitivities for control. While scientific machine learning offers powerful nonlinear modeling tools, the control-oriented implications of different differentiable paradigms remain insufficiently understood. This paper presents a comparative study of Physics-Informed Neural Networks (PINNs), Neural Ordinary Differential Equations (NODEs), and Differentiable Programming (DP) for modeling, identification, and control of power system dynamics. Using the Single Machine Infinite Bus (SMIB) system as a benchmark, we evaluate their performance in trajectory extrapolation, parameter estimation, and Linear Quadratic Regulator (LQR) synthesis. Our results highlight a fundamental trade-off between data-driven flexibility and physical structure. NODE exhibits superior extrapolation by capturing the underlying vector field, whereas PINN shows limited generalization due to its reliance on a time-dependent solution map. In the inverse problem of parameter identification, while both DP and PINN successfully recover the unknown parameters, DP achieves significantly faster convergence by enforcing governing equations as hard constraints. Most importantly, for control synthesis, the DP framework yields closed-loop stability comparable to the theoretical optimum. Furthermore, we demonstrate that NODE serves as a viable data-driven surrogate when governing equations are unavailable.
Authors: Jin Li, Kleanthis Malialis, Marios Polycarpou
Abstract: In today's connected world, the generation of massive streaming data across diverse domains has become commonplace. In the presence of concept drift, class imbalance, label scarcity, and new class emergence, they jointly degrade representation stability, bias learning toward outdated distributions, and reduce the resilience and reliability of detection in dynamic environments. This paper proposes SCIL (Streaming Class-Incremental Learning) to address these challenges. The SCIL framework integrates an autoencoder (AE) with a multi-layer perceptron for multi-class prediction, uses a dual-loss strategy (classification and reconstruction) for prediction and new class detection, employs corrected pseudo-labels for online training, manages classes with queues, and applies oversampling to handle imbalance. The rationale behind the method's structure is elucidated through ablation studies and a comprehensive experimental evaluation is performed using both real-world and synthetic datasets that feature class imbalance, incremental classes, and concept drifts. Our results demonstrate that SCIL outperforms strong baselines and state-of-the-art methods. Based on our commitment to Open Science, we make our code and datasets available to the community.
Authors: Alireza Abdollahpoorrostam, Nikolaos Dimitriadis, Adam Hazimeh, Pascal Frossard
Abstract: Fine-tuning large pre-trained models on a target distribution often improves in-distribution (ID) accuracy, but at the cost of out-of-distribution (OOD) robustness as representations specialize to the fine-tuning data. Weight-space ensembling methods, such as Model Soups, mitigate this effect by averaging multiple checkpoints, but they are computationally prohibitive, requiring the training and storage of dozens of fine-tuned models. In this paper, we introduce MonoSoup, a simple, data-free, hyperparameter-free, post-hoc method that achieves a strong ID-OOD balance using only a single checkpoint. Our method applies Singular Value Decomposition (SVD) to each layer's update and decomposes it into high-energy directions that capture task-specific adaptation and low-energy directions that introduce noise but may still encode residual signals useful for robustness. MonoSoup then uses entropy-based effective rank to automatically re-weigh these components with layer-wise coefficients that account for the spectral and geometric structure of the model. Experiments on CLIP models fine-tuned on ImageNet and evaluated under natural distribution shifts, as well as on Qwen language models tested on mathematical reasoning and multiple-choice benchmarks, show that this plug-and-play approach is a practical and effective alternative to multi-checkpoint methods, retaining much of their benefits without their computational overhead.
Authors: Lingpei Zhang, Qingming Li, Yong Yang, Jiahao Chen, Rui Zeng, Chenyang Lyu, Shouling Ji
Abstract: Univariate time series (UTS), where each timestamp records a single variable, serve as crucial indicators in web systems and cloud servers. Anomaly detection in UTS plays an essential role in both data mining and system reliability management. However, existing reconstruction-based and prediction-based methods struggle to capture certain subtle anomalies, particularly small point anomalies and slowly rising anomalies. To address these challenges, we propose a novel prediction-based framework named Contextual and Seasonal LSTMs (CS-LSTMs). CS-LSTMs are built upon a noise decomposition strategy and jointly leverage contextual dependencies and seasonal patterns, thereby strengthening the detection of subtle anomalies. By integrating both time-domain and frequency-domain representations, CS-LSTMs achieve more accurate modeling of periodic trends and anomaly localization. Extensive evaluations on public benchmark datasets demonstrate that CS-LSTMs consistently outperform state-of-the-art methods, highlighting their effectiveness and practical value in robust time series anomaly detection.
Authors: Davide Gallon, Philippe von Wurstemberger, Patrick Cheridito, Arnulf Jentzen
Abstract: We propose a methodology that combines generative latent diffusion models with physics-informed machine learning to generate solutions of parametric partial differential equations (PDEs) conditioned on partial observations, which includes, in particular, forward and inverse PDE problems. We learn the joint distribution of PDE parameters and solutions via a diffusion process in a latent space of scaled spectral representations, where Gaussian noise corresponds to functions with controlled regularity. This spectral formulation enables significant dimensionality reduction compared to grid-based diffusion models and ensures that the induced process in function space remains within a class of functions for which the PDE operators are well defined. Building on diffusion posterior sampling, we enforce physics-informed constraints and measurement conditions during inference, applying Adam-based updates at each diffusion step. We evaluate the proposed approach on Poisson, Helmholtz, and incompressible Navier--Stokes equations, demonstrating improved accuracy and computational efficiency compared with existing diffusion-based PDE solvers, which are state of the art for sparse observations. Code is available at https://github.com/deeplearningmethods/PISD.
Authors: Justin Dachille, Aurora Rossi, Sunil Kumar Maurya, Frederik Mallmann-Trenn, Xin Liu, Fr\'ed\'eric Giroire, Tsuyoshi Murata, Emanuele Natale
Abstract: Computing node importance in networks is a long-standing fundamental problem that has driven extensive study of various centrality measures. A particularly well-known centrality measure is betweenness centrality, which becomes computationally prohibitive on large-scale networks. Graph Neural Network (GNN) models have thus been proposed to predict node rankings according to their relative betweenness centrality. However, state-of-the-art methods fail to generalize to high-diameter graphs such as road networks. We propose BRAVA-GNN, a lightweight GNN architecture that leverages the empirically observed correlation linking betweenness centrality to degree-based quantities, in particular multi-hop degree mass. This correlation motivates the use of degree masses as size-invariant node features and synthetic training graphs that closely match the degree distributions of real networks. Furthermore, while previous work relies on scale-free synthetic graphs, we leverage the hyperbolic random graph model, which reproduces power-law exponents outside the scale-free regime, better capturing the structure of real-world graphs like road networks. This design enables BRAVA-GNN to generalize across diverse graph families while using 54x fewer parameters than the most lightweight existing GNN baseline. Extensive experiments on 19 real-world networks, spanning social, web, email, and road graphs, show that BRAVA-GNN achieves up to 214% improvement in Kendall-Tau correlation and up to 70x speedup in inference time over state-of-the-art GNN-based approaches, particularly on challenging road networks.
Authors: Hanyong Wang, Menglong Yang
Abstract: Deep reinforcement learning has been able to solve various tasks successfully, however, due to the construction of policy gradient and training dynamics, tuning deep reinforcement learning models remains challenging. As one of the most successful deep reinforcement-learning algorithm, the Proximal Policy Optimization algorithm (PPO) clips the policy gradient within a conservative on-policy updates, which ensures reliable and stable policy improvement. However, this training pattern may sacrifice sample efficiency. On the other hand, off-policy methods make more adequate use of data through sample reuse, though at the cost of increased the estimation variance and bias. To leverage the advantages of both, in this paper, we propose a new PPO variant based on the stability guarantee from conservative on-policy iteration with a more efficient off-policy data utilization. Specifically, we first derive an extended off-policy improvement from an expectation form of generalized policy improvement lower bound. Then, we extend the clipping mechanism with segmented exponential functions for a suitable surrogate objective function. Third, the trajectories generated by the past $M$ policies are organized in the replay buffer for off-policy training. We refer to this method as Extended Off-policy Proximal Policy Optimization (ExO-PPO). Compared with PPO and some other state-of-the-art variants, we demonstrate an improved performance of ExO-PPO with balanced sample efficiency and stability on varied tasks in the empirical experiments.
Authors: Mihnea Ghitu, Matthew Wicker
Abstract: Understanding the reliability of natural language generation is critical for deploying foundation models in security-sensitive domains. While certified poisoning defenses provide provable robustness bounds for classification tasks, they are fundamentally ill-equipped for autoregressive generation: they cannot handle sequential predictions or the exponentially large output space of language models. To establish a framework for certified natural language generation, we formalize two security properties: stability (robustness to any change in generation) and validity (robustness to targeted, harmful changes in generation). We introduce Targeted Partition Aggregation (TPA), the first algorithm to certify validity/targeted attacks by computing the minimum poisoning budget needed to induce a specific harmful class, token, or phrase. Further, we extend TPA to provide tighter guarantees for multi-turn generations using mixed integer linear programming (MILP). Empirically, we demonstrate TPA's effectiveness across diverse settings including: certifying validity of agent tool-calling when adversaries modify up to 0.5% of the dataset and certifying 8-token stability horizons in preference-based alignment. Though inference-time latency remains an open challenge, our contributions enable certified deployment of language models in security-critical applications.
Authors: Matteo Pannacci, Andrea Fanti, Elena Umili, Roberto Capobianco
Abstract: In this work we address the problem of training a Reinforcement Learning agent to follow multiple temporally-extended instructions expressed in Linear Temporal Logic in sub-symbolic environments. Previous multi-task work has mostly relied on knowledge of the mapping between raw observations and symbols appearing in the formulae. We drop this unrealistic assumption by jointly training a multi-task policy and a symbol grounder with the same experience. The symbol grounder is trained only from raw observations and sparse rewards via Neural Reward Machines in a semi-supervised fashion. Experiments on vision-based environments show that our method achieves performance comparable to using the true symbol grounding and significantly outperforms state-of-the-art methods for sub-symbolic environments.
Authors: Surjo Dey, Pallabi Saikia
Abstract: This study investigates the explainability of generative diffusion models in the context of medical imaging, focusing on Magnetic resonance imaging (MRI) synthesis. Although diffusion models have shown strong performance in generating realistic medical images, their internal decision making process remains largely opaque. We present a faithfulness-based explainability framework that analyzes how prototype-based explainability methods like ProtoPNet (PPNet), Enhanced ProtoPNet (EPPNet), and ProtoPool can link the relationship between generated and training features. Our study focuses on understanding the reasoning behind image formation through denoising trajectory of diffusion model and subsequently prototype explainability with faithfulness analysis. Experimental analysis shows that EPPNet achieves the highest faithfulness (with score 0.1534), offering more reliable insights, and explainability into the generative process. The results highlight that diffusion models can be made more transparent and trustworthy through faithfulness-based explanations, contributing to safer and more interpretable applications of generative AI in healthcare.
Authors: Kun Chen, Peng Shi, Fanfan Liu, Haibo Qiu, Zhixiong Zeng, Siqi Yang, Wenji Mao
Abstract: Reinforcement Learning with Verifiable Rewards (RLVR) has emerged as a critical method for enhancing the reasoning capabilities of Large Language Models (LLMs). However, continuous training often leads to policy entropy collapse, characterized by a rapid decay in entropy that results in premature overconfidence, reduced output diversity, and vanishing gradient norms that inhibit learning. Gradient-Preserving Clipping is a primary factor influencing these dynamics, but existing mitigation strategies are largely static and lack a framework connecting clipping mechanisms to precise entropy control. This paper proposes reshaping entropy control in RL from the perspective of Gradient-Preserving Clipping. We first theoretically and empirically verify the contributions of specific importance sampling ratio regions to entropy growth and reduction. Leveraging these findings, we introduce a novel regulation mechanism using dynamic clipping threshold to precisely manage entropy. Furthermore, we design and evaluate dynamic entropy control strategies, including increase-then-decrease, decrease-increase-decrease, and oscillatory decay. Experimental results demonstrate that these strategies effectively mitigate entropy collapse, and achieve superior performance across multiple benchmarks.
Authors: Andres Saurez, Yousung Lee, Dongsoo Har
Abstract: Linear probes and sparse autoencoders consistently recover meaningful structure from transformer representations -- yet why should such simple methods succeed in deep, nonlinear systems? We show this is not merely an empirical regularity but a consequence of architectural necessity: transformers communicate information through linear interfaces (attention OV circuits, unembedding matrices), and any semantic feature decoded through such an interface must occupy a context-invariant linear subspace. We formalize this as the \emph{Invariant Subspace Necessity} theorem and derive the \emph{Self-Reference Property}: tokens directly provide the geometric direction for their associated features, enabling zero-shot identification of semantic structure without labeled data or learned probes. Empirical validation in eight classification tasks and four model families confirms the alignment between class tokens and semantically related instances. Our framework provides \textbf{a principled architectural explanation} for why linear interpretability methods work, unifying linear probes and sparse autoencoders.
Authors: Andres Saurez, Neha Sengar, Dongsoo Har
Abstract: Circuit discovery and activation steering in transformers have developed as separate research threads, yet both operate on the same representational space. Are they two views of the same underlying structure? We show they follow a single geometric principle: answer tokens, processed in isolation, encode the directions that would produce them. This Circuit Fingerprint hypothesis enables circuit discovery without gradients or causal intervention -- recovering comparable structure to gradient-based methods through geometric alignment alone. We validate this on standard benchmarks (IOI, SVA, MCQA) across four model families, achieving circuit discovery performance comparable to gradient-based methods. The same directions that identify circuit components also enable controlled steering -- achieving 69.8\% emotion classification accuracy versus 53.1\% for instruction prompting while preserving factual accuracy. Beyond method development, this read-write duality reveals that transformer circuits are fundamentally geometric structures: interpretability and controllability are two facets of the same object.
Authors: Ruishan Guo, Yibing Liu, Guoxin Ma, Yan Wang, Yueyang Zhang, Long Xia, Kecheng Chen, Zhiyuan Sun, Daiting Shi
Abstract: Scaling up model parameters has long been a prevalent training paradigm driven by the assumption that larger models yield superior generation capabilities. However, under lossy context compression in a compressor-decoder setup, we observe a Size-Fidelity Paradox: increasing the compressor size can lessen the faithfulness of reconstructed contexts though training loss decreases. Through extensive experiments across models from 0.6B to 90B, we coin this paradox arising from two dominant factors: 1) knowledge overwriting: larger models increasingly replace source facts with their own prior beliefs, e.g., ``the white strawberry'' $\to$ ``the red strawberry''; and 2) semantic drift: larger models tend to paraphrase or restructure content instead of reproducing it verbatim, e.g., ``Alice hit Bob'' $\to$ ``Bob hit Alice''. By holding model size fixed, we reflect on the emergent properties of compressed context representations. We show that the culprit is not parameter count itself, but the excessive semantic capacity and amplified generative uncertainty that accompany scaling. Specifically, the increased rank of context embeddings facilitates prior knowledge intrusion, whereas higher entropy over token prediction distributions promotes rewriting. Our results complement existing evaluations over context compression paradigm, underpinning a breakdown in scaling laws for faithful preservation in open-ended generation.
Authors: Jesper Str{\o}m, Casper Skj{\ae}rb{\ae}k, Natasha Becker Bertelsen, Steffen Torpe Simonsen, Niels Okkels, David Bertram, Sinah R\"ottgen, Konstantin Kufer, Kaare B. Mikkelsen, Marit Otto, Poul J{\o}rgen Jennum, Per Borghammer, Michael Sommerauer, Preben Kidmose
Abstract: Isolated REM sleep behavior disorder (iRBD) is a key prodromal marker of Parkinson's disease (PD), and video-polysomnography (vPSG) remains the diagnostic gold standard. However, manual sleep staging is particularly challenging in neurodegenerative diseases due to EEG abnormalities and fragmented sleep, making PSG assessments a bottleneck for deploying new RBD screening technologies at scale. We adapted U-Sleep, a deep neural network, for generalizable sleep staging in PD and iRBD. A pretrained U-Sleep model, based on a large publicly available, multisite non-neurodegenerative dataset (PUB; 19,236 PSGs across 12 sites), was fine-tuned on research datasets from two centers (Lundbeck Foundation Parkinson's Disease Research Center (PACE) and the Cologne-Bonn Cohort (CBC); 112 PD, 138 iRBD, 89 age-matched controls. The resulting model was evaluated on an independent dataset from the Danish Center for Sleep Medicine (DCSM; 81 PD, 36 iRBD, 87 sleep-clinic controls). A subset of PSGs with low agreement between the human rater and the model (\k{appa} < 0.6) was re-scored by a second blinded human rater to identify sources of disagreement. Finally, we applied confidence-based thresholds to optimize REM sleep staging. The pretrained model achieved mean \k{appa} = 0.81 in PUB, but \k{appa} = 0.66 when applied directly to PACE/CBC. By fine-tuning the model, we developed a generalized model with \k{appa} = 0.74 on PACE/CBC (p < 0.001 vs. the pretrained model). In DCSM, mean and median \k{appa} increased from 0.60 to 0.64 (p < 0.001) and 0.64 to 0.69 (p < 0.001), respectively. In the interrater study, PSGs with low agreement between the model and the initial scorer showed similarly low agreement between human scorers. Applying a confidence threshold increased the proportion of correctly identified REM sleep epochs from 85% to 95.5%, while preserving sufficient (> 5 min) REM sleep for 95% of subjects.
Authors: Azka Nasir, Fatima Dossa, Muhammad Ahmed Atif, Mohammad Ahmed Atif
Abstract: Transfer learning in deep reinforcement learning is often motivated by improved stability and reduced training cost, but it can also fail under substantial domain shift. This paper presents a controlled empirical study examining how architectural differences between Double Deep Q-Networks (DDQN) and Dueling DQN influence transfer behavior across environments. Using CartPole as a source task and LunarLander as a structurally distinct target task, we evaluate a fixed layer-wise representation transfer protocol under identical hyperparameters and training conditions, with baseline agents trained from scratch used to contextualize transfer effects. Empirical results show that DDQN consistently avoids negative transfer under the examined setup and maintains learning dynamics comparable to baseline performance in the target environment. In contrast, Dueling DQN consistently exhibits negative transfer under identical conditions, characterized by degraded rewards and unstable optimization behavior. Statistical analysis across multiple random seeds confirms a significant performance gap under transfer. These findings suggest that architectural inductive bias is strongly associated with robustness to cross-environment transfer in value-based deep reinforcement learning under the examined transfer protocol.
Authors: Xuhang Wu, Zhuoxuan Liang, Wei Li, Xiaohua Jia, Sumi Helal
Abstract: With the rapid advancement of IoT and edge computing, sensor networks have become indispensable, driving the need for large-scale sensor deployment. However, the high deployment cost hinders their scalability. To tackle the issues, Spatial Interpolation (SI) introduces virtual sensors to infer readings from observed sensors, leveraging graph structure. However, current graph-based SI methods rely on pre-trained models, lack adaptation to larger and unseen graphs at test-time, and overlook test data utilization. To address these issues, we propose PlugSI, a plug-and-play framework that refines test-time graph through two key innovations. First, we design an Unknown Topology Adapter (UTA) that adapts to the new graph structure of each small-batch at test-time, enhancing the generalization of SI pre-trained models. Second, we introduce a Temporal Balance Adapter (TBA) that maintains a stable historical consensus to guide UTA adaptation and prevent drifting caused by noise in the current batch. Empirically, extensive experiments demonstrate PlugSI can be seamlessly integrated into existing graph-based SI methods and provide significant improvement (e.g., a 10.81% reduction in MAE).
Authors: Beicheng Xu, Keyao Ding, Wei Liu, Yupeng Lu, Bin Cui
Abstract: Feature Engineering (FE) is pivotal in automated machine learning (AutoML) but remains a bottleneck for traditional methods, which treat it as a black-box search, operating within rigid, predefined search spaces and lacking domain awareness. While Large Language Models (LLMs) offer a promising alternative by leveraging semantic reasoning to generate unbounded operators, existing methods fail to construct free-form FE pipelines, remaining confined to isolated subtasks such as feature generation. Most importantly, they are rarely optimized jointly with hyperparameter optimization (HPO) of the ML model, leading to greedy "FE-then-HPO" workflows that cannot capture strong FE-HPO interactions. In this paper, we present CoFEH, a collaborative framework that interleaves LLM-based FE and Bayesian HPO for robust end-to-end AutoML. CoFEH uses an LLM-driven FE optimizer powered by Tree of Thought (ToT) to explore flexible FE pipelines, a Bayesian optimization (BO) module to solve HPO, and a dynamic optimizer selector that realizes interleaved optimization by adaptively scheduling FE and HPO steps. Crucially, we introduce a mutual conditioning mechanism that shares context between LLM and BO, enabling mutually informed decisions. Experiments show that CoFEH not only outperforms traditional and LLM-based FE baselines, but also achieves superior end-to-end performance under joint optimization.
Authors: Beno\^it Hurpeau
Abstract: Clustering heterogeneous relational data remains a central challenge in graph learning, particularly when interactions involve more than two types of entities. While differentiable modularity objectives such as DMoN have enabled end-to-end community detection on homogeneous and bipartite graphs, extending these approaches to higher-order relational structures remains non-trivial. In this work, we introduce a differentiable formulation of tripartite modularity for graphs composed of three node types connected through mediated interactions. Community structure is defined in terms of weighted co-paths across the tripartite graph, together with an exact factorized computation that avoids the explicit construction of dense third-order tensors. A structural normalization at pivot nodes is introduced to control extreme degree heterogeneity and ensure stable optimization. The resulting objective can be optimized jointly with a graph neural network in an end-to-end manner, while retaining linear complexity in the number of edges. We validate the proposed framework on large-scale urban cadastral data, where it exhibits robust convergence behavior and produces spatially coherent partitions. These results highlight differentiable tripartite modularity as a generic methodological building block for unsupervised clustering of heterogeneous graphs.
Authors: Cyril Garcia, Guillaume Remy
Abstract: We study the performance of transformer architectures for multivariate time-series forecasting in low-data regimes consisting of only a few years of daily observations. Using synthetically generated processes with known temporal and cross-sectional dependency structures and varying signal-to-noise ratios, we conduct bootstrapped experiments that enable direct evaluation via out-of-sample correlations with the optimal ground-truth predictor. We show that two-way attention transformers, which alternate between temporal and cross-sectional self-attention, can outperform standard baselines-Lasso, boosting methods, and fully connected multilayer perceptrons-across a wide range of settings, including low signal-to-noise regimes. We further introduce a dynamic sparsification procedure for attention matrices applied during training, and demonstrate that it becomes significantly effective in noisy environments, where the correlation between the target variable and the optimal predictor is on the order of a few percent. Analysis of the learned attention patterns reveals interpretable structure and suggests connections to sparsity-inducing regularization in classical regression, providing insight into why these models generalize effectively under noise.
Authors: Anna Bodonhelyi, Mengdi Wang, Efe Bozkir, Babette B\"uhler, Enkelejda Kasneci
Abstract: Since the COVID-19 pandemic, online courses have expanded access to education, yet the absence of direct instructor support challenges learners' ability to self-regulate attention and engagement. Mind wandering and disengagement can be detrimental to learning outcomes, making their automated detection via video-based indicators a promising approach for real-time learner support. However, machine learning-based approaches often require sharing sensitive data, raising privacy concerns. Federated learning offers a privacy-preserving alternative by enabling decentralized model training while also distributing computational load. We propose a framework exploiting cross-device federated learning to address different manifestations of behavioral and cognitive disengagement during remote learning, specifically behavioral disengagement, mind wandering, and boredom. We fit video-based cognitive disengagement detection models using facial expressions and gaze features. By adopting federated learning, we safeguard users' data privacy through privacy-by-design and introduce a novel solution with the potential for real-time learner support. We further address challenges posed by eyeglasses by incorporating related features, enhancing overall model performance. To validate the performance of our approach, we conduct extensive experiments on five datasets and benchmark multiple federated learning algorithms. Our results show great promise for privacy-preserving educational technologies promoting learner engagement.
Authors: Daanish Aleem Qureshi, Khemraj Shukla, Vikas Srivastava
Abstract: Accurate modeling of drug release is essential for designing and developing controlled-release systems. Classical models (Fick, Higuchi, Peppas) rely on simplifying assumptions that limit their accuracy in complex geometries and release mechanisms. Here, we propose a novel approach using Physics-Informed Neural Networks (PINNs) and Bayesian PINNs (BPINNs) for predicting release from planar, 1D-wrinkled, and 2D-crumpled films. This approach uniquely integrates Fick's diffusion law with limited experimental data to enable accurate long-term predictions from short-term measurements, and is systematically benchmarked against classical drug release models. We embedded Fick's second law into PINN as loss with 10,000 Latin-hypercube collocation points and utilized previously published experimental datasets to assess drug release performance through mean absolute error (MAE) and root mean square error (RMSE), considering noisy conditions and limited-data scenarios. Our approach reduced mean error by up to 40% relative to classical baselines across all film types. The PINN formulation achieved RMSE <0.05 utilizing only the first 6% of the release time data (reducing 94% of release time required for the experiments) for the planar film. For wrinkled and crumpled films, the PINN reached RMSE <0.05 in 33% of the release time data. BPINNs provide tighter and more reliable uncertainty quantification under noise. By combining physical laws with experimental data, the proposed framework yields highly accurate long-term release predictions from short-term measurements, offering a practical route for accelerated characterization and more efficient early-stage drug release system formulation.
Authors: Varun Gupta, Vijay Kamble
Abstract: We study a canonical multi-task demand learning problem motivated by retail pricing, in which a firm seeks to estimate heterogeneous linear price-response functions across a large collection of decision contexts. Each context is characterized by rich observable covariates yet typically exhibits only limited historical price variation, motivating the use of multi-task learning to borrow strength across tasks. A central challenge in this setting is endogeneity: historical prices are chosen by managers or algorithms and may be arbitrarily correlated with unobserved, task-level demand determinants. Under such confounding by latent fundamentals, commonly used approaches, such as pooled regression and meta-learning, fail to identify causal price effects. We propose a new estimation framework that achieves causal identification despite arbitrary dependence between prices and latent task structure. Our approach, Decision-Conditioned Masked-Outcome Meta-Learning (DCMOML), involves carefully designing the information set of a meta-learner to leverage cross-task heterogeneity while accounting for endogenous decision histories. Under a mild restriction on price adaptivity in each task, we establish that this method identifies the conditional mean of the task-specific causal parameters given the designed information set. Our results provide guarantees for large-scale demand estimation with endogenous prices and small per-task samples, offering a principled foundation for deploying causal, data-driven pricing models in operational environments.
Authors: Enzo Nicolas Spotorno, Josafat Ribeiro Leal, Antonio Augusto Frohlich
Abstract: Standard Physics-Informed Neural Networks (PINNs) often face challenges when modeling parameterized dynamical systems with sharp regime transitions, such as bifurcations. In these scenarios, the continuous mapping from parameters to solutions can result in spectral bias or "mode collapse", where the network averages distinct physical behaviors. We propose a Topology-Aware PINN (TAPINN) that aims to mitigate this challenge by structuring the latent space via Supervised Metric Regularization. Unlike standard parametric PINNs that map physical parameters directly to solutions, our method conditions the solver on a latent state optimized to reflect the metric-based separation between regimes, showing ~49% lower physics residual (0.082 vs. 0.160). We train this architecture using a phase-based Alternating Optimization (AO) schedule to manage gradient conflicts between the metric and physics objectives. Preliminary experiments on the Duffing Oscillator demonstrate that while standard baselines suffer from spectral bias and high-capacity Hypernetworks overfit (memorizing data while violating physics), our approach achieves stable convergence with 2.18x lower gradient variance than a multi-output Sobolev Error baseline, and 5x fewer parameters than a hypernetwork-based alternative.
Authors: Alexander Fertig, Karthikeyan Chandra Sekaran, Lakshman Balasubramanian, Michael Botsch
Abstract: As autonomous vehicles are rolled out, measures must be taken to ensure their safe operation. In order to supervise a system that is already in operation, monitoring frameworks are frequently employed. These run continuously online in the background, supervising the system status and recording anomalies. This work proposes an online monitoring framework to detect anomalies in object state representations. Thereby, a key challenge is creating a framework for anomaly detection without anomaly labels, which are usually unavailable for unknown anomalies. To address this issue, this work applies a self-supervised embedding method to translate object data into a latent representation space. For this, a JEPA-based self-supervised prediction task is constructed, allowing training without anomaly labels and the creation of rich object embeddings. The resulting expressive JEPA embeddings serve as input for established anomaly detection methods, in order to identify anomalies within object state representations. This framework is particularly useful for applications in real-world environments, where new or unknown anomalies may occur during operation for which there are no labels available. Experiments performed on the publicly available, real-world nuScenes dataset illustrate the framework's capabilities.
Authors: J Rosser, Robert Kirk, Edward Grefenstette, Jakob Foerster, Laura Ruis
Abstract: Influence functions are commonly used to attribute model behavior to training documents. We explore the reverse: crafting training data that induces model behavior. Our framework, Infusion, uses scalable influence-function approximations to compute small perturbations to training documents that induce targeted changes in model behavior through parameter shifts. We evaluate Infusion on data poisoning tasks across vision and language domains. On CIFAR-10, we show that making subtle edits via Infusion to just 0.2% (100/45,000) of the training documents can be competitive with the baseline of inserting a small number of explicit behavior examples. We also find that Infusion transfers across architectures (ResNet $\leftrightarrow$ CNN), suggesting a single poisoned corpus can affect multiple independently trained models. In preliminary language experiments, we characterize when our approach increases the probability of target behaviors and when it fails, finding it most effective at amplifying behaviors the model has already learned. Taken together, these results show that small, subtle edits to training data can systematically shape model behavior, underscoring the importance of training data interpretability for adversaries and defenders alike. We provide the code here: https://github.com/jrosseruk/infusion.
Authors: Enzo Nicolas Spotorno, Josafat Leal Filho, Antonio Augusto Medeiros Frohlich
Abstract: We investigate the integration of Kolmogorov-Arnold Networks (KANs) into hard-constrained recurrent physics-informed architectures (HRPINN) to evaluate the fidelity of learned residual manifolds in oscillatory systems. Motivated by the Kolmogorov-Arnold representation theorem and preliminary gray-box results, we hypothesized that KANs would enable efficient recovery of unknown terms compared to MLPs. Through initial sensitivity analysis on configuration sensitivity, parameter scale, and training paradigm, we found that while small KANs are competitive on univariate polynomial residuals (Duffing), they exhibit severe hyperparameter fragility, instability in deeper configurations, and consistent failure on multiplicative terms (Van der Pol), generally outperformed by standard MLPs. These empirical challenges highlight limitations of the additive inductive bias in the original KAN formulation for state coupling and provide preliminary empirical evidence of inductive bias limitations for future hybrid modeling.
Authors: Shijie Zhang, Xiang Guo, Rujun Guo, Shaoyu Liu, Xiaozhao Wang, Guanjun Jiang, Kevin Zhang
Abstract: Building a search relevance model that achieves both low latency and high performance is a long-standing challenge in the search industry. To satisfy the millisecond-level response requirements of online systems while retaining the interpretable reasoning traces of Large Language Models (LLMs), we propose a novel \textbf{Answer-First, Reason Later (AFRL)} paradigm. This paradigm requires the model to output the definitive relevance score in the very first token, followed by a structured logical explanation. Inspired by the success of reasoning models, we adopt a "Supervised Fine-Tuning (SFT) + Reinforcement Learning (RL)" pipeline to achieve AFRL. However, directly applying existing RL training often leads to \textbf{mode collapse} in the search relevance task, where the model forgets complex long-tail rules in pursuit of high rewards. From an information theory perspective: RL inherently minimizes the \textbf{Reverse KL divergence}, which tends to seek probability peaks (mode-seeking) and is prone to "reward hacking." On the other hand, SFT minimizes the \textbf{Forward KL divergence}, forcing the model to cover the data distribution (mode-covering) and effectively anchoring expert rules. Based on this insight, we propose a \textbf{Mode-Balanced Optimization} strategy, incorporating an SFT auxiliary loss into Stepwise-GRPO training to balance these two properties. Furthermore, we construct an automated instruction evolution system and a multi-stage curriculum to ensure expert-level data quality. Extensive experiments demonstrate that our 32B teacher model achieves state-of-the-art performance. Moreover, the AFRL architecture enables efficient knowledge distillation, successfully transferring expert-level logic to a 0.6B model, thereby reconciling reasoning depth with deployment latency.
Authors: Chenruo Liu, Yijun Dong, Yiqiu Shen, Qi Lei
Abstract: Iterative self-improvement fine-tunes an autoregressive large language model (LLM) on reward-verified outputs generated by the LLM itself. In contrast to the empirical success of self-improvement, the theoretical foundation of this generative, iterative procedure in a practical, finite-sample setting remains limited. We make progress toward this goal by modeling each round of self-improvement as maximum-likelihood fine-tuning on a reward-filtered distribution and deriving finite-sample guarantees for the expected reward. Our analysis reveals an explicit feedback loop where better models accept more data per iteration, supporting sustained self-improvement while explaining eventual saturation of such improvement. Adopting a task-centric view by considering reasoning tasks with multiple difficulty levels, we further prove quantifiable conditions on model initialization, task difficulty, and sample budget where easy-to-hard curricula provably achieve better guarantees than training on fixed mixtures of tasks. Our analyses are validated via Monte-Carlo simulations and controlled experiments on graph-based reasoning tasks.
Authors: Qingnan Ren, Shiting Huang, Zhen Fang, Zehui Chen, Lin Chen, Lijun Li, Feng Zhao
Abstract: Reinforcement learning has become a cornerstone technique for developing reasoning models in complex tasks, ranging from mathematical problem-solving to imaginary reasoning. The optimization of these models typically relies on policy gradient methods, whose efficacy hinges on the accurate estimation of an advantage function. However, prevailing methods typically employ static advantage estimation, a practice that leads to inefficient credit assignment by neglecting the dynamic utility of training samples over time. This limitation results in suboptimal policy updates, which in turn manifest as slower convergence rates and increased learning instability, as models fail to adapt to evolving sample utilities effectively. To address this problem, we introduce \textbf{ADORA} (\textbf{A}dvantage \textbf{D}ynamics via \textbf{O}nline \textbf{R}ollout \textbf{A}daptation), a novel framework for policy optimization. ADORA dynamically adjusts the advantage function's weighting by adaptively categorizing training data into temporarily advantageous and disadvantageous samples, based on their evolving utility during online model rollouts. This tailored data differentiation strategy allows ADORA to be seamlessly integrated into existing policy optimization algorithms without significant architectural modifications, enabling the policy to prioritize learning from more informative experiences and thereby achieve more efficient policy updates. Extensive evaluations across diverse model families and varying data scales demonstrate that ADORA is a robust and efficient framework. It significantly enhances long reasoning in both geometric and mathematical tasks, consistently achieving notable performance gains without requiring sensitive hyperparameter tuning.
Authors: Antonis Vasileiou, Juan Cervino, Pascal Frossard, Charilaos I. Kanatsoulis, Christopher Morris, Michael T. Schaub, Pierre Vandergheynst, Zhiyang Wang, Guy Wolf, Ron Levie
Abstract: Graph neural networks (GNNs) are commonly divided into message-passing neural networks (MPNNs) and spectral graph neural networks, reflecting two largely separate research traditions in machine learning and signal processing. This paper argues that this divide is mostly artificial, hindering progress in the field. We propose a viewpoint in which both MPNNs and spectral GNNs are understood as different parametrizations of permutation-equivariant operators acting on graph signals. From this perspective, many popular architectures are equivalent in expressive power, while genuine gaps arise only in specific regimes. We further argue that MPNNs and spectral GNNs offer complementary strengths. That is, MPNNs provide a natural language for discrete structure and expressivity analysis using tools from logic and graph isomorphism research, while the spectral perspective provides principled tools for understanding smoothing, bottlenecks, stability, and community structure. Overall, we posit that progress in graph learning will be accelerated by clearly understanding the key similarities and differences between these two types of GNNs, and by working towards unifying these perspectives within a common theoretical and conceptual framework rather than treating them as competing paradigms.
Authors: Tetsuro Abe, Masashi Yamashita, Shu Tanaka
Abstract: In black-box combinatorial optimization, objective evaluations are often expensive, so high quality solutions must be found under a limited budget. Factorization machine with quantum annealing (FMQA) builds a quadratic surrogate model from evaluated samples and optimizes it on an Ising machine. However, FMQA requires binary decision variables, and for nonbinary structures such as integer permutations, the choice of binary encoding strongly affects search efficiency. If the encoding fails to reflect the original neighborhood structure, small Hamming moves may not correspond to meaningful modifications in the original solution space, and constrained problems can yield many infeasible candidates that waste evaluations. Recent work combines FMQA with a binary autoencoder (bAE) that learns a compact binary latent code from feasible solutions, yet the mechanism behind its performance gains is unclear. Using a small traveling salesman problem as an interpretable testbed, we show that the bAE reconstructs feasible tours accurately and, compared with manually designed encodings at similar compression, better aligns tour distances with latent Hamming distances, yields smoother neighborhoods under small bit flips, and produces fewer local optima. These geometric properties explain why bAE+FMQA improves the approximation ratio faster while maintaining feasibility throughout optimization, and they provide guidance for designing latent representations for black-box optimization.
Authors: Akshay Mete, Shahid Aamir Sheikh, Tzu-Hsiang Lin, Dileep Kalathil, P. R. Kumar
Abstract: Efficient exploration remains a central challenge in reinforcement learning (RL), particularly in sparse-reward environments. We introduce Optimistic World Models (OWMs), a principled and scalable framework for optimistic exploration that brings classical reward-biased maximum likelihood estimation (RBMLE) from adaptive control into deep RL. In contrast to upper confidence bound (UCB)-style exploration methods, OWMs incorporate optimism directly into model learning by augmentation with an optimistic dynamics loss that biases imagined transitions toward higher-reward outcomes. This fully gradient-based loss requires neither uncertainty estimates nor constrained optimization. Our approach is plug-and-play with existing world model frameworks, preserving scalability while requiring only minimal modifications to standard training procedures. We instantiate OWMs within two state-of-the-art world model architectures, leading to Optimistic DreamerV3 and Optimistic STORM, which demonstrate significant improvements in sample efficiency and cumulative return compared to their baseline counterparts.
Authors: Xinchen Han, Hossam Afifi, Michel Marot, Xilu Wang, Lu Yin
Abstract: Large Language Models (LLMs) often generate unnecessarily verbose Chain-of-Thought (CoT) reasoning that increases computational costs and latency without proportional performance gains. In this paper, we propose \textbf{F}ine-grained \textbf{G}roup policy \textbf{O}ptimization (\textbf{FGO}), a Reinforcement Learning (RL) algorithm that refines group responses by subdividing them and assigning appropriate weights based on length and entropy, thereby enabling effective CoT compression. Meanwhile, as an enhanced variant of Group Relative Policy Optimization (GRPO), FGO successfully addresses two major limitations of the GRPO: inefficient data utilization and entropy collapse. We evaluate FGO on multiple reasoning LLMs and benchmarks, including MATH500, AIME24, AMC23, and Minerva. Experimental results show that FGO achieves efficient CoT compression without degrading performance, and simultaneously resolves the key limitations of GRPO.
Authors: Tobias Schr\"oder, Lester Mackey
Abstract: We introduce WildCat, a high-accuracy, low-cost approach to compressing the attention mechanism in neural networks. While attention is a staple of modern network architectures, it is also notoriously expensive to deploy due to resource requirements that scale quadratically with the input sequence length $n$. WildCat avoids these quadratic costs by only attending over a small weighted coreset. Crucially, we select the coreset using a fast but spectrally-accurate subsampling algorithm -- randomly pivoted Cholesky -- and weight the elements optimally to minimise reconstruction error. Remarkably, given bounded inputs, WildCat approximates exact attention with super-polynomial $O(n^{-\sqrt{\log(\log(n))}})$ error decay while running in near-linear $O(n^{1+o(1)})$ time. In contrast, prior practical approximations either lack error guarantees or require quadratic runtime to guarantee such high fidelity. We couple this advance with a GPU-optimized PyTorch implementation and a suite of benchmark experiments demonstrating the benefits of WildCat for image generation, image classification, and language model KV cache compression.
Authors: Amey P. Pasarkar, Adji Bousso Dieng
Abstract: Out-of-distribution (OOD) detection is critical for the safe deployment of machine learning systems. Existing post-hoc detectors typically rely on model confidence scores or likelihood estimates in feature space, often under restrictive distributional assumptions. In this work, we introduce a third paradigm and formulate OOD detection from a diversity perspective. We propose the Vendi Novelty Score (VNS), an OOD detector based on the Vendi Scores (VS), a family of similarity-based diversity metrics. VNS quantifies how much a test sample increases the VS of the in-distribution feature set, providing a principled notion of novelty that does not require density modeling. VNS is linear-time, non-parametric, and naturally combines class-conditional (local) and dataset-level (global) novelty signals. Across multiple image classification benchmarks and network architectures, VNS achieves state-of-the-art OOD detection performance. Remarkably, VNS retains this performance when computed using only 1% of the training data, enabling deployment in memory- or access-constrained settings.
Authors: Aaditya Vikram Prasad, Connor Watts, Jack Merullo, Dhruvil Gala, Owen Lewis, Thomas McGrath, Ekdeep Singh Lubana
Abstract: Language models trained on large-scale datasets have been shown to learn features that encode abstract concepts such as factuality or intent. Such features are traditionally used for test-time monitoring or steering. We present an alternative affordance: features as scalable supervision for open-ended tasks. We consider the case of hallucination-reduction as a desirable, yet open-ended behavior and design a reinforcement learning (RL) pipeline, titled RLFR (Reinforcement Learning from Feature Rewards), that uses features as reward functions. Grounded in a novel probing framework that identifies candidate hallucinated claims, our pipeline teaches a model to intervene and correct its completions when it is uncertain of their factuality. Furthermore, the pipeline enables scalable test-time compute, guided once more by our reward features. This end-to-end process operationalized on Gemma-3-12B-IT results in a policy that is 58% less likely to hallucinate compared to the original model, while preserving performance on standard benchmarks. Taken together, by grounding supervision in the language of features, this paper introduces a novel paradigm in the use of interpretability for learning open-ended tasks.
Authors: Georgios Kaissis, David Mildenberger, Juan Felipe Gomez, Martin J. Menten, Eleni Triantafillou
Abstract: We study how individual training examples shape the internal computation of looped transformers, where a shared block is applied for $\tau$ recurrent iterations to enable latent reasoning. Existing training-data influence estimators such as TracIn yield a single scalar score that aggregates over all loop iterations, obscuring when during the recurrent computation a training example matters. We introduce \textit{Step-Decomposed Influence (SDI)}, which decomposes TracIn into a length-$\tau$ influence trajectory by unrolling the recurrent computation graph and attributing influence to specific loop iterations. To make SDI practical at transformer scale, we propose a TensorSketch implementation that never materialises per-example gradients. Experiments on looped GPT-style models and algorithmic reasoning tasks show that SDI scales excellently, matches full-gradient baselines with low error and supports a broad range of data attribution and interpretability tasks with per-step insights into the latent reasoning process.
Authors: Amandeep Kumar, Vishal M. Patel
Abstract: Leveraging representation encoders for generative modeling offers a path for efficient, high-fidelity synthesis. However, standard diffusion transformers fail to converge on these representations directly. While recent work attributes this to a capacity bottleneck proposing computationally expensive width scaling of diffusion transformers we demonstrate that the failure is fundamentally geometric. We identify Geometric Interference as the root cause: standard Euclidean flow matching forces probability paths through the low-density interior of the hyperspherical feature space of representation encoders, rather than following the manifold surface. To resolve this, we propose Riemannian Flow Matching with Jacobi Regularization (RJF). By constraining the generative process to the manifold geodesics and correcting for curvature-induced error propagation, RJF enables standard Diffusion Transformer architectures to converge without width scaling. Our method RJF enables the standard DiT-B architecture (131M parameters) to converge effectively, achieving an FID of 3.37 where prior methods fail to converge. Code: https://github.com/amandpkr/RJF
Authors: J\'ulio Oliveira, Rodrigo Ferreira, Andr\'e Riker, Glaucio H. S. Carvalho, Eirini Eleni Tsilopoulou
Abstract: Data privacy and eXplainable Artificial Intelligence (XAI) are two important aspects for modern Machine Learning systems. To enhance data privacy, recent machine learning models have been designed as a Federated Learning (FL) system. On top of that, additional privacy layers can be added, via Differential Privacy (DP). On the other hand, to improve explainability, ML must consider more interpretable approaches with reduced number of features and less complex internal architecture. In this context, this paper aims to achieve a machine learning (ML) model that combines enhanced data privacy with explainability. So, we propose a FL solution, called Federated EXplainable Trees with Differential Privacy (FEXT-DP), that: (i) is based on Decision Trees, since they are lightweight and have superior explainability than neural networks-based FL systems; (ii) provides additional layer of data privacy protection applying Differential Privacy (DP) to the Tree-Based model. However, there is a side effect adding DP: it harms the explainability of the system. So, this paper also presents the impact of DP protection on the explainability of the ML model. The carried out performance assessment shows improvements of FEXT-DP in terms of a faster training, i.e., numbers of rounds, Mean Squared Error and explainability.
Authors: Iv\'an Arcuschin, David Chanin, Adri\`a Garriga-Alonso, Oana-Maria Camburu
Abstract: Large Language Models (LLMs) often provide chain-of-thought (CoT) reasoning traces that appear plausible, but may hide internal biases. We call these *unverbalized biases*. Monitoring models via their stated reasoning is therefore unreliable, and existing bias evaluations typically require predefined categories and hand-crafted datasets. In this work, we introduce a fully automated, black-box pipeline for detecting task-specific unverbalized biases. Given a task dataset, the pipeline uses LLM autoraters to generate candidate bias concepts. It then tests each concept on progressively larger input samples by generating positive and negative variations, and applies statistical techniques for multiple testing and early stopping. A concept is flagged as an unverbalized bias if it yields statistically significant performance differences while not being cited as justification in the model's CoTs. We evaluate our pipeline across six LLMs on three decision tasks (hiring, loan approval, and university admissions). Our technique automatically discovers previously unknown biases in these models (e.g., Spanish fluency, English proficiency, writing formality). In the same run, the pipeline also validates biases that were manually identified by prior work (gender, race, religion, ethnicity). More broadly, our proposed approach provides a practical, scalable path to automatic task-specific bias discovery.
Authors: Maria De La Luz Lomboy Toledo, Daniel Onah
Abstract: Diabetes is a chronic metabolic disorder characterized by elevated blood glucose levels due to impaired insulin production or function. Two main forms are recognized: type 1 diabetes (T1D), which involves autoimmune destruction of insulin-producing \b{eta}-cells, and type 2 diabetes (T2D), which arises from insulin resistance and progressive \b{eta}-cell dysfunction. Understanding the molecular mechanisms underlying these diseases is essential for the development of improved therapeutic strategies, particularly those targeting \b{eta}-cell dysfunction. To investigate these mechanisms in a controlled and biologically interpretable setting, mouse models have played a central role in diabetes research. Owing to their genetic and physiological similarity to humans, together with the ability to precisely manipulate their genome, mice enable detailed investigation of disease progression and gene function. In particular, mouse models have provided critical insights into \b{eta}-cell development, cellular heterogeneity, and functional failure under diabetic conditions. Building on these experimental advances, this study applies machine learning methods to single-cell transcriptomic data from mouse pancreatic islets. Specifically, we evaluate two supervised approaches identified in the literature; Extra Trees Classifier (ETC) and Partial Least Squares Discriminant Analysis (PLS-DA), to assess their ability to identify T2D-associated gene expression signatures at single-cell resolution. Model performance is evaluated using standard classification metrics, with an emphasis on interpretability and biological relevance
Authors: Georgios Ioannides, Adrian Kieback, Judah Goldfeder, Linsey Pang, Aman Chadha, Aaron Elkins, Yann LeCun, Ravid Shwartz-Ziv
Abstract: Joint Embedding Predictive Architectures (JEPA) offer a promising approach to self-supervised speech representation learning, but suffer from representation collapse without explicit grounding. We propose GMM-Anchored JEPA, which fits a Gaussian Mixture Model once on log-mel spectrograms and uses its frozen soft posteriors as auxiliary targets throughout training. A decaying supervision schedule allows GMM regularization to dominate early training before gradually yielding to the JEPA objective. Unlike HuBERT and WavLM, which require iterative re-clustering, our approach clusters input features once with soft rather than hard assignments. On ~50k hours of speech, GMM anchoring improves ASR (28.68% vs. 33.22% WER), emotion recognition (67.76% vs. 65.46%), and slot filling (64.7% vs. 59.1% F1) compared to a WavLM-style baseline with matched compute. Cluster analysis shows GMM-anchored representations achieve up to 98% entropy compared to 31% for WavLM-style, indicating substantially more uniform cluster utilization. Code is made available at https://github.com/gioannides/clustering-anchored-jepa.
URLs: https://github.com/gioannides/clustering-anchored-jepa.
Authors: Aditya Srinivas Menon, Kumud Tripathi, Raj Gohil, Pankaj Wasnik
Abstract: Self-supervised learning (SSL) has advanced speech processing but suffers from quadratic complexity due to self-attention. To address this, SummaryMixing (SM) has been proposed as a linear-time alternative that summarizes entire utterances using mean pooling but lacks sufficient local context. In this work, we introduce Windowed SummaryMixing (WSM), which enhances SM by integrating local neighborhood summaries alongside the global summary, maintaining efficiency while improving temporal dependencies. Additionally, we introduce a selective fine-tuning approach, replacing self-attention layers in SSL models with WSM blocks and fine-tuning only these blocks in low-resource settings. Our approach improves ASR performance while reducing peak VRAM usage by 40\% in the SSL models. WSM blocks have linear-time complexity with enhanced context awareness. Selectively replacing some attention layers reduces compute, memory, and latency, making it ideal for low-resource speech recognition.
Authors: Zhipeng Chang, Wenrui Hao, Nian Liu
Abstract: This paper introduces a novel optimization algorithm designed for nonlinear least-squares problems. The method is derived by preconditioning the gradient descent direction using the Singular Value Decomposition (SVD) of the Jacobian. This SVD-based preconditioner is then integrated with the first- and second-moment adaptive learning rate mechanism of the Adam optimizer. We establish the local linear convergence of the proposed method under standard regularity assumptions and prove global convergence for a modified version of the algorithm under suitable conditions. The effectiveness of the approach is demonstrated experimentally across a range of tasks, including function approximation, partial differential equation (PDE) solving, and image classification on the CIFAR-10 dataset. Results show that the proposed method consistently outperforms standard Adam, achieving faster convergence and lower error in both regression and classification settings.
Authors: Matteo Rucco
Abstract: Persistent entropy (PE) is an information-theoretic summary statistic of persistence barcodes that has been widely used to detect regime changes in complex systems. Despite its empirical success, a general theoretical understanding of when and why persistent entropy reliably detects phase transitions has remained limited, particularly in stochastic and data-driven settings. In this work, we establish a general, model-independent theorem providing sufficient conditions under which persistent entropy provably separates two phases. We show that persistent entropy exhibits an asymptotically non-vanishing gap across phases. The result relies only on continuity of persistent entropy along the convergent diagram sequence, or under mild regularization, and is therefore broadly applicable across data modalities, filtrations, and homological degrees. To connect asymptotic theory with finite-time computations, we introduce an operational framework based on topological stabilization, defining a topological transition time by stabilizing a chosen topological statistic over sliding windows, and a probability-based estimator of critical parameters within a finite observation horizon. We validate the framework on the Kuramoto synchronization transition, the Vicsek order-to-disorder transition in collective motion, and neural network training dynamics across multiple datasets and architectures. Across all experiments, stabilization of persistent entropy and collapse of variability across realizations provide robust numerical signatures consistent with the theoretical mechanism.
Authors: S M Rakib Ul Karim, Wenyi Lu, Enock Kasaadha, Sean Goggins
Abstract: Open Source Software (OSS) projects follow diverse lifecycle trajectories shaped by evolving patterns of contribution, coordination, and community engagement. Understanding these trajectories is essential for stakeholders seeking to assess project organization and health at scale. However, prior work has largely relied on static or aggregated metrics, such as project age or cumulative activity, providing limited insight into how OSS sustainability unfolds over time. In this paper, we propose a hierarchical predictive framework that models OSS projects as belonging to distinct lifecycle stages grounded in established socio-technical categorizations of OSS development. Rather than treating sustainability solely as project longevity, these lifecycle stages operationalize sustainability as a multidimensional construct integrating contribution activity, community participation, and maintenance dynamics. The framework combines engineered tabular indicators with 24-month temporal activity sequences and employs a multi-stage classification pipeline to distinguish lifecycle stages associated with different coordination and participation regimes. To support transparency, we incorporate explainable AI techniques to examine the relative contribution of feature categories to model predictions. Evaluated on a large corpus of OSS repositories, the proposed approach achieves over 94\% overall accuracy in lifecycle stage classification. Attribution analyses consistently identify contribution activity and community-related features as dominant signals, highlighting the central role of collective participation dynamics.
Authors: Stefano Balla (DISI), Stefano Zacchiroli (IP Paris, LTCI, ACES, INFRES), Thomas Degueule (LaBRI, UB), Jean-R\'emy Falleri (LaBRI, UB), Romain Robbes (LaBRI, UB)
Abstract: The ability to automatically classify source code repositories with ''topics'' that reflect their content and purpose is very useful, especially when navigating or searching through large software collections. However, existing approaches often rely heavily on README files and other metadata, which are frequently missing, limiting their applicability in real-world large-scale settings. We present DRAGON, a repository classifier designed for very large and diverse software collections. It operates entirely on lightweight signals commonly stored in version control systems: file and directory names, and optionally the README when available. In repository classification at scale, DRAGON improves F1@5 from 54.8% to 60.8%, surpassing the state of the art. DRAGON remains effective even when README files are absent, with performance degrading by only 6% w.r.t. when they are present. This robustness makes it practical for real-world settings where documentation is sparse or inconsistent. Furthermore, many of the remaining classification errors are near misses, where predicted labels are semantically close to the correct topics. This property increases the practical value of the predictions in real-world software collections, where suggesting a few related topics can still guide search and discovery. As a byproduct of developing DRAGON, we also release the largest open dataset to date for repository classification, consisting of 825 thousand repositories with associated ground-truth topics, sourced from the Software Heritage archive, providing a foundation for future large-scale and language-agnostic research on software repository understanding.
Authors: Veuns-Team, :, Changlong Gao, Zhangxuan Gu, Yulin Liu, Xinyu Qiu, Shuheng Shen, Yue Wen, Tianyu Xia, Zhenyu Xu, Zhengwen Zeng, Beitong Zhou, Xingran Zhou, Weizhi Chen, Sunhao Dai, Jingya Dou, Yichen Gong, Yuan Guo, Zhenlin Guo, Feng Li, Qian Li, Jinzhen Lin, Yuqi Zhou, Linchao Zhu, Liang Chen, Zhenyu Guo, Changhua Meng, Weiqiang Wang
Abstract: GUI agents have emerged as a powerful paradigm for automating interactions in digital environments, yet achieving both broad generality and consistently strong task performance remains challenging.In this report, we present UI-Venus-1.5, a unified, end-to-end GUI Agent designed for robust real-world applications.The proposed model family comprises two dense variants (2B and 8B) and one mixture-of-experts variant (30B-A3B) to meet various downstream application scenarios.Compared to our previous version, UI-Venus-1.5 introduces three key technical advances: (1) a comprehensive Mid-Training stage leveraging 10 billion tokens across 30+ datasets to establish foundational GUI semantics; (2) Online Reinforcement Learning with full-trajectory rollouts, aligning training objectives with long-horizon, dynamic navigation in large-scale environments; and (3) a single unified GUI Agent constructed via Model Merging, which synthesizes domain-specific models (grounding, web, and mobile) into one cohesive checkpoint. Extensive evaluations demonstrate that UI-Venus-1.5 establishes new state-of-the-art performance on benchmarks such as ScreenSpot-Pro (69.6%), VenusBench-GD (75.0%), and AndroidWorld (77.6%), significantly outperforming previous strong baselines. In addition, UI-Venus-1.5 demonstrates robust navigation capabilities across a variety of Chinese mobile apps, effectively executing user instructions in real-world scenarios. Code: https://github.com/inclusionAI/UI-Venus; Model: https://huggingface.co/collections/inclusionAI/ui-venus
URLs: https://github.com/inclusionAI/UI-Venus;, https://huggingface.co/collections/inclusionAI/ui-venus
Authors: Max Geier, Liang Fu
Abstract: Computational discovery of magnetic materials remains challenging because magnetism arises from the competition between kinetic energy and Coulomb interaction that is often beyond the reach of standard electronic-structure methods. Here we tackle this challenge by directly solving the many-electron Schr\"odinger equation with neural-network variational Monte Carlo, which provides a highly expressive variational wavefunction for strongly correlated systems. Applying this technique to transition metal dichalcogenide moir\'e semicondutors, we predict itinerant ferromagnetism in WSe$_2$/WS$_2$ and an antiferromagnetic insulator in twisted $\Gamma$-valley homobilayer, using the same neural network without any physics input beyond the microscopic Hamiltonian. Crucially, both types of magnetic states are obtained from a single calculation within the $S_z=0$ sector, removing the need to compute and compare multiple $S_z$ sectors. This significantly reduces computational cost and paves the way for faster and more reliable magnetic material design.
Authors: Ahmed Rahu, Brian Shula, Brandon Combs, Aqsa Sultana, Surendra P. Singh, Vijayan K. Asari, Derrick Forchetti
Abstract: Colorectal cancer (CRC) remains a significant cause of cancer-related mortality, despite the widespread implementation of prophylactic initiatives aimed at detecting and removing precancerous polyps. Although screening effectively reduces incidence, a notable portion of patients initially diagnosed with low-grade adenomatous polyps will still develop CRC later in life, even without the presence of known high-risk syndromes. Identifying which low-risk patients are at higher risk of progression is a critical unmet need for tailored surveillance and preventative therapeutic strategies. Traditional histological assessment of adenomas, while fundamental, may not fully capture subtle architectural or cytological features indicative of malignant potential. Advancements in digital pathology and machine learning provide an opportunity to analyze whole-slide images (WSIs) comprehensively and objectively. This study investigates whether machine learning algorithms, specifically convolutional neural networks (CNNs), can detect subtle histological features in WSIs of low-grade tubular adenomas that are predictive of a patient's long-term risk of developing colorectal cancer.
Authors: Sherman Khoo, Dennis Prangle, Song Liu, Mark Beaumont
Abstract: Simulation-based inference (SBI) enables amortized Bayesian inference by first training a neural posterior estimator (NPE) on prior-simulator pairs, typically through low-dimensional summary statistics, which can then be cheaply reused for fast inference by querying it on new test observations. Because NPE is estimated under the training data distribution, it is susceptible to misspecification when observations deviate from the training distribution. Many robust SBI approaches address this by modifying NPE training or introducing error models, coupling robustness to the inference network and compromising amortization and modularity. We introduce minimum-distance summaries, a plug-in robust NPE method that adapts queried test-time summaries independently of the pretrained NPE. Leveraging the maximum mean discrepancy (MMD) as a distance between observed data and a summary-conditional predictive distribution, the adapted summary inherits strong robustness properties from the MMD. We demonstrate that the algorithm can be implemented efficiently with random Fourier feature approximations, yielding a lightweight, model-free test-time adaptation procedure. We provide theoretical guarantees for the robustness of our algorithm and empirically evaluate it on a range of synthetic and real-world tasks, demonstrating substantial robustness gains with minimal additional overhead.
Authors: Aditi Gupta, Raphael A. Meyer, Yotam Yaniv, Elynn Chen, N. Benjamin Erichson
Abstract: To ensure high quality outputs, it is important to quantify the epistemic uncertainty of diffusion models.Existing methods are often unreliable because they mix epistemic and aleatoric uncertainty. We introduce a method based on Fisher information that explicitly isolates epistemic variance, producing more reliable plausibility scores for generated data. To make this approach scalable, we propose FLARE (Fisher-Laplace Randomized Estimator), which approximates the Fisher information using a uniformly random subset of model parameters. Empirically, FLARE improves uncertainty estimation in synthetic time-series generation tasks, achieving more accurate and reliable filtering than other methods. Theoretically, we bound the convergence rate of our randomized approximation and provide analytic and empirical evidence that last-layer Laplace approximations are insufficient for this task.
Authors: Kotekar Annapoorna Prabhu, Andrew Gan, Zahra Ghodsi
Abstract: Machine learning relies on randomness as a fundamental component in various steps such as data sampling, data augmentation, weight initialization, and optimization. Most machine learning frameworks use pseudorandom number generators as the source of randomness. However, variations in design choices and implementations across different frameworks, software dependencies, and hardware backends along with the lack of statistical validation can lead to previously unexplored attack vectors on machine learning systems. Such attacks on randomness sources can be extremely covert, and have a history of exploitation in real-world systems. In this work, we examine the role of randomness in the machine learning development pipeline from an adversarial point of view, and analyze the implementations of PRNGs in major machine learning frameworks. We present RNGGuard to help machine learning engineers secure their systems with low effort. RNGGuard statically analyzes a target library's source code and identifies instances of random functions and modules that use them. At runtime, RNGGuard enforces secure execution of random functions by replacing insecure function calls with RNGGuard's implementations that meet security specifications. Our evaluations show that RNGGuard presents a practical approach to close existing gaps in securing randomness sources in machine learning systems.
Authors: Jie Lu, Peihao Yan, Huacheng Zeng
Abstract: With over 3.5 million 5G base stations deployed globally, their collective energy consumption (projected to exceed 131 TWh annually) raises significant concerns over both operational costs and environmental impacts. In this paper, we present EExAPP, a deep reinforcement learning (DRL)-based xApp for 5G Open Radio Access Network (O-RAN) that jointly optimizes radio unit (RU) sleep scheduling and distributed unit (DU) resource slicing. EExAPP uses a dual-actor-dual-critic Proximal Policy Optimization (PPO) architecture, with dedicated actor-critic pairs targeting energy efficiency and quality-of-service (QoS) compliance. A transformer-based encoder enables scalable handling of variable user equipment (UE) populations by encoding all-UE observations into fixed-dimensional representations. To coordinate the two optimization objectives, a bipartite Graph Attention Network (GAT) is used to modulate actor updates based on both critic outputs, enabling adaptive tradeoffs between power savings and QoS. We have implemented EExAPP and deployed it on a real-world 5G O-RAN testbed with live traffic, commercial RU and smartphones. Extensive over-the-air experiments and ablation studies confirm that EExAPP significantly outperforms existing methods in reducing the energy consumption of RU while maintaining QoS.
Authors: Yihan Zhang, Hong Chang Ji, Ramji Venkataramanan, Marco Mondelli
Abstract: We consider the problem of parameter estimation from a generalized linear model with a random design matrix that is orthogonally invariant in law. Such a model allows the design have an arbitrary distribution of singular values and only assumes that its singular vectors are generic. It is a vast generalization of the i.i.d. Gaussian design typically considered in the theoretical literature, and is motivated by the fact that real data often have a complex correlation structure so that methods relying on i.i.d. assumptions can be highly suboptimal. Building on the paradigm of spectrally-initialized iterative optimization, this paper proposes optimal spectral estimators and combines them with an approximate message passing (AMP) algorithm, establishing rigorous performance guarantees for these two algorithmic steps. Both the spectral initialization and the subsequent AMP meet existing conjectures on the fundamental limits to estimation -- the former on the optimal sample complexity for efficient weak recovery, and the latter on the optimal errors. Numerical experiments suggest the effectiveness of our methods and accuracy of our theory beyond orthogonally invariant data.
Authors: Archiki Prasad, Mandar Joshi, Kenton Lee, Mohit Bansal, Peter Shaw
Abstract: Chain-of-thought (CoT) reasoning and its variants have substantially improved the performance of language models on complex reasoning tasks, yet the precise mechanisms by which different strategies facilitate generalization remain poorly understood. While current explanations often point to increased test-time computation or structural guidance, establishing a consistent, quantifiable link between these factors and generalization remains challenging. In this work, we identify intrinsic dimensionality as a quantitative measure for characterizing the effectiveness of reasoning chains. Intrinsic dimensionality quantifies the minimum number of model dimensions needed to reach a given accuracy threshold on a given task. By keeping the model architecture fixed and varying the task formulation through different reasoning strategies, we demonstrate that effective reasoning strategies consistently reduce the intrinsic dimensionality of the task. Validating this on GSM8K with Gemma-3 1B and 4B, we observe a strong inverse correlation between the intrinsic dimensionality of a reasoning strategy and its generalization performance on both in-distribution and out-of-distribution data. Our findings suggest that effective reasoning chains facilitate learning by better compressing the task using fewer parameters, offering a new quantitative metric for analyzing reasoning processes.
Authors: Minh Vu, Xiaoliang Wan, Shuangqing Wei
Abstract: The $\beta$-VAE is a foundational framework for unsupervised disentanglement, using $\beta$ to regulate the trade-off between latent factorization and reconstruction fidelity. Empirically, however, disentanglement performance exhibits a pervasive non-monotonic trend: benchmarks such as MIG and SAP typically peak at intermediate $\beta$ and collapse as regularization increases. We demonstrate that this collapse is a fundamental information-theoretic failure, where strong Kullback-Leibler pressure promotes marginal independence at the expense of the latent channel's semantic informativeness. By formalizing this mechanism in a linear-Gaussian setting, we prove that for $\beta > 1$, stationarity-induced dynamics trigger a spectral contraction of the encoder gain, driving latent-factor mutual information to zero. To resolve this, we introduce the $\lambda\beta$-VAE, which decouples regularization pressure from informational collapse via an auxiliary $L_2$ reconstruction penalty $\lambda$. Extensive experiments on dSprites, Shapes3D, and MPI3D-real confirm that $\lambda > 0$ stabilizes disentanglement and restores latent informativeness over a significantly broader range of $\beta$, providing a principled theoretical justification for dual-parameter regularization in variational inference backbones.
Authors: Pranav Kulkarni, Junfeng Guo, Heng Huang
Abstract: High-quality medical imaging datasets are essential for training deep learning models, but their unauthorized use raises serious copyright and ethical concerns. Medical imaging presents a unique challenge for existing dataset ownership verification methods designed for natural images, as static watermark patterns generated in fixed-scale images scale poorly dynamic and high-resolution scans with limited visual diversity and subtle anatomical structures, while preserving diagnostic quality. In this paper, we propose X-Mark, a sample-specific clean-label watermarking method for chest x-ray copyright protection. Specifically, X-Mark uses a conditional U-Net to generate unique perturbations within salient regions of each sample. We design a multi-component training objective to ensure watermark efficacy, robustness against dynamic scaling processes while preserving diagnostic quality and visual-distinguishability. We incorporate Laplacian regularization into our training objective to penalize high-frequency perturbations and achieve watermark scale-invariance. Ownership verification is performed in a black-box setting to detect characteristic behaviors in suspicious models. Extensive experiments on CheXpert verify the effectiveness of X-Mark, achieving WSR of 100% and reducing probability of false positives in Ind-M scenario by 12%, while demonstrating resistance to potential adaptive attacks.
Authors: Can Polat, Erchin Serpedin, Mustafa Kurban, Hasan Kurban
Abstract: Every generative model for crystalline materials harbors a critical structure size beyond which its outputs quietly become unreliable -- we call this the extrapolation frontier. Despite its direct consequences for nanomaterial design, this frontier has never been systematically measured. We introduce RADII, a radius-resolved benchmark of ${\sim}$75,000 nanoparticle structures (55-11,298 atoms) that treats radius as a continuous scaling knob to trace generation quality from in-distribution to out-of-distribution regimes under leakage-free splits. RADII provides frontier-specific diagnostics: per-radius error profiles pinpoint each architecture's scaling ceiling, surface-interior decomposition tests whether failures originate at boundaries or in bulk, and cross-metric failure sequencing reveals which aspect of structural fidelity breaks first. Benchmarking five state-of-the-art architectures, we find that: (i) all models degrade by ${\sim}13\%$ in global positional error beyond training radii, yet local bond fidelity diverges wildly across architectures -- from near-zero to over $2\times$ collapse; (ii) no two architectures share the same failure sequence, revealing the frontier as a multi-dimensional surface shaped by model family; and (iii) well-behaved models obey a power-law scaling exponent $\alpha \approx 1/3$ whose in-distribution fit accurately predicts out-of-distribution error, making their frontiers quantitatively forecastable. These findings establish output scale as a first-class evaluation axis for geometric generative models. The dataset and code are available at https://github.com/KurbanIntelligenceLab/RADII.
Authors: Shu-Ting Pi, Pradeep Bagavan, Yejia Li, Disha, Qun Liu
Abstract: Utilizing Large Language Models (LLM) as chatbots in diverse business scenarios often presents the challenge of maintaining topic continuity. Abrupt shifts in topics can lead to poor user experiences and inefficient utilization of computational resources. In this paper, we present a topic continuity model aimed at assessing whether a response aligns with the initial conversation topic. Our model is built upon the expansion of the corresponding natural language understanding (NLU) model into quantifiable terms using a Naive Bayes approach. Subsequently, we have introduced an attention mechanism and logarithmic nonlinearity to enhance its capability to capture topic continuity. This approach allows us to convert the NLU model into an interpretable analytical formula. In contrast to many NLU models constrained by token limits, our proposed model can seamlessly handle conversations of any length with linear time complexity. Furthermore, the attention mechanism significantly improves the model's ability to identify topic continuity in complex conversations. According to our experiments, our model consistently outperforms traditional methods, particularly in handling lengthy and intricate conversations. This unique capability offers us an opportunity to ensure the responsible and interpretable use of LLMs.
Authors: Mykola Khandoga, Rui Yuan, Vinay Kumar Sankarapu
Abstract: Policy gradient methods for language model reasoning, such as GRPO and DAPO, assign uniform credit to all generated tokens - the filler phrase "Let me think" receives the same gradient update as the critical calculation "23 + 45 = 68." We propose counterfactual importance weighting: mask reasoning spans, measure the drop in answer probability, and upweight tokens accordingly during policy gradient updates. Our method requires no auxiliary models or external annotation, instead importance is estimated directly from the policy model's own probability shifts. Experiments on GSM8K across three models spanning the Qwen and Llama families demonstrate consistent improvements over uniform baselines and faster convergence to equivalent accuracy. Inverting the importance signal hurts performance, confirming we capture genuine causal structure rather than noise. Analysis shows the method correctly prioritizes calculation steps over scaffolding text. We view these findings as establishing counterfactual importance weighting as a foundation for further research rather than a complete solution.
Authors: Waris Quamer, Mu-Ruei Tseng, Ghady Nasrallah, Ricardo Gutierrez-Osuna
Abstract: Real-time voice conversion and speaker anonymization require causal, low-latency synthesis without sacrificing intelligibility or naturalness. Current systems have a core representational mismatch: content is time-varying, while speaker identity is injected as a static global embedding. We introduce a streamable speech synthesizer that aligns the temporal granularity of identity and content via a content-synchronous, time-varying timbre (TVT) representation. A Global Timbre Memory expands a global timbre instance into multiple compact facets; frame-level content attends to this memory, a gate regulates variation, and spherical interpolation preserves identity geometry while enabling smooth local changes. In addition, a factorized vector-quantized bottleneck regularizes content to reduce residual speaker leakage. The resulting system is streamable end-to-end, with <80 ms GPU latency. Experiments show improvements in naturalness, speaker transfer, and anonymization compared to SOTA streaming baselines, establishing TVT as a scalable approach for privacy-preserving and expressive speech synthesis under strict latency budgets.
Authors: Seyed Morteza Emadi
Abstract: Manufacturing lines, service journeys, supply chains, and AI reasoning chains share a common challenge: attributing a terminal outcome to the intermediate stage that caused it. We establish an information-theoretic barrier to this credit assignment problem: the signal connecting early steps to final outcomes decays exponentially with depth, creating a critical horizon beyond which no algorithm can learn from endpoint data alone. We prove four results. First, a Signal Decay Bound: sample complexity for attributing outcomes to early stages grows exponentially in the number of intervening steps. Second, Width Limits: parallel rollouts provide only logarithmic relief, with correlation capping the effective number of independent samples. Third, an Objective Mismatch: additive reward aggregation optimizes the wrong quantity when sequential validity requires all steps to be correct. Fourth, Optimal Inspection Design: uniform checkpoint spacing is minimax-optimal under homogeneous signal attenuation, while a greedy algorithm yields optimal non-uniform schedules under heterogeneous attenuation. Together, these results provide a common analytical foundation for inspection design in operations and supervision design in AI.
Authors: Chen Cheng, Rina Foygel Barber
Abstract: We examine the connection between training error and generalization error for arbitrary estimating procedures, working in an overparameterized linear model under general priors in a Bayesian setup. We find determining factors inherent to the prior distribution $\pi$, giving explicit conditions under which optimal generalization necessitates that the training error be (i) near interpolating relative to the noise size (i.e., memorization is necessary), or (ii) close to the noise level (i.e., overfitting is harmful). Remarkably, these phenomena occur when the noise reaches thresholds determined by the Fisher information and the variance parameters of the prior $\pi$.
Authors: Xinyu Wang, Ke Deng, Fei Dou, Jinbo Bi, Jin Lu
Abstract: Model merging aims to combine multiple fine-tuned models into a single multi-task model without access to training data. Existing task-vector merging methods such as TIES, TSV-M, and Iso-C/CTS differ in their aggregation rules but treat all layers nearly uniformly. This assumption overlooks the strong layer-wise heterogeneity in large vision transformers, where shallow layers are sensitive to interference while deeper layers encode stable task-specific features. We introduce LARV, a training-free, data-free, merger-agnostic Layer-wise Adaptive Rescaling Veneer that plugs into any task-vector merger and assigns a per-layer scale to each task vector before aggregation, and show it consistently boosts diverse merging rules. LARV adaptively suppresses shallow-layer interference and amplifies deeper-layer alignment using a simple deterministic schedule, requiring no retraining or modification to existing mergers. To our knowledge, this is the first work to perform layer-aware scaling for task-vector merging. LARV computes simple data-free layer proxies and turns them into scales through a lightweight rule; we study several instantiations within one framework (e.g., tiered two/three-level scaling with fixed values, or continuous mappings) and show that tiered choices offer the best robustness, while continuous mappings remain an ablation. LARV is orthogonal to the base merger and adds negligible cost. On FusionBench with Vision Transformers, LARV consistently improves all task-vector baselines across 8/14/20-task settings; for example, Iso-C + LARV reaches 85.9% on ViT-B/32, 89.2% on ViT-B/16, and 92.6% on ViT-L/14. Layerwise analysis and corruption tests further indicate that LARV suppresses shallow-layer interference while modestly amplifying deeper, task-stable features, turning model merging into a robust, layer-aware procedure rather than a uniform one.
Authors: Yiqiao Li, Bo Shang, Jie Wei
Abstract: Fine-grained truck classification is critical for intelligent transportation systems (ITS), yet current LiDAR-based methods face scalability challenges due to their reliance on supervised deep learning and labor-intensive manual annotation. Vision-Language Models (VLMs) offer promising few-shot generalization, but their application to roadside LiDAR is limited by a modality gap between sparse 3D point clouds and dense 2D imagery. We propose a framework that bridges this gap by adapting off-the-shelf VLMs for fine-grained truck classification without parameter fine-tuning. Our new depth-aware image generation pipeline applies noise removal, spatial and temporal registration, orientation rectification, morphological operations, and anisotropic smoothing to transform sparse, occluded LiDAR scans into depth-encoded 2D visual proxies. Validated on a real-world dataset of 20 vehicle classes, our approach achieves competitive classification accuracy with as few as 16-30 examples per class, offering a scalable alternative to data-intensive supervised baselines. We further observe a "Semantic Anchor" effect: text-based guidance regularizes performance in ultra-low-shot regimes $k < 4$, but degrades accuracy in more-shot settings due to semantic mismatch. Furthermore, we demonstrate the efficacy of this framework as a Cold Start strategy, using VLM-generated labels to bootstrap lightweight supervised models. Notably, the few-shot VLM-based model achieves over correct classification rate of 75 percent for specific drayage categories (20ft, 40ft, and 53ft containers) entirely without the costly training or fine-tuning, significantly reducing the intensive demands of initial manual labeling, thus achieving a method of practical use in ITS applications.
Authors: Mohammad Masudur Rahman, Md. Rashedur Rahman, Ashraful Islam, Saadia B Alam, M Ashraful Amin
Abstract: Urban Visual Pollution (UVP) has emerged as a critical concern, yet research on automatic detection and application remains fragmented. This scoping review maps the existing deep learning-based approaches for detecting, classifying, and designing a comprehensive application framework for visual pollution management. Following the PRISMA-ScR guidelines, seven academic databases (Scopus, Web of Science, IEEE Xplore, ACM DL, ScienceDirect, SpringerNatureLink, and Wiley) were systematically searched and reviewed, and 26 articles were found. Most research focuses on specific pollutant categories and employs variations of YOLO, Faster R-CNN, and EfficientDet architectures. Although several datasets exist, they are limited to specific areas and lack standardized taxonomies. Few studies integrate detection into real-time application systems, yet they tend to be geographically skewed. We proposed a framework for monitoring visual pollution that integrates a visual pollution index to assess the severity of visual pollution for a certain area. This review highlights the need for a unified UVP management system that incorporates pollutant taxonomy, a cross-city benchmark dataset, a generalized deep learning model, and an assessment index that supports sustainable urban aesthetics and enhances the well-being of urban dwellers.
Authors: Xincan Feng, Noriki Nishida, Yusuke Sakai, Yuji Matsumoto
Abstract: Prior work reports conflicting results on query diversity in synthetic data generation for dense retrieval. We identify this conflict and design Q-D metrics to quantify diversity's impact, making the problem measurable. Through experiments on 4 benchmark types (31 datasets), we find query diversity especially benefits multi-hop retrieval. Deep analysis on multi-hop data reveals that diversity benefit correlates strongly with query complexity ($r$$\geq$0.95, $p$$<$0.05 in 12/14 conditions), measured by content words (CW). We formalize this as the Complexity-Diversity Principle (CDP): query complexity determines optimal diversity. CDP provides actionable thresholds (CW$>$10: use diversity; CW$<$7: avoid it). Guided by CDP, we propose zero-shot multi-query synthesis for multi-hop tasks, achieving state-of-the-art performance.
Authors: Haoran Sun, Xuanzhi Xia, Xu Chu, Xiaotie Deng
Abstract: Affine Maximizer Auctions (AMAs), a generalized mechanism family from VCG, are widely used in automated mechanism design due to their inherent dominant-strategy incentive compatibility (DSIC) and individual rationality (IR). However, as the payment form is fixed, AMA's expressiveness is restricted, especially in distributions where bidders' valuations are correlated. In this paper, we propose Correlation-Aware AMA (CA-AMA), a novel framework that augments AMA with a new correlation-aware payment. We show that any CA-AMA preserves the DSIC property and formalize finding optimal CA-AMA as a constraint optimization problem subject to the IR constraint. Then, we theoretically characterize scenarios where classic AMAs can perform arbitrarily poorly compared to the optimal revenue, while the CA-AMA can reach the optimal revenue. For optimizing CA-AMA, we design a practical two-stage training algorithm. We derive that the target function's continuity and the generalization bound on the degree of deviation from strict IR. Finally, extensive experiments showcase that our algorithm can find an approximate optimal CA-AMA in various distributions with improved revenue and a low degree of violation of IR.
Authors: Shinsaku Sakaue, Yuichi Yoshida
Abstract: We study online learning in the random-order model, where the multiset of loss functions is chosen adversarially but revealed in a uniformly random order. Building on the batch-to-online conversion by Dong and Yoshida (2023), we show that if an offline algorithm admits a $(1+\varepsilon)$-approximation guarantee and the effect of $\varepsilon$ on its average sensitivity is characterized by a function $\varphi(\varepsilon)$, then an adaptive choice of $\varepsilon$ yields a small-loss regret bound of $\tilde O(\varphi^{\star}(\mathrm{OPT}_T))$, where $\varphi^{\star}$ is the concave conjugate of $\varphi$, $\mathrm{OPT}_T$ is the offline optimum over $T$ rounds, and $\tilde O$ hides polylogarithmic factors in $T$. Our method requires no regularity assumptions on loss functions, such as smoothness, and can be viewed as a generalization of the AdaGrad-style tuning applied to the approximation parameter $\varepsilon$. Our result recovers and strengthens the $(1+\varepsilon)$-approximate regret bounds of Dong and Yoshida (2023) and yields small-loss regret bounds for online $k$-means clustering, low-rank approximation, and regression. We further apply our framework to online submodular function minimization using $(1\pm\varepsilon)$-cut sparsifiers of submodular hypergraphs, obtaining a small-loss regret bound of $\tilde O(n^{3/4}(1 + \mathrm{OPT}_T^{3/4}))$, where $n$ is the ground-set size. Our approach sheds light on the power of sparsification and related techniques in establishing small-loss regret bounds in the random-order model.
Authors: James Burgess, Rameen Abdal, Dan Stoddart, Sergey Tulyakov, Serena Yeung-Levy, Kuan-Chieh Jackson Wang
Abstract: Modern image generators produce strikingly realistic images, where only artifacts like distorted hands or warped objects reveal their synthetic origin. Detecting these artifacts is essential: without detection, we cannot benchmark generators or train reward models to improve them. Current detectors fine-tune VLMs on tens of thousands of labeled images, but this is expensive to repeat whenever generators evolve or new artifact types emerge. We show that pretrained VLMs already encode the knowledge needed to detect artifacts - with the right scaffolding, this capability can be unlocked using only a few hundred labeled examples per artifact category. Our system, ArtifactLens, achieves state-of-the-art on five human artifact benchmarks (the first evaluation across multiple datasets) while requiring orders of magnitude less labeled data. The scaffolding consists of a multi-component architecture with in-context learning and text instruction optimization, with novel improvements to each. Our methods generalize to other artifact types - object morphology, animal anatomy, and entity interactions - and to the distinct task of AIGC detection.
Authors: Vid Kocijan, Jinu Sunil, Jan Eric Lenssen, Viman Deb, Xinwei Xe, Federco Reyes Gomez, Matthias Fey, Jure Leskovec
Abstract: The purpose of predictive modeling on relational data is to predict future or missing values in a relational database, for example, future purchases of a user, risk of readmission of the patient, or the likelihood that a financial transaction is fraudulent. Typically powered by machine learning methods, predictive models are used in recommendations, financial fraud detection, supply chain optimization, and other systems, providing billions of predictions every day. However, training a machine learning model requires manual work to extract the required training examples - prediction entities and target labels - from the database, which is slow, laborious, and prone to mistakes. Here, we present the Predictive Query Language (PQL), a SQL-inspired declarative language for defining predictive tasks on relational databases. PQL allows specifying a predictive task in a single declarative query, enabling the automatic computation training labels for a large variety of machine learning tasks, such as regression, classification, time-series forecasting, and recommender systems. PQL is already successfully integrated and used in a collection of use cases as part of a predictive AI platform. The versatility of the language can be demonstrated through its many ongoing use cases, including financial fraud, item recommendations, and workload prediction. We demonstrate its versatile design through two implementations; one for small-scale, low-latency use and one that can handle large-scale databases.
Authors: Sora Miyamoto, Daisuke Oba, Naoaki Okazaki
Abstract: Tree-search decoding is an effective form of test-time scaling for large language models (LLMs), but real-world deployment imposes a fixed per-query token budget that varies across settings. Existing tree-search policies are largely budget-agnostic, treating the budget as a termination condition, which can lead to late-stage over-branching or premature termination. We propose {Budget-Guided MCTS} (BG-MCTS), a tree-search decoding algorithm that aligns its search policy with the remaining token budget: it starts with broad exploration, then prioritizes refinement and answer completion as the budget depletes while reducing late-stage branching from shallow nodes. BG-MCTS consistently outperforms budget-agnostic tree-search baselines across different budgets on MATH500 and AIME24/25 with open-weight LLMs.
Authors: Chenyu Yang, Denis Tarasov, Davide Liconti, Hehui Zheng, Robert K. Katzschmann
Abstract: Real-world fine-tuning of dexterous manipulation policies remains challenging due to limited real-world interaction budgets and highly multimodal action distributions. Diffusion-based policies, while expressive, do not permit conservative likelihood-based updates during fine-tuning because action probabilities are intractable. In contrast, conventional Gaussian policies collapse under multimodality, particularly when actions are executed in chunks, and standard per-step critics fail to align with chunked execution, leading to poor credit assignment. We present SOFT-FLOW, a sample-efficient off-policy fine-tuning framework with normalizing flow (NF) to address these challenges. The normalizing flow policy yields exact likelihoods for multimodal action chunks, allowing conservative, stable policy updates through likelihood regularization and thereby improving sample efficiency. An action-chunked critic evaluates entire action sequences, aligning value estimation with the policy's temporal structure and improving long-horizon credit assignment. To our knowledge, this is the first demonstration of a likelihood-based, multimodal generative policy combined with chunk-level value learning on real robotic hardware. We evaluate SOFT-FLOW on two challenging dexterous manipulation tasks in the real world: cutting tape with scissors retrieved from a case, and in-hand cube rotation with a palm-down grasp -- both of which require precise, dexterous control over long horizons. On these tasks, SOFT-FLOW achieves stable, sample-efficient adaptation where standard methods struggle.
Authors: Daisuke Nohara, Taishi Nakamura, Rio Yokota
Abstract: Reinforcement learning substantially improves reasoning in large language models, but it also tends to lengthen chain of thought outputs and increase computational cost during both training and inference. Though length control methods have been proposed, it remains unclear what the optimal output length is for balancing efficiency and performance. In this work, we compare several length control methods on two models, Qwen3-1.7B Base and DeepSeek-R1-Distill-Qwen-1.5B. Our results indicate that length penalties may hinder reasoning acquisition, while properly tuned length control can improve efficiency for models with strong prior reasoning. By extending prior work to RL trained policies, we identify two failure modes, 1) long outputs increase dispersion, and 2) short outputs lead to under-thinking.
Authors: Tobias W\"ohrer, Christian Kuehn
Abstract: We investigate finite-time Lyapunov exponents (FTLEs), a measure for exponential separation of input perturbations, of deep neural networks within the framework of continuous-depth neural ODEs. We demonstrate that FTLEs are powerful organizers for input-output dynamics, allowing for better interpretability and the comparison of distinct model architectures. We establish a direct connection between Lyapunov exponents and adversarial vulnerability, and propose a novel training algorithm that improves robustness by FTLE regularization. The key idea is to suppress exponents far from zero in the early stage of the input dynamics. This approach enhances robustness and reduces computational cost compared to full-interval regularization, as it avoids a full ``double'' backpropagation.
Authors: R E Zera Marveen Lyngkhoi, Chirag Chawla, Pratinav Seth, Utsav Avaiya, Soham Bhattacharjee, Mykola Khandoga, Rui Yuan, Vinay Kumar Sankarapu
Abstract: Post-training alignment is central to deploying large language models (LLMs), yet practical workflows remain split across backend-specific tools and ad-hoc glue code, making experiments hard to reproduce. We identify backend interference, reward fragmentation, and irreproducible pipelines as key obstacles in alignment research. We introduce AlignTune, a modular toolkit exposing a unified interface for supervised fine-tuning (SFT) and RLHF-style optimization with interchangeable TRL and Unsloth backends. AlignTune standardizes configuration, provides an extensible reward layer (rule-based and learned), and integrates evaluation over standard benchmarks and custom tasks. By isolating backend-specific logic behind a single factory boundary, AlignTune enables controlled comparisons and reproducible alignment experiments.
Authors: Florian Handke, Dejan Stan\v{c}evi\'c, Felix Koulischer, Thomas Demeester, Luca Ambrogioni
Abstract: Diffusion models do not recover semantic structure uniformly over time. Instead, samples transition from semantic ambiguity to class commitment within a narrow regime. Recent theoretical work attributes this transition to dynamical instabilities along class-separating directions, but practical methods to detect and exploit these windows in trained models are still limited. We show that tracking the class-conditional entropy of a latent semantic variable given the noisy state provides a reliable signature of these transition regimes. By restricting the entropy to semantic partitions, the entropy can furthermore resolve semantic decisions at different levels of abstraction. We analyze this behavior in high-dimensional Gaussian mixture models and show that the entropy rate concentrates on the same logarithmic time scale as the speciation symmetry-breaking instability previously identified in variance-preserving diffusion. We validate our method on EDM2-XS and Stable Diffusion 1.5, where class-conditional entropy consistently isolates the noise regimes critical for semantic structure formation. Finally, we use our framework to quantify how guidance redistributes semantic information over time. Together, these results connect information-theoretic and statistical physics perspectives on diffusion and provide a principled basis for time-localized control.
Authors: Joseph Attieh, Timothee Mickus, Anne-Laure Ligozat, Aur\'elie N\'ev\'eol, J\"org Tiedemann
Abstract: Knowledge distillation (KD) is a tool to compress a larger system (teacher) into a smaller one (student). In machine translation, studies typically report only the translation quality of the student and omit the computational complexity of performing KD, making it difficult to select among the many available KD choices under compute-induced constraints. In this study, we evaluate representative KD methods by considering both translation quality and computational cost. We express computational cost as a carbon footprint using the machine learning life cycle assessment (MLCA) tool. This assessment accounts for runtime operational emissions and amortized hardware production costs throughout the KD model life cycle (teacher training, distillation, and inference). We find that (i) distillation overhead dominates the total footprint at small deployment volumes, (ii) inference dominates at scale, making KD beneficial only beyond a task-dependent usage threshold, and (iii) word-level distillation typically offers more favorable footprint-quality trade-offs than sequence-level distillation. Our protocol provides reproducible guidance for selecting KD methods under explicit quality and compute-induced constraints.
Authors: Jialiang Tang, Jialin Zhang, Xiaoming Sun
Abstract: Quantum machine learning (QML), as an interdisciplinary field bridging quantum computing and machine learning, has garnered significant attention in recent years. Currently, the field as a whole faces challenges due to incomplete theoretical foundations for the expressivity of quantum neural networks (QNNs). In this paper we propose a constructive QNN model and demonstrate that it possesses the universal approximation property (UAP), which means it can approximate any square-integrable function up to arbitrary accuracy. Furthermore, it supports switching function bases, thus adaptable to various scenarios in numerical approximation and machine learning. Our model has asymptotic advantages over the best classical feed-forward neural networks in terms of circuit size and achieves optimal parameter complexity when approximating Sobolev functions under $L_2$ norm.
Authors: Pablo Garc\'ia-Santaclara, Bruno Fern\'andez-Castro, RebecaP. D\'iaz-Redondo, Mart\'in Alonso-Gamarra
Abstract: Data streams are rarely static in dynamic environments like Industry 4.0. Instead, they constantly change, making traditional offline models outdated unless they can quickly adjust to the new data. This need can be adequately addressed by continual learning (CL), which allows systems to gradually acquire knowledge without incurring the prohibitive costs of retraining them from scratch. Most research on continual learning focuses on classification problems, while very few studies address regression tasks. We propose the first prototype-based generative replay framework designed for online task-free continual regression. Our approach defines an adaptive output-space discretization model, enabling prototype-based generative replay for continual regression without storing raw data. Evidence obtained from several benchmark datasets shows that our framework reduces forgetting and provides more stable performance than other state-of-the-art solutions.
Authors: Laura Paul, Holger Rauhut, Martin Burger, Samira Kabri, Tim Roith
Abstract: Recent advances in imaging technologies, deep learning and numerical performance have enabled non-invasive detailed analysis of artworks, supporting their documentation and conservation. In particular, automated detection of craquelure in digitized paintings is crucial for assessing degradation and guiding restoration, yet remains challenging due to the possibly complex scenery and the visual similarity between cracks and crack-like artistic features such as brush strokes or hair. We propose a hybrid approach that models crack detection as an inverse problem, decomposing an observed image into a crack-free painting and a crack component. A deep generative model is employed as powerful prior for the underlying artwork, while crack structures are captured using a Mumford--Shah-type variational functional together with a crack prior. Joint optimization yields a pixel-level map of crack localizations in the painting.
Authors: Daan Otto, Jannis Kurtz, Dick den Hertog, Ilker Birbil
Abstract: In model extraction attacks, the goal is to reveal the parameters of a black-box machine learning model by querying the model for a selected set of data points. Due to an increasing demand for explanations, this may involve counterfactual queries besides the typically considered factual queries. In this work, we consider linear models and three types of queries: factual, counterfactual, and robust counterfactual. First, for an arbitrary set of queries, we derive novel mathematical formulations for the classification regions for which the decision of the unknown model is known, without recovering any of the model parameters. Second, we derive bounds on the number of queries needed to extract the model's parameters for (robust) counterfactual queries under arbitrary norm-based distances. We show that the full model can be recovered using just a single counterfactual query when differentiable distance measures are employed. In contrast, when using polyhedral distances for instance, the number of required queries grows linearly with the dimension of the data space. For robust counterfactuals, the latter number of queries doubles. Consequently, the applied distance function and robustness of counterfactuals have a significant impact on the model's security.
Authors: Kawtar Zaher, Ilyass Moummad, Olivier Buisson, Alexis Joly
Abstract: Most self-supervised learning (SSL) methods learn continuous visual representations by aligning different views of the same input, offering limited control over how information is structured across representation dimensions. In this work, we frame visual self-supervised learning as a discrete communication process between a teacher and a student network, where semantic information is transmitted through a fixed-capacity binary channel. Rather than aligning continuous features, the student predicts multi-label binary messages produced by the teacher. Discrete agreement is enforced through an element-wise binary cross-entropy objective, while a coding-rate regularization term encourages effective utilization of the constrained channel, promoting structured representations. We further show that periodically reinitializing the projection head strengthens this effect by encouraging embeddings that remain predictive across multiple discrete encodings. Extensive experiments demonstrate consistent improvements over continuous agreement baselines on image classification, retrieval, and dense visual prediction tasks, as well as under domain shift through self-supervised adaptation. Beyond backbone representations, we analyze the learned binary codes and show that they form a compact and informative discrete language, capturing semantic factors reusable across classes.
Authors: Vladimir R. Kostic, Karim Lounici, Massimiliano Pontil
Abstract: We introduce a Toeplitz-based framework for data-driven spectral estimation of linear evolution operators in dynamical systems. Focusing on transfer and Koopman operators from equilibrium trajectories without access to the underlying equations of motion, our method applies Toeplitz filters to the infinitesimal generator to extract eigenvalues, eigenfunctions, and spectral measures. Structural prior knowledge, such as self-adjointness or skew-symmetry, can be incorporated by design. The approach is statistically consistent and computationally efficient, leveraging both primal and dual algorithms commonly used in statistical learning. Numerical experiments on deterministic and chaotic systems demonstrate that the framework can recover spectral properties beyond the reach of standard data-driven methods.
Authors: Daniel Kaiser, Arnoldo Frigessi, Ali Ramezani-Kebrya, Benjamin Ricaud
Abstract: Large language models trained for reasoning trade off inference tokens against accuracy, yet standard evaluations report only final accuracy, obscuring where tokens are spent or wasted. We introduce a trace-optional framework that decomposes token efficiency into interpretable factors: completion under a fixed token budget (avoiding truncation), conditional correctness given completion, and verbosity (token usage). When benchmark metadata provides per-instance workload proxies, we further factor verbosity into two components: mean verbalization overhead (tokens per work unit) and a coupling coefficient capturing how overhead scales with task workload. When reasoning traces are available, we add deterministic trace-quality measures (grounding, repetition, prompt copying) to separate degenerate looping from verbose-but-engaged reasoning, avoiding human labeling and LLM judges. Evaluating 25 models on CogniLoad, we find that accuracy and token-efficiency rankings diverge (Spearman $\rho=0.63$), efficiency gaps are often driven by conditional correctness, and verbalization overhead varies by about 9 times (only weakly related to model scale). Our decomposition reveals distinct bottleneck profiles that suggest different efficiency interventions.
Authors: Emanuel Figetakis, Ahmed Refaey Hussein
Abstract: The convergence of AI and 6G network automation introduces new challenges in maintaining transparency, fairness, and accountability across multivendor management systems. Although closed-loop AI orchestration improves adaptability and self-optimization, it also creates a responsibility gap, where violations of SLAs cannot be causally attributed to specific agents or vendors. This paper presents a hybrid responsible AI-stochastic learning framework that embeds fairness, robustness, and auditability directly into the network control loop. The framework integrates RAI games with stochastic optimization, enabling dynamic adversarial reweighting and probabilistic exploration across heterogeneous vendor domains. An RAAP continuously records AI-driven decision trajectories and produces dual accountability reports: user-level SLA summaries and operator-level responsibility analytics. Experimental evaluations on synthetic two-class multigroup datasets demonstrate that the proposed hybrid model improves the accuracy of the worst group by up to 10.5\%. Specifically, hybrid RAI achieved a WGAcc of 60.5\% and an AvgAcc of 72.7\%, outperforming traditional RAI-GA (50.0\%) and ERM (21.5\%). The audit mechanism successfully traced 99\% simulated SLA violations to the AI entities responsible, producing both vendor and agent-level accountability indices. These results confirm that the proposed hybrid approach enhances fairness and robustness as well as establishes a concrete accountability framework for autonomous SLA assurance in multivendor 6G networks.
Authors: Fabian Schaipp, Robert M. Gower, Adrien Taylor
Abstract: We present a theoretical analysis of stochastic optimization methods in terms of their sensitivity with respect to the step size. We identify a key quantity that, for each method, describes how the performance degrades as the step size becomes too large. For convex problems, we show that this quantity directly impacts the suboptimality bound of the method. Most importantly, our analysis provides direct theoretical evidence that adaptive step-size methods, such as SPS or NGN, are more robust than SGD. This allows us to quantify the advantage of these adaptive methods beyond empirical evaluation. Finally, we show through experiments that our theoretical bound qualitatively mirrors the actual performance as a function of the step size, even for nonconvex problems.
Authors: Alireza Tabarraei
Abstract: Conditional Value-at-Risk (CVaR) is a central tail-risk measure in stochastic structural mechanics, yet its accurate evaluation under high-dimensional, spatially correlated material uncertainty remains computationally prohibitive for classical Monte Carlo methods. Leveraging bounded-expectation reformulations of CVaR compatible with quantum amplitude estimation, we develop a quantum-enhanced inference framework that casts CVaR evaluation as a statistically consistent, confidence-constrained maximum-likelihood amplitude estimation problem. The proposed method extends iterative quantum amplitude estimation (IQAE) by embedding explicit maximum-likelihood inference within a rigorously controlled interval-tracking architecture. To ensure global correctness under finite-shot noise and the non-injective oscillatory response induced by Grover amplification, we introduce a stabilized inference scheme incorporating multi-hypothesis feasibility tracking, periodic low-depth disambiguation, and a bounded restart mechanism governed by an explicit failure-probability budget. This formulation preserves the quadratic oracle-complexity advantage of amplitude estimation while providing finite-sample confidence guarantees and reduced estimator variance. The framework is demonstrated on benchmark problems with spatially correlated lognormal Young's modulus fields generated using a Nystrom low-rank Gaussian kernel model. Numerical results show that the proposed estimator achieves substantially lower oracle complexity than classical Monte Carlo CVaR estimation at comparable confidence levels, while maintaining rigorous statistical reliability. This work establishes a practically robust and theoretically grounded quantum-enhanced methodology for tail-risk quantification in stochastic continuum mechanics.
Authors: Shixiong Wang, Wei Dai, Li-Chun Wang, Geoffrey Ye Li
Abstract: This tutorial-style overview article examines the fundamental principles and methods of robustness, using wireless sensing and communication (WSC) as the narrative and exemplifying framework. First, we formalize the conceptual and mathematical foundations of robustness, highlighting the interpretations and relations across robust statistics, optimization, and machine learning. Key techniques, such as robust estimation and testing, distributionally robust optimization, and regularized and adversary training, are investigated. Together, the costs of robustness in system design, for example, the compromised nominal performances and the extra computational burdens, are discussed. Second, we review recent robust signal processing solutions for WSC that address model mismatch, data scarcity, adversarial perturbation, and distributional shift. Specific applications include robust ranging-based localization, modality sensing, channel estimation, receive combining, waveform design, and federated learning. Through this effort, we aim to introduce the classical developments and recent advances in robustness theory to the general signal processing community, exemplifying how robust statistical, optimization, and machine learning approaches can address the uncertainties inherent in WSC systems.
Authors: Shih-Lun Wu, Ge Zhu, Juan-Pablo Caceres, Cheng-Zhi Anna Huang, Nicholas J. Bryan
Abstract: Music stem generation, the task of producing musically-synchronized and isolated instrument audio clips, offers the potential of greater user control and better alignment with musician workflows compared to conventional text-to-music models. Existing stem generation approaches, however, either rely on fixed architectures that output a predefined set of stems in parallel, or generate only one stem at a time, resulting in slow inference despite flexibility in stem combination. We propose Stemphonic, a diffusion-/flow-based framework that overcomes this trade-off and generates a variable set of synchronized stems in one inference pass. During training, we treat each stem as a batch element, group synchronized stems in a batch, and apply a shared noise latent to each group. At inference-time, we use a shared initial noise latent and stem-specific text inputs to generate synchronized multi-stem outputs in one pass. We further expand our approach to enable one-pass conditional multi-stem generation and stem-wise activity controls to empower users to iteratively generate and orchestrate the temporal layering of a mix. We benchmark our results on multiple open-source stem evaluation sets and show that Stemphonic produces higher-quality outputs while accelerating the full mix generation process by 25 to 50%. Demos at: https://stemphonic-demo.vercel.app.
Authors: Rafid Mahmood
Abstract: To mitigate the trade-offs between performance and costs, LLM providers route user tasks to different models based on task difficulty and latency. We study the effect of LLM routing with respect to user behavior. We propose a game between an LLM provider with two models (standard and reasoning) and a user who can re-prompt or abandon tasks if the routed model cannot solve them. The user's goal is to maximize their utility minus the delay from using the model, while the provider minimizes the cost of servicing the user. We solve this Stackelberg game by fully characterizing the user best response and simplifying the provider problem. We observe that in nearly all cases, the optimal routing policy involves a static policy with no cascading that depends on the expected utility of the models to the user. Furthermore, we reveal a misalignment gap between the provider-optimal and user-preferred routes when the user's and provider's rankings of the models with respect to utility and cost differ. Finally, we demonstrate conditions for extreme misalignment where providers are incentivized to throttle the latency of the models to minimize their costs, consequently depressing user utility. The results yield simple threshold rules for single-provider, single-user interactions and clarify when routing, cascading, and throttling help or harm.
Authors: William Lugoloobi, Thomas Foster, William Bankes, Chris Russell
Abstract: Running LLMs with extended reasoning on every problem is expensive, but determining which inputs actually require additional compute remains challenging. We investigate whether their own likelihood of success is recoverable from their internal representations before generation, and if this signal can guide more efficient inference. We train linear probes on pre-generation activations to predict policy-specific success on math and coding tasks, substantially outperforming surface features such as question length and TF-IDF. Using E2H-AMC, which provides both human and model performance on identical problems, we show that models encode a model-specific notion of difficulty that is distinct from human difficulty, and that this distinction increases with extended reasoning. Leveraging these probes, we demonstrate that routing queries across a pool of models can exceed the best-performing model whilst reducing inference cost by up to 70\% on MATH, showing that internal representations enable practical efficiency gains even when they diverge from human intuitions about difficulty. Our code is available at: https://github.com/KabakaWilliam/llms_know_difficulty
Authors: Roy R. Lederman, David Silva-S\'anchez, Ziling Chen, Gilles Mordant, Amnon Balanov, Tamir Bendory
Abstract: Lloyd's k-means algorithm is one of the most widely used clustering methods. We prove that in high-dimensional, high-noise settings, the algorithm exhibits catastrophic failure: with high probability, essentially every partition of the data is a fixed point. Consequently, Lloyd's algorithm simply returns its initial partition - even when the underlying clusters are trivially recoverable by other methods. In contrast, we prove that Hartigan's k-means algorithm does not exhibit this pathology. Our results show the stark difference between these algorithms and offer a theoretical explanation for the empirical difficulties often observed with k-means in high dimensions.
Authors: Hugo Latourelle-Vigeant, Theodor Misiakiewicz
Abstract: We study the problem of learning multi-index models (MIMs), where the label depends on the input $\boldsymbol{x} \in \mathbb{R}^d$ only through an unknown $\mathsf{s}$-dimensional projection $\boldsymbol{W}_*^\mathsf{T} \boldsymbol{x} \in \mathbb{R}^\mathsf{s}$. Exploiting the equivariance of this problem under the orthogonal group $\mathcal{O}_d$, we obtain a sharp harmonic-analytic characterization of the learning complexity for MIMs with spherically symmetric inputs -- which refines and generalizes previous Gaussian-specific analyses. Specifically, we derive statistical and computational complexity lower bounds within the Statistical Query (SQ) and Low-Degree Polynomial (LDP) frameworks. These bounds decompose naturally across spherical harmonic subspaces. Guided by this decomposition, we construct a family of spectral algorithms based on harmonic tensor unfolding that sequentially recover the latent directions and (nearly) achieve these SQ and LDP lower bounds. Depending on the choice of harmonic degree sequence, these estimators can realize a broad range of trade-offs between sample and runtime complexity. From a technical standpoint, our results build on the semisimple decomposition of the $\mathcal{O}_d$-action on $L^2 (\mathbb{S}^{d-1})$ and the intertwining isomorphism between spherical harmonics and traceless symmetric tensors.
Authors: Calvin Yeung, Ali Zakeri, Zhuowen Zou, Mohsen Imani
Abstract: Many visual scenes can be described as compositions of latent factors. Effective recognition, reasoning, and editing often require not only forming such compositional representations, but also solving the decomposition problem. One popular choice for constructing these representations is through the binding operation. Resonator networks, which can be understood as coupled Hopfield networks, were proposed as a way to perform decomposition on such bound representations. Recent works have shown notable similarities between Hopfield networks and diffusion models. Motivated by these observations, we introduce a framework for semantic decomposition using coupled inference in diffusion models. Our method frames semantic decomposition as an inverse problem and couples the diffusion processes using a reconstruction-driven guidance term that encourages the composition of factor estimates to match the bound vector. We also introduce a novel iterative sampling scheme that improves the performance of our model. Finally, we show that attention-based resonator networks are a special case of our framework. Empirically, we demonstrate that our coupled inference framework outperforms resonator networks across a range of synthetic semantic decomposition tasks.
Authors: Kerri Lu, Dan M. Kluger, Stephen Bates, Sherrie Wang
Abstract: Current instance segmentation models achieve high performance on average predictions, but lack principled uncertainty quantification: their outputs are not calibrated, and there is no guarantee that a predicted mask is close to the ground truth. To address this limitation, we introduce a conformal prediction algorithm to generate adaptive confidence sets for instance segmentation. Given an image and a pixel coordinate query, our algorithm generates a confidence set of instance predictions for that pixel, with a provable guarantee for the probability that at least one of the predictions has high Intersection-Over-Union (IoU) with the true object instance mask. We apply our algorithm to instance segmentation examples in agricultural field delineation, cell segmentation, and vehicle detection. Empirically, we find that our prediction sets vary in size based on query difficulty and attain the target coverage, outperforming existing baselines such as Learn Then Test, Conformal Risk Control, and morphological dilation-based methods. We provide versions of the algorithm with asymptotic and finite sample guarantees.
Authors: Laura Ib\'a\~nez-Mart\'inez, Chukwuemeka Nkama, Andrea Poltronieri, Xavier Serra, Mart\'in Rocamora
Abstract: Recent approaches in music generation rely on disentangled representations, often labeled as structure and timbre or local and global, to enable controllable synthesis. Yet the underlying properties of these embeddings remain underexplored. In this work, we evaluate such disentangled representations in a set of music audio models for controllable generation using a probing-based framework that goes beyond standard downstream tasks. The selected models reflect diverse unsupervised disentanglement strategies, including inductive biases, data augmentations, adversarial objectives, and staged training procedures. We further isolate specific strategies to analyze their effect. Our analysis spans four key axes: informativeness, equivariance, invariance, and disentanglement, which are assessed across datasets, tasks, and controlled transformations. Our findings reveal inconsistencies between intended and actual semantics of the embeddings, suggesting that current strategies fall short of producing truly disentangled representations, and prompting a re-examination of how controllability is approached in music generation.
Authors: Zhaoyang Wang, Canwen Xu, Boyi Liu, Yite Wang, Siwei Han, Zhewei Yao, Huaxiu Yao, Yuxiong He
Abstract: Recent advances in large language model (LLM) have empowered autonomous agents to perform complex tasks that require multi-turn interactions with tools and environments. However, scaling such agent training is limited by the lack of diverse and reliable environments. In this paper, we propose Agent World Model (AWM), a fully synthetic environment generation pipeline. Using this pipeline, we scale to 1,000 environments covering everyday scenarios, in which agents can interact with rich toolsets (35 tools per environment on average) and obtain high-quality observations. Notably, these environments are code-driven and backed by databases, providing more reliable and consistent state transitions than environments simulated by LLMs. Moreover, they enable more efficient agent interaction compared with collecting trajectories from realistic environments. To demonstrate the effectiveness of this resource, we perform large-scale reinforcement learning for multi-turn tool-use agents. Thanks to the fully executable environments and accessible database states, we can also design reliable reward functions. Experiments on three benchmarks show that training exclusively in synthetic environments, rather than benchmark-specific ones, yields strong out-of-distribution generalization. The code is available at https://github.com/Snowflake-Labs/agent-world-model.
Authors: Xingjian Bai, Guande He, Zhengqi Li, Eli Shechtman, Xun Huang, Zongze Wu
Abstract: Causality -- referring to temporal, uni-directional cause-effect relationships between components -- underlies many complex generative processes, including videos, language, and robot trajectories. Current causal diffusion models entangle temporal reasoning with iterative denoising, applying causal attention across all layers, at every denoising step, and over the entire context. In this paper, we show that the causal reasoning in these models is separable from the multi-step denoising process. Through systematic probing of autoregressive video diffusers, we uncover two key regularities: (1) early layers produce highly similar features across denoising steps, indicating redundant computation along the diffusion trajectory; and (2) deeper layers exhibit sparse cross-frame attention and primarily perform intra-frame rendering. Motivated by these findings, we introduce Separable Causal Diffusion (SCD), a new architecture that explicitly decouples once-per-frame temporal reasoning, via a causal transformer encoder, from multi-step frame-wise rendering, via a lightweight diffusion decoder. Extensive experiments on both pretraining and post-training tasks across synthetic and real benchmarks show that SCD significantly improves throughput and per-frame latency while matching or surpassing the generation quality of strong causal diffusion baselines.
Authors: Yuxin Jiang, Yuchao Gu, Ivor W. Tsang, Mike Zheng Shou
Abstract: Scaling action-controllable world models is limited by the scarcity of action labels. While latent action learning promises to extract control interfaces from unlabeled video, learned latents often fail to transfer across contexts: they entangle scene-specific cues and lack a shared coordinate system. This occurs because standard objectives operate only within each clip, providing no mechanism to align action semantics across contexts. Our key insight is that although actions are unobserved, their semantic effects are observable and can serve as a shared reference. We introduce Seq$\Delta$-REPA, a sequence-level control-effect alignment objective that anchors integrated latent action to temporal feature differences from a frozen, self-supervised video encoder. Building on this, we present Olaf-World, a pipeline that pretrains action-conditioned video world models from large-scale passive video. Extensive experiments demonstrate that our method learns a more structured latent action space, leading to stronger zero-shot action transfer and more data-efficient adaptation to new control interfaces than state-of-the-art baselines.
Authors: Nicholas E. Corrado, Josiah P. Hanna
Abstract: On-policy reinforcement learning (RL) algorithms are typically characterized as algorithms that perform policy updates using i.i.d. trajectories collected by the agent's current policy. However, after observing only a finite number of trajectories, such on-policy sampling may produce data that fails to match the expected on-policy data distribution. This sampling error leads to high-variance gradient estimates that yield data-inefficient on-policy learning. Recent work in the policy evaluation setting has shown that non-i.i.d., off-policy sampling can produce data with lower sampling error w.r.t. the expected on-policy distribution than on-policy sampling can produce (Zhong et. al, 2022). Motivated by this observation, we introduce an adaptive, off-policy sampling method to reduce sampling error during on-policy policy gradient RL training. Our method, Proximal Robust On-Policy Sampling (PROPS), reduces sampling error by collecting data with a behavior policy that increases the probability of sampling actions that are under-sampled w.r.t. the current policy. We empirically evaluate PROPS on continuous-action MuJoCo benchmark tasks as well as discrete-action tasks and demonstrate that (1) PROPS decreases sampling error throughout training and (2) increases the data efficiency of on-policy policy gradient algorithms.
Authors: Bo Peng, Yadan Luo, Yonggang Zhang, Yixuan Li, Zhen Fang
Abstract: Post-hoc out-of-distribution (OOD) detection has garnered intensive attention in reliable machine learning. Many efforts have been dedicated to deriving score functions based on logits, distances, or rigorous data distribution assumptions to identify low-scoring OOD samples. Nevertheless, these estimate scores may fail to accurately reflect the true data density or impose impractical constraints. To provide a unified perspective on density-based score design, we propose a novel theoretical framework grounded in Bregman divergence, which extends distribution considerations to encompass an exponential family of distributions. Leveraging the conjugation constraint revealed in our theorem, we introduce a \textsc{ConjNorm} method, reframing density function design as a search for the optimal norm coefficient $p$ against the given dataset. In light of the computational challenges of normalization, we devise an unbiased and analytically tractable estimator of the partition function using the Monte Carlo-based importance sampling technique. Extensive experiments across OOD detection benchmarks empirically demonstrate that our proposed \textsc{ConjNorm} has established a new state-of-the-art in a variety of OOD detection setups, outperforming the current best method by up to 13.25$\%$ and 28.19$\%$ (FPR95) on CIFAR-100 and ImageNet-1K, respectively.
Authors: Yue Jian, Curtis Wu, Danny Reidenbach, Aditi S. Krishnapriyan
Abstract: Structure-based drug design (SBDD) aims to generate ligands that bind strongly and specifically to target protein pockets. Recent diffusion models have advanced SBDD by capturing the distributions of atomic positions and types, yet they often underemphasize binding affinity control during generation. To address this limitation, we introduce \textbf{\textnormal{\textbf{BADGER}}}, a general \textbf{binding-affinity guidance framework for diffusion models in SBDD}. \textnormal{\textbf{BADGER} }incorporates binding affinity awareness through two complementary strategies: (1) \textit{classifier guidance}, which applies gradient-based affinity signals during sampling in a plug-and-play fashion, and (2) \textit{classifier-free guidance}, which integrates affinity conditioning directly into diffusion model training. Together, these approaches enable controllable ligand generation guided by binding affinity. \textnormal{\textbf{BADGER} } can be added to any diffusion model and achieves up to a \textbf{60\% improvement in ligand--protein binding affinity} of sampled molecules over prior methods. Furthermore, we extend the framework to \textbf{multi-constraint diffusion guidance}, jointly optimizing for binding affinity, drug-likeness (QED), and synthetic accessibility (SA) to design realistic and synthesizable drug candidates.
Authors: Cheng Zeng, Zulqarnain Khan, Nathan L. Post
Abstract: Inverse materials design has proven successful in accelerating novel material discovery. Many inverse materials design methods use unsupervised learning where a latent space is learned to offer a compact description of materials representations. A latent space learned this way is likely to be entangled, in terms of the target property and other properties of the materials. This makes the inverse design process ambiguous. Here, we present a semi-supervised learning approach based on a disentangled variational autoencoder to learn a probabilistic relationship between features, latent variables and target properties. This approach is data efficient because it combines all labelled and unlabelled data in a coherent manner, and it uses expert-informed prior distributions to improve model robustness even with limited labelled data. It is in essence interpretable, as the learnable target property is disentangled out of the other properties of the materials, and an extra layer of interpretability can be provided by a post-hoc analysis of the classification head of the model. We demonstrate this new approach on an experimental high-entropy alloy dataset with chemical compositions as input and single-phase formation as the single target property. High-entropy alloys were chosen as example materials because of the vast chemical space of their possible combinations of compositions and atomic configurations. While single property is used in this work, the disentangled model can be extended to customize for inverse design of materials with multiple target properties.
Authors: Gustav Wagner Zakarias, Lars Kai Hansen, Zheng-Hua Tan
Abstract: Models initialized from self-supervised pretraining may suffer from poor alignment with downstream tasks, reducing the extent to which subsequent fine-tuning can adapt pretrained features toward downstream objectives. To mitigate this, we introduce BiSSL, a novel bilevel training framework that enhances the alignment of self-supervised pretrained models with downstream tasks prior to fine-tuning. BiSSL acts as an intermediate training stage conducted after conventional self-supervised pretraining and is tasked with solving a bilevel optimization problem that incorporates the pretext and downstream training objectives in its lower- and upper-level objectives, respectively. This approach explicitly models the interdependence between the pretraining and fine-tuning stages within the conventional self-supervised learning pipeline, facilitating enhanced information sharing between them that ultimately leads to a model initialization better aligned with the downstream task. We propose a general training algorithm for BiSSL that is compatible with a broad range of pretext and downstream tasks. Using SimCLR and Bootstrap Your Own Latent to pretrain ResNet-50 backbones on the ImageNet dataset, we demonstrate that our proposed framework significantly improves accuracy on the vast majority of 12 downstream image classification datasets, as well as on object detection. Exploratory analyses alongside investigative experiments further provide compelling evidence that BiSSL enhances downstream alignment.
Authors: Prabhat Lankireddy, Jayakrishnan Nair, D Manjunath
Abstract: We analyze the unintended effects that recommender systems have on the preferences of users that they are learning. We consider a contextual multi-armed bandit recommendation algorithm that learns optimal product recommendations based on user and product attributes. It is well known that the sequence of recommendations affects user preferences. However, typical learning algorithms treat the user attributes as static and disregard the impact of their recommendations on user preferences. Our interest is to analyze the effect of this mismatch between the model assumption of a static environment and the reality of an evolving environment affected by the recommendations. To perform this analysis, we introduce a model for the coupled evolution of a linear bandit recommendation system and its users, whose preferences are drawn towards the recommendations made by the algorithm. We describe a method, that is grounded in stochastic approximation theory, to come up with a dynamical system model that asymptotically approximates the mean behavior of the stochastic model. The resulting dynamical system captures the coupled evolution of the population preferences and the learning algorithm. Analyzing this dynamical system gives insight into the long-term properties of user preferences and the learning algorithm. Under certain conditions, we show that the RS is able to learn the population preferences in spite of the model mismatch. We discuss and characterize the relation between various parameters of the model and the long term preferences of users in this work. A key observation is that the exploration-exploitation tradeoff used by the recommendation algorithm significantly affects the long term preferences of users. Algorithms that exploit more can polarize user preferences, leading to the well-known filter bubble phenomenon.
Authors: Arya Hatamian, Lionel Levine, Haniyeh Ehsani Oskouie, Majid Sarrafzadeh
Abstract: Having a sufficient quantity of quality data is a critical enabler of training effective machine learning models. Being able to effectively determine the adequacy of a dataset prior to training and evaluating a model's performance would be an essential tool for anyone engaged in experimental design or data collection. However, despite the need for it, the ability to prospectively assess data sufficiency remains an elusive capability. We report here on two experiments undertaken in an attempt to better ascertain whether or not basic descriptive statistical measures can be indicative of how effective a dataset will be at training a resulting model. Leveraging the effect size of our features, this work first explores whether or not a correlation exists between effect size, and resulting model performance (theorizing that the magnitude of the distinction between classes could correlate to a classifier's resulting success). We then explore whether or not the magnitude of the effect size will impact the rate of convergence of our learning rate, (theorizing again that a greater effect size may indicate that the model will converge more rapidly, and with a smaller sample size needed). Our results appear to indicate that this is not an effective heuristic for determining adequate sample size or projecting model performance, and therefore that additional work is still needed to better prospectively assess adequacy of data.
Authors: Kaixi Bao, Chenhao Li, Yarden As, Andreas Krause, Marco Hutter
Abstract: Training reinforcement learning (RL) policies for legged locomotion often requires extensive environment interactions, which are costly and time-consuming. We propose Symmetry-Guided Memory Augmentation (SGMA), a framework that improves training efficiency by combining structured experience augmentation with memory-based context inference. Our method leverages robot and task symmetries to generate additional, physically consistent training experiences without requiring extra interactions. To avoid the pitfalls of naive augmentation, we extend these transformations to the policy's memory states, enabling the agent to retain task-relevant context and adapt its behavior accordingly. We evaluate the approach on quadruped and humanoid robots in simulation, as well as on a real quadruped platform. Across diverse locomotion tasks involving joint failures and payload variations, our method achieves efficient policy training while maintaining robust performance, demonstrating a practical route toward data-efficient RL for legged robots.
Authors: Nikunj Gupta, James Zachary Hare, Jesse Milzman, Rajgopal Kannan, Viktor Prasanna
Abstract: This paper presents deep meta coordination graphs (DMCG) for learning cooperative policies in multi-agent reinforcement learning (MARL). Coordination graph formulations encode local interactions and accordingly factorize the joint value function of all agents to improve efficiency in MARL. Through DMCG, we dynamically compose what we refer to as \textit{meta coordination graphs}, to learn a more expressive representation of agent interactions and use them to integrate agent information through graph convolutional networks. The goal is to enable an evolving coordination graph to guide effective coordination in cooperative MARL tasks. The graphs are jointly optimized with agents' value functions to learn to implicitly reason about joint actions, facilitating the end-to-end learning of interaction representations and coordinated policies. We demonstrate that DMCG consistently achieves state-of-the-art coordination performance and sample efficiency on challenging cooperative tasks, outperforming several prior graph-based and non-graph-based MARL baselines. Through several ablations, we also isolate the impact of individual components in DMCG, showing that the observed improvements are due to the meaningful design choices in this approach. We also include an analysis of its computational complexity to discuss its practicality in real-world applications. All codes can be found here: {\color{blue}{https://github.com/Nikunj-Gupta/dmcg-marl}.
Authors: Linus Aronsson, Arman Rahbar, Morteza Haghir Chehreghani
Abstract: Active feature acquisition (AFA) studies how to sequentially acquire features for each data instance to trade off predictive performance against acquisition cost. This survey offers the first unified treatment of AFA via an explicit partially observable Markov decision process (POMDP) formulation. We place this formulation in the broader literature on optimal information acquisition and, more specifically, in a family of structured POMDPs (for example, information-gathering and sensing POMDPs) whose assumptions and algorithmic tools directly apply to AFA. This connection provides a common language for comparing problem settings and methods, and it highlights where AFA can leverage established results in structured POMDP planning and approximation. Building on this perspective, we present an up-to-date taxonomy of AFA methods that (roughly) mirrors standard approaches to solving POMDPs: (i) embedded cost-aware predictors (notably cost-sensitive decision trees and ensembles), (ii) model-based methods that plan using learned probabilistic components, (iii) model-free methods that learn acquisition policies from simulated episodes, and (iv) hybrid methods that combine the strengths of model-based and model-free approaches. We argue that this POMDP-centric view clarifies connections among existing methods and motivates more principled algorithm design. Since much prior work is heuristic and lacks formal guarantees, we also outline routes to guarantees by connecting AFA to adaptive stochastic optimization. We conclude by highlighting open challenges and promising directions for future research.
Authors: Ali Eslamian, Qiang Cheng
Abstract: Tabular data poses unique challenges for deep learning due to its heterogeneous feature types, lack of spatial structure, and often limited sample sizes. We propose TabNSA, a novel deep learning framework that integrates Native Sparse Attention (NSA) with a TabMixer backbone to efficiently model tabular data. TabNSA tackles computational and representational challenges by dynamically focusing on relevant feature subsets per instance. The NSA module employs a hierarchical sparse attention mechanism, including token compression, selective preservation, and localized sliding windows, to significantly reduce the quadratic complexity of standard attention operations while addressing feature heterogeneity. Complementing this, the TabMixer backbone captures complex, non-linear dependencies through parallel multilayer perceptron (MLP) branches with independent parameters. These modules are synergistically combined via element-wise summation and mean pooling, enabling TabNSA to model both global context and fine-grained interactions. Extensive experiments across supervised and transfer learning settings show that TabNSA consistently outperforms state-of-the-art deep learning models. Furthermore, by augmenting TabNSA with a fine-tuned large language model (LLM), we enable it to effectively address Few-Shot Learning challenges through language-guided generalization on diverse tabular benchmarks. Code available on: https://github.com/aseslamian/TabNSA
Authors: Jisoo Kim, Sungmin Kang, Sunwoo Lee
Abstract: Expensive communication cost is a common performance bottleneck in Federated Learning (FL), which makes it less appealing in real-world applications. Many communication-efficient FL methods focus on discarding a part of model updates mostly based on gradient magnitude. In this study, we find that recycling previous updates, rather than simply dropping them, more effectively reduces the communication cost while maintaining FL performance. We propose FedLUAR, a Layer-wise Update Aggregation with Recycling scheme for communication-efficient FL. We first define a useful metric that quantifies the extent to which the aggregated gradients influences the model parameter values in each layer. FedLUAR selects a few layers based on the metric and recycles their previous updates on the server side. Our extensive empirical study demonstrates that the update recycling scheme significantly reduces the communication cost while maintaining model accuracy. For example, our method achieves nearly the same AG News accuracy as FedAvg, while reducing the communication cost to just 17%.
Authors: Samira Alkaee Taleghan, Morteza Karimzadeh, Andrew P. Barrett, Walter N. Meier, Farnoush Banaei-Kashani
Abstract: Accurate segmentation and mapping of sea ice types is crucial for safe polar navigation, offshore operations, and climate monitoring. While deep learning has demonstrated strong potential for automating sea ice type segmentation, its success often relies on access to extensive expert labeled datasets, which is both resource intensive and time consuming to create. However, foundation models (FMs), recently developed through self-supervised training on large-scale datasets, have demonstrated impressive performance. Nevertheless, their applicability to sea ice type segmentation based on Synthetic Aperture Radar (SAR) imagery remains uncertain due to the unique challenges posed by sea ice such as intricate geophysical patterns, pronounced seasonal variability, and SAR-specific artifacts like banding, scalloping, and heterogeneous backscatter as well as the fact that SAR data in polar regions are often acquired using specialized sensor modes that differ markedly from those used to collect FM training data at lower latitudes, limiting their direct transferability to polar environments. To address this gap, we contribute: (1) IceFMBench, a comprehensive benchmark framework for evaluation of the state-of-the-art remote sensing FMs on the sea ice type segmentation task using Sentinel1 SAR imagery, where IceFMBench is composed of a widely used standardized dataset, diverse evaluation metrics, and a representative set of selected remote sensing FM models suitable for sea ice type segmentation, with the ability to include new models side by side the existing models; (2) an extensive comparative evaluation of the representative FMs using IceFMBench, with additional case studies to assess performance of the top-performing model in terms of transferability across temporal and spatial domains and (3) a multi teacher knowledge distillation approach to address lack of spatiotemporal transferability.
Authors: Mingqi Yuan, Qi Wang, Guozheng Ma, Caihao Sun, Bo Li, Xin Jin, Yunbo Wang, Xiaokang Yang, Wenjun Zeng, Dacheng Tao, Jiayu Chen
Abstract: Developing lifelong learning agents is crucial for artificial general intelligence (AGI). However, deep reinforcement learning (RL) systems often suffer from plasticity loss, where neural networks gradually lose their ability to adapt during training. Despite its significance, this field lacks unified benchmarks and evaluation protocols. We introduce Plasticine, the first open-source framework for benchmarking plasticity optimization in deep RL. Plasticine provides single-file implementations of over 13 mitigation methods, 6 evaluation metrics, and learning scenarios with increasing non-stationarity levels from standard to continually varying environments. This framework enables researchers to systematically quantify plasticity loss, evaluate mitigation strategies, and analyze plasticity dynamics across different contexts. Our documentation, examples, and source code are available at https://github.com/RLE-Foundation/Plasticine.
Authors: Wilson Wongso, Hao Xue, Flora D. Salim
Abstract: Understanding human mobility through Point-of-Interest (POI) trajectory modeling is increasingly important for applications such as urban planning, personalized services, and generative agent simulation. However, progress in this field is hindered by two key challenges: the over-reliance on older datasets from 2012-2013 and the lack of reproducible, city-level check-in datasets that reflect diverse global regions. To address these gaps, we present Massive-STEPS (Massive Semantic Trajectories for Understanding POI Check-ins), a large-scale, publicly available benchmark dataset built upon the Semantic Trails dataset and enriched with semantic POI metadata. Massive-STEPS spans 15 geographically and culturally diverse cities and features more recent (2017-2018) and longer-duration (24 months) check-in data than prior datasets. We benchmarked a wide range of POI models on Massive-STEPS using both supervised and zero-shot approaches, and evaluated their performance across multiple urban contexts. By releasing Massive-STEPS, we aim to facilitate reproducible and equitable research in human mobility and POI trajectory modeling. The dataset and benchmarking code are available at: https://github.com/cruiseresearchgroup/Massive-STEPS.
Authors: Zhichen Zeng, Ruizhong Qiu, Wenxuan Bao, Tianxin Wei, Xiao Lin, Yuchen Yan, Tarek F. Abdelzaher, Jiawei Han, Hanghang Tong
Abstract: Graph neural networks, despite their impressive performance, are highly vulnerable to distribution shifts on graphs. Existing graph domain adaptation (graph DA) methods often implicitly assume a mild shift between source and target graphs, limiting their applicability to real-world scenarios with large shifts. Gradual domain adaptation (GDA) has emerged as a promising approach for addressing large shifts by gradually adapting the source model to the target domain via a path of unlabeled intermediate domains. Existing GDA methods exclusively focus on independent and identically distributed (IID) data with a predefined path, leaving their extension to non-IID graphs without a given path an open challenge. To bridge this gap, we present Gadget, the first GDA framework for non-IID graph data. First (theoretical foundation), the Fused Gromov-Wasserstein (FGW) distance is adopted as the domain discrepancy for non-IID graphs, based on which, we derive an error bound on node, edge and graph-level tasks, showing that the target domain error is proportional to the length of the path. Second (optimal path), guided by the error bound, we identify the FGW geodesic as the optimal path, which can be efficiently generated by our proposed algorithm. The generated path can be seamlessly integrated with existing graph DA methods to handle large shifts on graphs, improving state-of-the-art graph DA methods by up to 6.8% in accuracy on real-world datasets.
Authors: Zihan Gu, Ruoyu Chen, Han Zhang, Hua Zhang, Yue Hu
Abstract: Positional encodings enable Transformers to incorporate sequential information, yet their theoretical understanding remains limited to two properties: distance attenuation and translation invariance. Because natural language lacks purely positional data, the interplay between positional and semantic information is still underexplored. We address this gap by deconstructing the attention-logit computation and providing a structured analysis of positional encodings, categorizing them into additive and multiplicative forms. The differing properties of these forms lead to distinct mechanisms for capturing positional information. To probe this difference, we design a synthetic task that explicitly requires strong integration of positional and semantic cues. As predicted, multiplicative encodings achieve a clear performance advantage on this task. Moreover, our evaluation reveals a hidden training bias: an information aggregation effect in shallow layers that we term the single-head deposit pattern. Through ablation studies and theoretical analysis, we proved that this phenomenon is inherent in multiplicative encodings. These findings deepen the understanding of positional encodings and call for further study of their training dynamics.
Authors: Quentin Clark, Florian Shkurti
Abstract: In policy learning, stitching and compositional generalization refer to the extent to which the policy is able to piece together sub-trajectories of data it is trained on to generate new and diverse behaviours. While stitching has been identified as a significant strength of offline reinforcement learning, recent generative behavioural cloning (BC) methods have also shown proficiency at stitching. However, the main factors behind this are poorly understood, hindering the development of new algorithms that can reliably stitch by design. Focusing on diffusion planners trained via generative behavioural cloning, and without resorting to dynamic programming or TD-learning, we find three properties are key enablers for composition: shift equivariance, local receptive fields, and inference choices. We use these properties to explain architecture, data, and inference choices in existing generative BC methods based on diffusion planning including replanning frequency, data augmentation, and data scaling. Our experiments show that while local receptive fields are more important than shift equivariance in creating a diffusion planner capable of composition, both are crucial. Using findings from our experiments, we develop a new architecture for diffusion planners called Eq-Net, that is simple, produces diverse trajectories competitive with more computationally expensive methods such as replanning or scaling data, and can be guided to enable generalization in goal-conditioned settings. We show that Eq-Net exhibits significant compositional generalization in a variety of navigation and manipulation tasks designed to test planning diversity.
Authors: Kiljae Lee, Ziqi Liu, Weijing Tang, Yuan Zhang
Abstract: Data Shapley is an important tool for data valuation, which quantifies the contribution of individual data points to machine learning models. In practice, group-level data valuation is desirable when data providers contribute data in batch. However, we identify that existing group-level extensions of Data Shapley are vulnerable to shell company attacks, where strategic group splitting can unfairly inflate valuations. We propose Faithful Group Shapley Value (FGSV) that uniquely defends against such attacks. Building on original mathematical insights, we develop a provably fast and accurate approximation algorithm for computing FGSV. Empirical experiments demonstrate that our algorithm significantly outperforms state-of-the-art methods in computational efficiency and approximation accuracy, while ensuring faithful group-level valuation.
Authors: Paul Setinek, Gianluca Galletti, Thomas Gross, Dominik Schn\"urer, Johannes Brandstetter, Werner Zellinger
Abstract: Neural surrogates for Partial Differential Equations (PDEs) often suffer significant performance degradation when evaluated on problem configurations outside their training distribution, such as new initial conditions or structural dimensions. While Unsupervised Domain Adaptation (UDA) techniques have been widely used in vision and language to generalize across domains without additional labeled data, their application to complex engineering simulations remains largely unexplored. In this work, we address this gap through two focused contributions. First, we introduce SIMSHIFT, a novel benchmark dataset and evaluation suite composed of four industrial simulation tasks spanning diverse processes and physics: hot rolling, sheet metal forming, electric motor design and heatsink design. Second, we extend established UDA methods to state-of-the-art neural surrogates and systematically evaluate them. Extensive experiments on SIMSHIFT highlight the challenges of out-of-distribution neural surrogate modeling, demonstrate the potential of UDA in simulation, and reveal open problems in achieving robust neural surrogates under distribution shifts in industrially relevant scenarios. Our codebase is available at https://github.com/psetinek/simshift
Authors: Jackson W. Burns, Akshat Shirish Zalte, Charlles R. A. Abreu, Jochen Sieg, Christian Feldmann, Miriam Mathea, William H. Green
Abstract: Fast and accurate data-driven prediction of molecular properties is pivotal to scientific advancements across myriad chemical domains. Deep learning methods have recently garnered much attention, despite their inability to outperform classical machine learning methods when tested on practical, real-world benchmarks with limited training data. This study seeks to bridge this gap with CheMeleon, a O(10M) parameter foundation model that enables directed message-passing neural networks to finally exceed the performance of classical methods. Evaluated on 58 benchmark datasets from Polaris and MoleculeACE, CheMeleon achieves a win rate of 75% on Polaris tasks, outperforming baselines like Random Forest (68%), fastprop (36%), and Chemprop (32%), and a 97% win rate on MoleculeACE assays, surpassing Random Forest (50%) and other foundation models. Unlike conventional pre-training approaches that rely on noisy experimental data or biased quantum mechanical simulations, CheMeleon utilizes low-noise molecular descriptors to learn rich and highly transferable molecular representations, suggesting a new avenue for foundation model pre-training.
Authors: Anish Dhir, Cristiana Diaconu, Valentinian Mihai Lungu, James Requeima, Richard E. Turner, Mark van der Wilk
Abstract: In scientific domains -- from biology to the social sciences -- many questions boil down to \textit{What effect will we observe if we intervene on a particular variable?} If the causal relationships (e.g.~a causal graph) are known, it is possible to estimate the intervention distributions. In the absence of this domain knowledge, the causal structure must be discovered from the available observational data. However, observational data are often compatible with multiple causal graphs, making methods that commit to a single structure prone to overconfidence. A principled way to manage this structural uncertainty is via Bayesian inference, which averages over a posterior distribution on possible causal structures and functional mechanisms. Unfortunately, the number of causal structures grows super-exponentially with the number of nodes in the graph, making computations intractable. We propose to circumvent these challenges by using meta-learning to create an end-to-end model: the Model-Averaged Causal Estimation Transformer Neural Process (MACE-TNP). The model is trained to predict the Bayesian model-averaged interventional posterior distribution, and its end-to-end nature bypasses the need for expensive calculations. Empirically, we demonstrate that MACE-TNP outperforms strong Bayesian baselines. Our work establishes meta-learning as a flexible and scalable paradigm for approximating complex Bayesian causal inference, that can be scaled to increasingly challenging settings in the future.
Authors: Md Arafat Hossain, Xingfu Wu, Valerie Taylor, Ali Jannesari
Abstract: Despite recent advancements of large language models (LLMs), optimally predicting the model size for LLM pretraining or allocating optimal resources still remains a challenge. Several efforts have addressed the challenge by proposing different empirical scaling laws, but almost all of them are architecture-specific (dense or sparse). In this work we revisit existing empirical scaling laws and propose a generalized scaling law to provide a unified framework that is applicable to both dense and sparse large language models. We evaluate and compare our proposed scaling law with existing scaling laws and demonstrate that our proposed scaling law captures the scaling behavior of existing scaling laws. Further, we show an IsoFLOP comparison between our proposed scaling law and the state-of-the-art scaling law to illustrate the effectiveness of our proposed scaling law for Mixture-of-Expert (MoE)-based very large LLMs like DeepSeek-V3. Our proposed scaling law can be used to estimate the best model hyperparameters (Model size, Tokens and Compute) for a given sparsity or to identify the optimal sparsity for the given model hyperparameters.
Authors: Valter Sch\"utz, Han Wu, Reza Rezvan, Linus Aronsson, Morteza Haghir Chehreghani
Abstract: In many real-world scenarios, acquiring all features of a data instance can be expensive or impractical due to monetary cost, latency, or privacy concerns. Active Feature Acquisition (AFA) addresses this challenge by dynamically selecting a subset of informative features for each data instance, trading predictive performance against acquisition cost. While numerous methods have been proposed for AFA, ranging from myopic information-theoretic strategies to non-myopic reinforcement learning approaches, fair and systematic evaluation of these methods has been hindered by a lack of standardized benchmarks. In this paper, we introduce AFABench, the first benchmark framework for AFA. Our benchmark includes a diverse set of synthetic and real-world datasets, supports a wide range of acquisition policies, and provides a modular design that enables easy integration of new methods and tasks. We implement and evaluate representative algorithms from all major categories, including static, myopic, and reinforcement learning-based approaches. To test the lookahead capabilities of AFA policies, we introduce a novel synthetic dataset, CUBE-NM, designed to expose the limitations of myopic selection. Our results highlight key trade-offs between different AFA strategies and provide actionable insights for future research. The benchmark code is available at: https://github.com/Linusaronsson/AFA-Benchmark.
Authors: Kushal Kapoor, Wyatt Mackey, Yiannis Aloimonos, Xiaomin Lin
Abstract: We propose HiCL, a novel hippocampal-inspired dual-memory continual learning architecture designed to mitigate catastrophic forgetting by using elements inspired by the hippocampal circuitry. Our system encodes inputs through a grid-cell-like layer, followed by sparse pattern separation using a dentate gyrus-inspired module with top-k sparsity. Episodic memory traces are maintained in a CA3-like autoassociative memory. Task-specific processing is dynamically managed via a DG-gated mixture-of-experts mechanism, wherein inputs are routed to experts based on cosine similarity between their normalized sparse DG representations and learned task-specific DG prototypes computed through online exponential moving averages. This biologically grounded yet mathematically principled gating strategy enables differentiable, scalable task-routing without relying on a separate gating network, and enhances the model's adaptability and efficiency in learning multiple sequential tasks. Cortical outputs are consolidated using Elastic Weight Consolidation weighted by inter-task similarity. Crucially, we incorporate prioritized replay of stored patterns to reinforce essential past experiences. Evaluations on standard continual learning benchmarks demonstrate the effectiveness of our architecture in reducing task interference, achieving near state-of-the-art results in continual learning tasks at lower computational costs. Our code is available here https://github.com/kushalk173-sc/HiCL.
Authors: Loka Li, Wong Yu Kang, Minghao Fu, Guangyi Chen, Zhenhao Chen, Gongxu Luo, Yuewen Sun, Salman Khan, Peter Spirtes, Kun Zhang
Abstract: Understanding human behavior traits is central to applications in human-computer interaction, computational social science, and personalized AI systems. Such understanding often requires integrating multiple modalities to capture nuanced patterns and relationships. However, existing resources rarely provide datasets that combine behavioral descriptors with complementary modalities such as facial attributes and biographical information. To address this gap, we present PersonaX, a curated collection of multimodal datasets designed to enable comprehensive analysis of public traits across modalities. PersonaX consists of (1) CelebPersona, featuring 9444 public figures from diverse occupations, and (2) AthlePersona, covering 4181 professional athletes across 7 major sports leagues. Each dataset includes behavioral trait assessments inferred by three high-performing large language models, alongside facial imagery and structured biographical features. We analyze PersonaX at two complementary levels. First, we abstract high-level trait scores from text descriptions and apply five statistical independence tests to examine their relationships with other modalities. Second, we introduce a novel causal representation learning (CRL) framework tailored to multimodal and multi-measurement data, providing theoretical identifiability guarantees. Experiments on both synthetic and real-world data demonstrate the effectiveness of our approach. By unifying structured and unstructured analysis, PersonaX establishes a foundation for studying LLM-inferred behavioral traits in conjunction with visual and biographical attributes, advancing multimodal trait analysis and causal reasoning. The code is available at https://github.com/lokali/PersonaX.
Authors: Yifan Hu, Jie Yang, Tian Zhou, Peiyuan Liu, Yujin Tang, Rong Jin, Liang Sun
Abstract: Although contrastive and other representation-learning methods have long been explored in vision and NLP, their adoption in modern time series forecasters remains limited. We believe they hold strong promise for this domain. To unlock this potential, we explicitly align past and future representations, thereby bridging the distributional gap between input histories and future targets. To this end, we introduce TimeAlign, a lightweight, plug-and-play framework that establishes a new representation paradigm, distinct from contrastive learning, by aligning auxiliary features via a simple reconstruction task and feeding them back into any base forecaster. Extensive experiments across eight benchmarks verify its superior performance. Further studies indicate that the gains arise primarily from correcting frequency mismatches between historical inputs and future outputs. Additionally, we provide two theoretical justifications for how reconstruction improves forecasting generalization and how alignment increases the mutual information between learned representations and predicted targets. The code is available at https://github.com/TROUBADOUR000/TimeAlign.
Authors: Taiga Kojima, Haruto Kajita, Ayato Kohara, Masayuki Karasuyama
Abstract: In the graph-level prediction task (predict a label for a given graph), the information contained in subgraphs of the input graph plays a key role. In this paper, we propose Exact subgraph Isomorphism Network (EIN), which combines the exact subgraph enumeration, a neural network, and a sparse regularization by the mixed $L_{0,2}$ norm constraint. In general, building a graph-level prediction model achieving high discriminative ability along with interpretability is still a challenging problem. Our combination of the subgraph enumeration and neural network contributes to high discriminative ability about the subgraph structure of the input graph. Further, the sparse regularization in EIN enables us 1) to derive an effective pruning strategy that mitigates computational difficulty of the enumeration while maintaining the prediction performance, and 2) to identify important subgraphs that contributes to high interpretability. We empirically show that EIN has sufficiently high prediction performance compared with standard graph neural network models, and also, we show examples of post-hoc analysis based on the selected subgraphs.
Authors: Wujiang Xu, Wentian Zhao, Zhenting Wang, Yu-Jhe Li, Can Jin, Mingyu Jin, Kai Mei, Kun Wan, Dimitris N. Metaxas
Abstract: Training LLM agents in multi-turn environments with sparse rewards, where completing a single task requires 30+ turns of interaction within an episode, presents a fundamental challenge for reinforcement learning. We identify a critical failure mode unique to this setting: the exploration-exploitation cascade failure. This cascade begins with early-stage policy premature convergence, where sparse feedback causes agents to commit to flawed, low-entropy strategies. Subsequently, agents enter late-stage policy collapse, where conventional entropy regularization becomes counterproductive, promoting chaotic exploration that destabilizes training. We propose Entropy-regularized Policy Optimization (EPO), a general framework that breaks this failure cycle through three synergistic mechanisms: (1) adopting entropy regularization in multi-turn settings to enhance exploration, (2) an entropy smoothing regularizer that bounds policy entropy within historical averages to prevent abrupt fluctuations, and (3) adaptive phase-based weighting that balances exploration and exploitation across training. Our analysis justifies that EPO guarantees monotonically decreasing entropy variance while maintaining convergence. EPO achieves up to 152% performance improvement on ScienceWorld and up to 19.8% on ALFWorld. Our work demonstrates that multi-turn sparse-reward settings require fundamentally different entropy control than traditional RL, with broad implications for LLM agent training.
Authors: Minh Le, Bao-Ngoc Dao, Huy Nguyen, Quyen Tran, Anh Nguyen, Nhat Ho
Abstract: Prompt-based methods have recently gained prominence in Continual Learning (CL) due to their strong performance and memory efficiency. A prevalent strategy in this paradigm assigns a dedicated subset of prompts to each task, which, while effective, incurs substantial computational overhead and causes memory requirements to scale linearly with the number of tasks. Conversely, approaches employing a single shared prompt across tasks offer greater efficiency but often suffer from degraded performance due to knowledge interference. To reconcile this trade-off, we propose SMoPE, a novel framework that integrates the benefits of both task-specific and shared prompt strategies. Inspired by recent findings on the relationship between Prefix Tuning and Mixture of Experts (MoE), SMoPE organizes a shared prompt into multiple "prompt experts" within a sparse MoE architecture. For each input, only a select subset of relevant experts is activated, effectively mitigating interference. To facilitate expert selection, we introduce a prompt-attention score aggregation mechanism that computes a unified proxy score for each expert, enabling dynamic and sparse activation. Additionally, we propose an adaptive noise mechanism to encourage balanced expert utilization while preserving knowledge from prior tasks. To further enhance expert specialization, we design a prototype-based loss function that leverages prefix keys as implicit memory representations. Extensive experiments across multiple CL benchmarks demonstrate that SMoPE consistently outperforms task-specific prompt methods and achieves performance competitive with state-of-the-art approaches, all while significantly reducing parameter counts and computational costs.
Authors: Mohammadreza Bakhtyari, Bogdan Mazoure, Renato Cordeiro de Amorim, Guillaume Rabusseau, Vladimir Makarenkov
Abstract: In unsupervised learning, identifying an effective clustering algorithm for a given tabular dataset remains a fundamental challenge. We introduce ClustRecNet, a novel end-to-end deep learning framework that recommends a suitable clustering algorithm by directly learning high-order representations of raw tabular data. To facilitate robust meta-learning, we construct a comprehensive repository of 34,000 synthetic datasets with diverse structures, run 10 prominent clustering algorithms, and use Adjusted Rand Index (ARI) to establish ground-truth labels. ClustRecNet integrates convolutional, residual, and attention mechanisms to capture both local/global structural patterns, effectively bypassing the knowledge bottleneck associated with manual feature engineering. Extensive evaluations on both synthetic and real-world benchmarks demonstrate that ClustRecNet consistently outperforms state-of-the-art Automated Machine Learning (AutoML) approaches, including ML2DAC and AutoML4Clust. Our framework achieves an average 0.497 ARI gain over the well-known Calinski-Harabasz cluster validity index on synthetic data and an average 15.3% ARI improvement over the leading AutoML approach (ML2DAC) on real-world benchmarks. To the best of our knowledge, we are the first to successively apply deep learning to automatically recommend suitable clustering algorithms for tabular data at hand.
Authors: Justin Chih-Yao Chen, Becky Xiangyu Peng, Prafulla Kumar Choubey, Kung-Hsiang Huang, Jiaxin Zhang, Mohit Bansal, Chien-Sheng Wu
Abstract: Current online reinforcement learning (RL) algorithms like GRPO share a key limitation in LLM reasoning: they cannot learn from problems that are "unsolvable" to the model. In other words, they can only improve performance on problems where the model is capable of exploring the correct answer. Consequently, the model's "upper limit" remains unchanged after RL training, even though the likelihood of solving easier, solvable problems may increase. These hard samples cannot contribute to training, as no rollouts yield rewards and thus no gradients are produced. To unlock learning from these hard samples, we propose NuRL, a "nudging" method that aims to push the upper bound of LLM reasoning using self-generated hints, i.e., abstract cues that help reduce the problem difficulty for the model. Given a question and its gold answer, the model generates a CoT and then produces a hint containing the core knowledge needed to solve the problem. During training, we generate G rollouts from the base policy and use the pass rate to decide whether the hint should be injected. For hard samples with a 0% pass rate, we inject the hint and regenerate a new batch of trajectories. This yields two benefits: (1) the hint boosts pass rates (from 0% to non-zero), thereby introducing training signals for previously unsolvable samples, and (2) the hints are self-generated, avoiding distributional shift and do not rely on external models. NuRL achieves consistent improvements across 6 benchmarks and 3 models, while remaining complementary to test-time scaling. Notably, NuRL can raise the model's upper limit, whereas GRPO leaves pass@1024 unchanged from the base model. Furthermore, we present a systematic study of what makes an effective hint and when hints are most useful. Interestingly, the best hints are abstract and high-level, and are most beneficial when applied necessarily and after GRPO has converged.
Authors: Hengwei Zhao, Zhengzhong Tu, Zhuo Zheng, Wei Wang, Junjue Wang, Rusty Feagin, Wenzhe Jiao
Abstract: Positive-Unlabeled (PU) learning aims to train a binary classifier (positive vs. negative) where only limited positive data and abundant unlabeled data are available. While widely applicable, state-of-the-art PU learning methods substantially underperform their supervised counterparts on complex datasets, especially without auxiliary negatives or pre-estimated parameters (e.g., a 14.26% gap on CIFAR-100 dataset). We identify the primary bottleneck as the challenge of learning discriminative representations under unreliable supervision. To tackle this challenge, we propose NcPU, a non-contrastive PU learning framework that requires no auxiliary information. NcPU combines a noisy-pair robust supervised non-contrastive loss (NoiSNCL), which aligns intra-class representations despite unreliable supervision, with a phantom label disambiguation (PLD) scheme that supplies conservative negative supervision via regret-based label updates. Theoretically, NoiSNCL and PLD can iteratively benefit each other from the perspective of the Expectation-Maximization framework. Empirically, extensive experiments demonstrate that: (1) NoiSNCL enables simple PU methods to achieve competitive performance; and (2) NcPU achieves substantial improvements over state-of-the-art PU methods across diverse datasets, including challenging datasets on post-disaster building damage mapping, highlighting its promise for real-world applications. Code: Code will be open-sourced after review.
Authors: Wanxin Li, Yongjin P. Park, Khanh Dao Duc
Abstract: Fairness testing evaluates whether a model satisfies a specified fairness criterion across different groups, yet most research has focused on classification models, leaving regression models underexplored. This paper introduces a framework for fairness testing in regression models, leveraging Wasserstein distance to project data distribution and focusing on expectation-based criteria. Upon categorizing fairness criteria for regression, we derive a Wasserstein projection test statistic from dual reformulation, and derive asymptotic bounds and limiting distributions, allowing us to formulate both a hypothesis-testing procedure and an optimal data perturbation method to improve fairness while balancing accuracy. Experiments on synthetic data demonstrate that the proposed hypothesis-testing approach offers higher specificity compared to permutation-based tests. To illustrate its potential applications, we apply our framework to two case studies on real data, showing (1) statistically significant gender disparities that appear on student performance data across multiple models, and (2) significant unfairness between pollution areas under multiple fairness criteria affecting housing price data, robust to different group divisions, with feature-level analysis identifying spatial and socioeconomic drivers.
Authors: Enrique Queipo-de-Llano, \'Alvaro Arroyo, Federico Barbero, Xiaowen Dong, Michael Bronstein, Yann LeCun, Ravid Shwartz-Ziv
Abstract: Attention sinks and compression valleys have attracted significant attention as two puzzling phenomena in large language models, but have been studied in isolation. In this work, we present a surprising connection between attention sinks and compression valleys, tracing both to the formation of massive activations in the residual stream. We prove theoretically that massive activations necessarily produce representational compression and establish bounds on the resulting entropy reduction. Through experiments across several models (410M-120B parameters), we confirm that when the beginning-of-sequence token develops extreme activation norms in the middle layers, both compression valleys and attention sinks emerge simultaneously. Targeted ablation studies validate our theoretical predictions. This unified view motivates us to propose the Mix-Compress-Refine theory of information flow, as an attempt to explain how LLMs organize their computation in depth by controlling attention and representational compression via massive activations. Specifically, we posit that Transformer-based LLMs process tokens in three distinct phases: (1) broad mixing in the early layers, (2) compressed computation with limited mixing in the middle layers, and (3) selective refinement in the late layers. Our framework helps explain why embedding tasks perform best at intermediate layers, whereas generation tasks benefit from full-depth processing, clarifying differences in task-dependent representations.
Authors: Omer Ben-Porat, Gur Keinan, Rotem Torkan
Abstract: We study an online stochastic matching problem in which an algorithm sequentially matches $U$ users to $K$ arms, aiming to maximize cumulative reward over $T$ rounds under budget constraints. Without structural assumptions, computing the optimal matching is NP-hard, making online learning computationally infeasible. To overcome this barrier, we focus on single-peaked preferences -- a well-established structure in social choice theory, where users' preferences are unimodal with respect to a common order over arms. We devise an efficient algorithm for the offline budgeted matching problem, and leverage it into an efficient online algorithm with a regret of $\tilde O(UKT^{2/3})$. Our approach relies on a novel PQ tree-based order approximation method. If the single-peaked structure is known, we develop an efficient UCB-like algorithm that achieves a regret bound of $\tilde O(U\sqrt{TK})$.
Authors: Konstantinos Barmpas, Na Lee, Alexandros Koliousis, Yannis Panagakis, Dimitrios A. Adamos, Nikolaos Laskaris, Stefanos Zafeiriou
Abstract: Electroencephalography (EEG) captures neural activity across multiple temporal and spectral scales, yielding signals that are rich but complex for representation learning. Recently, EEG foundation models trained to predict masked signal-tokens have shown promise for learning generalizable representations. However, their performance is hindered by their signal tokenization modules. Existing neural tokenizers fail to preserve high-frequency dynamics, limiting their ability to reconstruct EEG signals with high fidelity. We introduce NeuroRVQ, a scalable Large Brainwave Model (LBM) centered on a codebook-based tokenizer. Our tokenizer integrates: (i) multi-scale feature extraction modules that capture the full frequency neural spectrum; (ii) hierarchical residual vector quantization (RVQ) codebooks for high-resolution encoding; and, (iii) an EEG signal phase- and amplitude-aware loss function for efficient training. This design enables efficient EEG compression while supporting accurate reconstruction across all frequency bands, leading to robust generative masked modeling. Our empirical results demonstrate that NeuroRVQ achieves lower reconstruction error and outperforms existing LBMs on a variety of downstream tasks. More broadly, NeuroRVQ tokenizer establishes a strong prior for codebook-based general-purpose brainwave models, enabling advances in neural decoding, generative modeling and multimodal biosignal integration.
Authors: Rahul Vaze, Sumiran Mishra
Abstract: In this paper, we broaden the horizon of online convex optimization (OCO), and consider multi-objective OCO, where there are $K$ distinct loss function sequences, and an algorithm has to choose its action at time $t$, before the $K$ loss functions at time $t$ are revealed. To capture the tradeoff between tracking the $K$ different sequences, we consider the {\it min-max} regret, where the benchmark (optimal offline algorithm) takes a static action across all time slots that minimizes the maximum of the total loss (summed across time slots) incurred by each of the $K$ sequences. An online algorithm is allowed to change its action across time slots, and its {\it min-max} regret is defined as the difference between its {\it min-max} cost and that of the benchmark. The {\it min-max} regret is a stringent performance measure and an algorithm with small regret needs to `track' all loss functions simultaneously. We first show that with adversarial input, {\it min-max} regret scales linearly with the time horizon $T$ for any online algorithm. Consequently, we consider a stochastic i.i.d. input model where all loss functions are i.i.d. generated from an unknown joint distribution and propose a simple algorithm that combines the well-known {\it Hedge} and online gradient descent (OGD) and show via a remarkably simple proof that its expected {\it min-max} regret is $O(\sqrt{T \log (T K)})$. Analogous results are also derived for Martingale difference and Markov input models.
Authors: Jongmin Lee, Ernest K. Ryu
Abstract: The classical policy gradient method is the theoretical and conceptual foundation of modern policy-based reinforcement learning (RL) algorithms. Most rigorous analyses of such methods, particularly those establishing convergence guarantees, assume a discount factor $\gamma < 1$. In contrast, however, a recent line of work on policy-based RL for large language models uses the undiscounted total-reward setting with $\gamma = 1$, rendering much of the existing theory inapplicable. In this paper, we provide analyses of the policy gradient method for undiscounted expected total-reward infinite-horizon MDPs based on two key insights: (i) the classification of the MDP states into recurrent and transient states is invariant over the set of policies that assign strictly positive probability to every action (as is typical in deep RL models employing a softmax output layer) and (ii) the classical state visitation measure (which may be ill-defined when $\gamma = 1$) can be replaced with a new object that we call the transient visitation measure.
Authors: Zhaocheng Liu, Zhiwen Yu, Xiaoqing Liu
Abstract: Multimodal learning integrates diverse modalities but suffers from modality imbalance, where dominant modalities suppress weaker ones due to inconsistent convergence rates. Existing methods predominantly rely on static modulation or heuristics, overlooking sample-level distributional variations in prediction bias. Specifically, they fail to distinguish outlier samples where the modality gap is exacerbated by low data quality. We propose a framework to quantitatively diagnose and dynamically mitigate this imbalance at the sample level. We introduce the Modality Gap metric to quantify prediction discrepancies. Analysis reveals that this gap follows a bimodal distribution, indicating the coexistence of balanced and imbalanced sample subgroups. We employ a Gaussian Mixture Model (GMM) to explicitly model this distribution, leveraging Bayesian posterior probabilities for soft subgroup separation. Our two-stage framework comprises a Warm-up stage and an Adaptive Training stage. In the latter, a GMM-guided Adaptive Loss dynamically reallocates optimization priorities: it imposes stronger alignment penalties on imbalanced samples to rectify bias, while prioritizing fusion for balanced samples to maximize complementary information. Experiments on CREMA-D, AVE, and KineticSound demonstrate that our method significantly outperforms SOTA baselines. Furthermore, we show that fine-tuning on a GMM-filtered balanced subset serves as an effective data purification strategy, yielding substantial gains by eliminating extreme noisy samples even without the adaptive loss.
Authors: Yuxuan Tang, Yifan Feng
Abstract: Alignment of large language models (LLMs) has predominantly relied on pairwise preference optimization, where annotators select the better of two responses to a prompt. While simple, this approach overlooks the opportunity to learn from richer forms of human feedback, such as multiway comparisons and top-$k$ rankings. We introduce Ranked Choice Preference Optimization (RCPO), a unified framework that bridges preference optimization with (ranked) choice modeling via maximum likelihood estimation. RCPO supports both utility-based and rank-based models, subsumes several pairwise methods (such as DPO and SimPO) as special cases, and provides principled training objectives for richer feedback formats. We instantiate this framework with two representative models (Multinomial Logit and Mallows-RMJ). Experiments on Llama-3-8B-Instruct, Gemma-2-9B-it, and Mistral-7B-Instruct across in-distribution and out-of-distribution settings show that RCPO consistently outperforms competitive baselines. RCPO shows that directly leveraging ranked preference data, combined with the right choice models, yields more effective alignment. It offers an extensible foundation for incorporating (ranked) choice modeling into LLM training.
Authors: Hikaru Homma, Jun Ohkubo
Abstract: Initialization of neural network parameters, such as weights and biases, has a crucial impact on learning performance; if chosen well, we can even avoid the need for additional training with backpropagation. For example, algorithms based on the ridgelet transform or the SWIM (sampling where it matters) concept have been proposed for initialization. On the other hand, some works show hierarchical features in trained neural networks; neural networks tend to learn coarse information in the early-stage hidden layers. In this work, we investigate the effects of utilizing information on the hierarchical features in the initialization of neural networks. Hence, we propose a framework that adjusts the scale factors in the SWIM algorithm to capture low-frequency components in the early-stage hidden layers and to represent high-frequency components in the late-stage hidden layers. Numerical experiments on a one-dimensional regression task and the MNIST classification task demonstrate that the proposed method outperforms the conventional initialization algorithms. This work clarifies the importance of intrinsic hierarchical features in learning neural networks, and the finding yields an effective parameter initialization strategy that enhances their training performance.
Authors: Yinsicheng Jiang, Yeqi Huang, Liang Cheng, Cheng Deng, Xuan Sun, Luo Mai
Abstract: Retrieval-augmented generation (RAG) enhances large language models (LLMs) with retrieved context but often suffers from downgraded prefill performance as modern applications demand longer and more complex inputs. Existing caching techniques either preserve accuracy with low cache reuse or improve reuse at the cost of degraded reasoning quality. We present RAGBoost, an efficient RAG system that achieves high cache reuse without sacrificing accuracy through accuracy-preserving context reuse. RAGBoost detects overlapping retrieved items across concurrent sessions and multi-turn interactions, using efficient context indexing, ordering, and de-duplication to maximize reuse, while lightweight contextual hints maintain reasoning fidelity. It integrates seamlessly with existing LLM inference engines and improves their prefill performance by 1.5-3X over state-of-the-art methods, while preserving or even enhancing reasoning accuracy across diverse RAG and agentic AI workloads. Our code is released at: https://github.com/Edinburgh-AgenticAI/RAGBoost.
Authors: Zhehao Xu, Lok Ming Lui
Abstract: Free-boundary diffeomorphism optimization, an important and widely occurring task in geometric modeling, computer graphics, and biological imaging, requires simultaneously determining a planar target domain and a locally bijective map with well-controlled distortion. We formulate this task through the least-squares quasiconformal (LSQC) operator and establish key structural properties of the LSQC minimizer, including well-posedness under mild conditions, invariance under similarity transformations, and resolution-independent behavior with stability under mesh refinement. We further analyze the sensitivity of the LSQC solution with respect to the Beltrami coefficient, establishing stability and differentiability properties that enable gradient-based optimization over the space of Beltrami coefficients. To make this differentiable formulation practical at scale and to facilitate the optimization process, we introduce the Spectral Beltrami Network (SBN), a multiscale mesh-spectral surrogate that approximates the LSQC solution operator in a single differentiable forward pass. This yields SBN-Opt, an optimization framework that searches over admissible Beltrami coefficients and pinning conditions to solve free-boundary diffeomorphism objectives with explicit distortion control. Extensive experiments on equiareal parameterization and inconsistent surface registration demonstrate consistent improvements over traditional numerical algorithms.
Authors: Ankush Kadu, Ashwanth Krishnan
Abstract: Scaling inference-time compute has emerged as a powerful paradigm--yet deliberating longer is not the same as learning. Current approaches to extended reasoning in large language models allocate more computation to thinking but remain fundamentally static: they cannot adapt from mistakes encountered during execution. Online reinforcement learning offers adaptation but requires gradient updates at runtime--expensive, prone to catastrophic forgetting, and unstable in deployment. We introduce ReflexGrad, a gradient-free framework for genuine inference-time learning: adaptation without retraining, without weight updates, without demonstrations. Our key insight is that effective runtime learning requires two complementary mechanisms--rapid policy refinement during forward progress, and deliberate causal diagnosis when stuck--with intelligent routing between them. ReflexGrad implements this by optimizing a natural language "policy" through textual feedback while keeping model weights frozen. When failures occur, the system analyzes recent action-outcome sequences to identify root causes and immediately applies corrections within the same execution--eliminating the need for multiple trials. Evaluated zero-shot across diverse interactive tasks without task-specific engineering, ReflexGrad achieves strong single-execution performance, demonstrating that gradient-free inference-time learning is not just theoretically appealing but practically viable.
Authors: Xiwen Huang, Pierre Pinson
Abstract: We introduce and analyse active learning markets as a way to purchase labels, in situations where analysts aim to acquire additional data to improve model fitting, or to better train models for predictive analytics applications. This comes in contrast to the many proposals that already exist to purchase features and examples. By originally formalising the market clearing as an optimisation problem, we integrate budget constraints and improvement thresholds into the label acquisition process. We focus on a single-buyer-multiple-seller setup and propose the use of two active learning strategies (variance based and query-by-committee based), paired with distinct pricing mechanisms. They are compared to benchmark baselines including random sampling and a greedy knapsack heuristic. The proposed strategies are validated on real-world datasets from two critical application domains: real estate pricing and energy forecasting. Results demonstrate the robustness of our approach, consistently achieving superior performance with fewer labels acquired compared to conventional methods. Our proposal comprises an easy-to-implement practical solution for optimising data acquisition in resource-constrained environments.
Authors: Preksha Girish, Rachana Mysore, Kiran K. N., Hiranmayee R., Shipra Prashanth, Shrey Kumar
Abstract: Understanding the dynamic reorganization of brain networks is critical for predicting cognitive decline, neurological progression, and individual variability in clinical outcomes. This work proposes a multimodal graph neural network framework that integrates structural MRI, diffusion tensor imaging, and functional MRI to model spatiotemporal brain network reorganization. Brain regions are represented as nodes and structural and functional connectivity as edges, forming longitudinal brain graphs for each subject. Temporal evolution is captured via fractional stochastic differential operators embedded within graph-based recurrent networks, enabling the modeling of long-term dependencies and stochastic fluctuations in network dynamics. Attention mechanisms fuse multimodal information and generate interpretable biomarkers, including network energy entropy, graph curvature, fractional memory indices, and modality-specific attention scores. These biomarkers are combined into a composite prognostic index to quantify individual risk of network instability or cognitive decline. Experiments on longitudinal neuroimaging datasets demonstrate both predictive accuracy and interpretability. The results highlight the potential of mathematically rigorous, multimodal graph-based approaches for deriving clinically meaningful biomarkers from existing imaging data without requiring new data collection.
Authors: Abdul Matin, Rupasree Dey, Tanjim Bin Faruk, Shrideep Pallickara, Sangmi Lee Pallickara
Abstract: Integrating domain knowledge into deep learning has emerged as a promising direction for improving model interpretability, generalization, and data efficiency. In this work, we present a novel knowledge-guided ViT-based Masked Autoencoder that embeds scientific domain knowledge within the self-supervised reconstruction process. Instead of relying solely on data-driven optimization, our proposed approach incorporates the Linear Spectral Mixing Model (LSMM) as a physical constraint and physically-based Spectral Angle Mapper (SAM), ensuring that learned representations adhere to known structural relationships between observed signals and their latent components. The framework jointly optimizes LSMM and SAM loss with a conventional Huber loss objective, promoting both numerical accuracy and geometric consistency in the feature space. This knowledge-guided design enhances reconstruction fidelity, stabilizes training under limited supervision, and yields interpretable latent representations grounded in physical principles. The experimental findings indicate that the proposed model substantially enhances reconstruction quality and improves downstream task performance, highlighting the promise of embedding physics-informed inductive biases within transformer-based self-supervised learning.
Authors: Robert Reed, Luca Laurenti, Morteza Lahijanian
Abstract: Finite Abstraction methods provide a powerful formal framework for proving that systems satisfy their specifications. However, these techniques face scalability challenges for high-dimensional systems, as they rely on state-space discretization which grows exponentially with dimension. Learning-based approaches to dimensionality reduction, utilizing neural networks and autoencoders, have shown great potential to alleviate this problem. However, ensuring the correctness of the resulting verification results remains an open question. In this work, we provide a formal approach to reduce the dimensionality of systems via convex autoencoders and learn the dynamics in the latent space through a kernel-based method. We then construct a finite abstraction from the learned model in the latent space and guarantee that the abstraction contains the true behaviors of the original system. We show that the verification results in the latent space can be mapped back to the original system. Finally, we demonstrate the effectiveness of our approach on multiple systems, including a 26D system controlled by a neural network, showing significant scalability improvements without loss of rigor.
Authors: Turja Kundu, Sanjukta Bhowmick
Abstract: Graph neural networks (GNNs) excel on homophilic graphs where connected nodes share labels, but struggle with heterophilic graphs where edges do not imply similarity. Moreover, iterative message passing limits scalability due to neighborhood expansion overhead. We introduce ATLAS (Adaptive Topology-based Learning at Scale), a propagation-free framework that encodes graph structure through multi-resolution community features rather than message passing. We first prove that community refinement involves a fundamental trade-off: finer partitions increase label-community mutual information but also increase entropy. We formalize when refinement improves normalized mutual information, explaining why intermediate granularities are often most predictive. ATLAS employs modularity-guided adaptive search to automatically identify informative community scales, which are one-hot encoded, projected into learnable embeddings, and concatenated with node attributes for MLP classification. This enables standard mini-batch training and adjacency-free inference after one-time preprocessing. Across 13 benchmarks including million-node graphs, ATLAS achieves competitive or superior accuracy, up to 20-point gains over GCN on heterophilic datasets and 12-point gains over MLPs on homophilic graphs. By treating topology as explicit features, ATLAS adapts intelligently: leveraging structure when informative, remaining robust when weakly aligned, and avoiding propagation when structure misleads, providing both scalable performance and interpretable structural insights.
Authors: Chenggong Zhang
Abstract: Partial Differential Equations (PDEs) are central to modeling complex systems across physical, biological, and engineering domains, yet traditional numerical methods often struggle with high-dimensional or complex problems. Physics-Informed Neural Networks (PINNs) have emerged as an efficient alternative by embedding physics-based constraints into deep learning frameworks, but they face challenges in achieving high accuracy and handling complex boundary conditions. In this work, we extend the Time-Evolving Natural Gradient (TENG) framework to address Dirichlet boundary conditions, integrating natural gradient optimization with numerical time-stepping schemes, including Euler and Heun methods, to ensure both stability and accuracy. By incorporating boundary condition penalty terms into the loss function, the proposed approach enables precise enforcement of Dirichlet constraints. Experiments on the heat equation demonstrate the superior accuracy of the Heun method due to its second-order corrections and the computational efficiency of the Euler method for simpler scenarios. This work establishes a foundation for extending the framework to Neumann and mixed boundary conditions, as well as broader classes of PDEs, advancing the applicability of neural network-based solvers for real-world problems.
Authors: Xueming Yan, Boyan Xu, Yaochu Jin, Lixian Xiao, Wenlong Ye, Runyang Cai, Zeqi Zheng, Jingfa Liu, Aimin Yang, Yongduan Song
Abstract: Indonesian, spoken by over 200 million people, remains underserved in multimodal emotion recognition research despite its dominant presence on Southeast Asian social media platforms. We introduce IndoMER, the first multimodal emotion recognition benchmark for Indonesian, comprising 1,944 video segments from 203 speakers with temporally aligned text, audio, and visual annotations across seven emotion categories. The dataset exhibits realistic challenges including cross-modal inconsistency and long-tailed class distributions shaped by Indonesian cultural communication norms. To address these challenges, we propose OmniMER, a multimodal adaptation framework built upon Qwen2.5-Omni that enhances emotion recognition through three auxiliary modality-specific perception tasks: emotion keyword extraction for text, facial expression analysis for video, and prosody analysis for audio. These auxiliary tasks help the model identify emotion-relevant cues in each modality before fusion, reducing reliance on spurious correlations in low-resource settings. Experiments on IndoMER show that OmniMER achieves 0.582 Macro-F1 on sentiment classification and 0.454 on emotion recognition, outperforming the base model by 7.6 and 22.1 absolute points respectively. Cross-lingual evaluation on the Chinese CH-SIMS dataset further demonstrates the generalizability of the proposed framework. The dataset and code are publicly available. https://github.com/yanxm01/INDOMER
Authors: Nina Fatehi, Antar Kumar Biswas, Masoud H. Nazari
Abstract: This paper presents a novel learning based framework for predicting power outages caused by extreme events. The proposed approach targets low-probability high-consequence outage scenarios and leverages a comprehensive set of features derived from publicly available data sources. We integrate EAGLE-I outage records from 2014 to 2024 with weather, socioeconomic, infrastructure, and seasonal event data. Incorporating social and demographic indicators reveals patterns of community vulnerability and improves understanding of outage risk during extreme conditions. Four machine learning models are evaluated, including Random Forest (RF), Graph Neural Network (GNN), Adaptive Boosting (AdaBoost), and Long Short-Term Memory (LSTM). Experimental validation is performed on a large-scale dataset covering counties in the lower peninsula of Michigan. Among all models tested, the LSTM network achieves higher accuracy.
Authors: Yi-Han Wang, Peng Zhao, Zhi-Hua Zhou
Abstract: Online eXp-concave Optimization (OXO) is a fundamental problem in online learning, where the goal is to minimize regret when loss functions are exponentially concave. The standard algorithm, Online Newton Step (ONS), guarantees an optimal $O(d \log T)$ regret, where $d$ is the dimension and $T$ is the time horizon. Despite its simplicity, ONS may face a computational bottleneck due to the Mahalanobis projection at each round. This step costs $\Omega(d^\omega)$ arithmetic operations for bounded domains, even for simple domains such as the unit ball, where $\omega \in (2,3]$ is the matrix-multiplication exponent. As a result, the total runtime can reach $\tilde{O}(d^\omega T)$, particularly when iterates frequently oscillate near the domain boundary. This paper proposes a simple variant of ONS, called LightONS, which reduces the total runtime to $O(d^2 T + d^\omega \sqrt{T \log T})$ while preserving the optimal regret. Deploying LightONS with the online-to-batch conversion implies a method for stochastic exp-concave optimization with runtime $\tilde{O}(d^3/\epsilon)$, thereby answering an open problem posed by Koren [2013]. The design leverages domain-conversion techniques from parameter-free online learning and defers expensive Mahalanobis projections until necessary, thereby preserving the elegant structure of ONS and enabling LightONS to act as an efficient plug-in replacement in broader scenarios, including gradient-norm adaptivity, parametric stochastic bandits, and memory-efficient OXO.
Authors: Bryon Tjanaka, Henry Chen, Matthew C. Fontaine, Stefanos Nikolaidis
Abstract: Quality diversity (QD) optimization searches for a collection of solutions that optimize an objective while attaining diverse outputs of a user-specified, vector-valued measure function. Contemporary QD algorithms are typically limited to low-dimensional measures because high-dimensional measures are prone to distortion, where many solutions found by the QD algorithm map to similar measures. For example, the state-of-the-art CMA-MAE algorithm guides measure space exploration with a histogram in measure space that records so-called discount values. However, CMA-MAE stagnates in domains with high-dimensional measure spaces because solutions with similar measures fall into the same histogram cell and hence receive the same discount value. To address these limitations, we propose Discount Model Search (DMS), which guides exploration with a model that provides a smooth, continuous representation of discount values. In high-dimensional measure spaces, this model enables DMS to distinguish between solutions with similar measures and thus continue exploration. We show that DMS facilitates new capabilities for QD algorithms by introducing two new domains where the measure space is the high-dimensional space of images, which enables users to specify their desired measures by providing a dataset of images rather than hand-designing the measure function. Results in these domains and on high-dimensional benchmarks show that DMS outperforms CMA-MAE and other existing black-box QD algorithms.
Authors: Dylan Sam, Sachin Goyal, Pratyush Maini, Alexander Robey, J. Zico Kolter
Abstract: Prior work has shown that safety interventions applied during pretraining, such as removing and rephrasing harmful content, can substantially improve the robustness of the resulting models. In this paper, we study the fundamental question that prior work has overlooked: "When during pretraining should safety interventions be introduced?" We keep the underlying data sources and pretraining interventions fixed, varying the intervention start time (after 0%, 20%, or 60% of pretraining tokens). We find that the optimal start time is not one-size-fits-all: with standard top-k decoding, introducing interventions after a short initial phase of safe-only pretraining (20%-60%) often yields the strongest robustness, with the clearest benefits emerging after downstream, benign finetuning. In contrast, for safety-aware inference, interventions starting from the beginning improve steerability towards safer generations. Finally, we observe that earlier interventions reshape internal representations: linear probes more cleanly separate safe vs harmful examples. Our results are the first to establish intervention timing as a key curriculum design choice for safety.
Authors: Saurish Nagrath, Saroj Kumar Panigrahy
Abstract: Transformer-based models have shown strong performance in time-series forecasting by leveraging self-attention to model long-range temporal dependencies. However, their effectiveness depends critically on the quality and structure of input representations derived from raw multivariate time-series data, particularly as sequence length and data scale increase. This paper proposes a two-stage forecasting framework that explicitly separates local temporal representation learning from global dependency modelling. In the proposed approach, a convolutional neural network operates on fixed-length temporal patches to extract short-range temporal dynamics and non-linear feature interactions, producing compact patch-level token embeddings. Token-level self-attention is applied during representation learning to refine these embeddings, after which a Transformer encoder models inter-patch temporal dependencies to generate forecasts. The method is evaluated on a synthetic multivariate time-series dataset with controlled static and dynamic factors, using an extended sequence length and a larger number of samples. Experimental results demonstrate that the proposed framework consistently outperforms a convolutional baseline under increased temporal context and remains competitive with a strong patch-based Transformer model. These findings indicate that structured patch-level tokenization provides a scalable and effective representation for multivariate time-series forecasting, particularly when longer input sequences are considered.
Authors: Gorgi Pavlov
Abstract: Learning precise Boolean logic via gradient descent remains challenging: neural networks typically converge to "fuzzy" approximations that degrade under quantization. We introduce Hierarchical Spectral Composition, a differentiable architecture that selects spectral coefficients from a frozen Boolean Fourier basis and composes them via Sinkhorn-constrained routing with column-sign modulation. Our approach draws on recent insights from Manifold-Constrained Hyper-Connections (mHC), which demonstrated that projecting routing matrices onto the Birkhoff polytope preserves identity mappings and stabilizes large-scale training. We adapt this framework to logic synthesis, adding column-sign modulation to enable Boolean negation -- a capability absent in standard doubly stochastic routing. We validate our approach across four phases of increasing complexity: (1) For n=2 (16 Boolean operations over 4-dim basis), gradient descent achieves 100% accuracy with zero routing drift and zero-loss quantization to ternary masks. (2) For n=3 (10 three-variable operations), gradient descent achieves 76% accuracy, but exhaustive enumeration over 3^8 = 6561 configurations proves that optimal ternary masks exist for all operations (100% accuracy, 39% sparsity). (3) For n=4 (10 four-variable operations over 16-dim basis), spectral synthesis -- combining exact Walsh-Hadamard coefficients, ternary quantization, and MCMC refinement with parallel tempering -- achieves 100% accuracy on all operations. This progression establishes (a) that ternary polynomial threshold representations exist for all tested functions, and (b) that finding them requires methods beyond pure gradient descent as dimensionality grows. All operations enable single-cycle combinational logic inference at 10,959 MOps/s on GPU, demonstrating viability for hardware-efficient neuro-symbolic logic synthesis.
Authors: Wei Lin, Yining Jiang, Qingyu Song, Qiao Xiang, Hong Xu
Abstract: Zeroth-Order (ZO) optimization has emerged as a promising solution for fine-tuning LLMs under strict memory constraints, as it avoids the prohibitive memory cost of storing activations for backpropagation. However, existing ZO methods typically employ isotropic perturbations, neglecting the rich structural information available during the forward pass. In this paper, we identify a crucial link between gradient formation and activation structure: the gradient of a linear layer is confined to the subspace spanned by its input activations. Leveraging this insight, we propose Activation-Guided Zeroth-Order optimization (AGZO). Unlike prior methods, AGZO extracts a compact, activation-informed subspace on the fly during the forward pass and restricts perturbations to this low-rank subspace. We provide a theoretical framework showing that AGZO optimizes a subspace-smoothed objective and provably yields update directions with higher cosine similarity to the true gradient than isotropic baselines. Empirically, we evaluate AGZO on Qwen3 and Pangu models across various benchmarks. AGZO consistently outperforms state-of-the-art ZO baselines and significantly narrows the performance gap with first-order fine-tuning, while maintaining almost the same peak memory footprint as other ZO methods.
Authors: Rohan Asthana, Vasileios Belagiannis
Abstract: Diffusion-based image generative models produce high-fidelity images through iterative denoising but remain vulnerable to memorization, where they unintentionally reproduce exact copies or parts of training images. Recent memorization detection methods are primarily based on the norm of score difference as indicators of memorization. We prove that such norm-based metrics are mainly effective under the assumption of isotropic log-probability distributions, which generally holds at high or medium noise levels. In contrast, analyzing the anisotropic regime reveals that memorized samples exhibit strong angular alignment between the guidance vector and unconditional scores in the low-noise setting. Through these insights, we develop a memorization detection metric by integrating isotropic norm and anisotropic alignment. Our detection metric can be computed directly on pure noise inputs via two conditional and unconditional forward passes, eliminating the need for costly denoising steps. Detection experiments on Stable Diffusion v1.4 and v2 show that our metric outperforms existing denoising-free detection methods while being at least approximately 5x faster than the previous best approach. Finally, we demonstrate the effectiveness of our approach by utilizing a mitigation strategy that adapts memorized prompts based on our developed metric. The code is available at https://github.com/rohanasthana/memorization-anisotropy .
URLs: https://github.com/rohanasthana/memorization-anisotropy
Authors: Nikita Makarov, Maria Bordukova, Lena Voith von Voithenberg, Estrella Pivel-Villanueva, Sabrina Mielke, Jonathan Wickes, Hanchen Wang, Mingyu Derek Ma, Keunwoo Choi, Kyunghyun Cho, Stephen Ra, Raul Rodriguez-Esteban, Fabian Schmich, Michael Menden
Abstract: Precision oncology requires forecasting clinical events and trajectories, yet modeling sparse, multi-modal clinical time series remains a critical challenge. We introduce TwinWeaver, an open-source framework that serializes longitudinal patient histories into text, enabling unified event prediction as well as forecasting with large language models, and use it to build Genie Digital Twin (GDT) on 93,054 patients across 20 cancer types. In benchmarks, GDT significantly reduces forecasting error, achieving a median Mean Absolute Scaled Error (MASE) of 0.87 compared to 0.97 for the strongest time-series baseline (p<0.001). Furthermore, GDT improves risk stratification, achieving an average concordance index (C-index) of 0.703 across survival, progression, and therapy switching tasks, surpassing the best baseline of 0.662. GDT also generalizes to out-of-distribution clinical trials, matching trained baselines at zero-shot and surpassing them with fine-tuning, achieving a median MASE of 0.75-0.88 and outperforming the strongest baseline in event prediction with an average C-index of 0.672 versus 0.648. Finally, TwinWeaver enables an interpretable clinical reasoning extension, providing a scalable and transparent foundation for longitudinal clinical modeling.
Authors: Sichen Zhao, Zhiming Xue, Yalun Qi, Xianling Zeng, Zihan Yu
Abstract: Malicious bots pose a growing threat to e-commerce platforms by scraping data, hoarding inventory, and perpetrating fraud. Traditional bot mitigation techniques, including IP blacklists and CAPTCHA-based challenges, are increasingly ineffective or intrusive, as modern bots leverage proxies, botnets, and AI-assisted evasion strategies. This work proposes a non-intrusive graph-based bot detection framework for e-commerce that models user session behavior through a graph representation and applies an inductive graph neural network for classification. The approach captures both relational structure and behavioral semantics, enabling accurate identification of subtle automated activity that evades feature-based methods. Experiments on real-world e-commerce traffic demonstrate that the proposed inductive graph model outperforms a strong session-level multilayer perceptron baseline in terms of AUC and F1 score. Additional adversarial perturbation and cold-start simulations show that the model remains robust under moderate graph modifications and generalizes effectively to previously unseen sessions and URLs. The proposed framework is deployment-friendly, integrates with existing systems without client-side instrumentation, and supports real-time inference and incremental updates, making it suitable for practical e-commerce security deployments.
Authors: Jie Xiao, Meng Chen, Qingnan Ren, Jingwei Song, Jiaqi Huang, Yangshen Deng, Chris Tong, Wanyi Chen, Suli Wang, Ziqian Bi, Shuo Lu, Yiqun Duan, Xu Wang, Rymon Yu, Ween Yang, Lynn Ai, Eric Yang, Bill Shi
Abstract: Reinforcement learning (RL) is a critical stage in post-training large language models (LLMs), involving repeated interaction between rollout generation, reward evaluation, and centralized learning. Distributing rollout execution offers opportunities to leverage more cost-efficient inference resources, but introduces challenges in wide-area coordination and policy dissemination. We present ECHO-2, a distributed RL framework for post-training with remote inference workers and non-negligible dissemination latency. ECHO-2 combines centralized learning with distributed rollouts and treats bounded policy staleness as a user-controlled parameter, enabling rollout generation, dissemination, and training to overlap. We introduce an overlap-based capacity model that relates training time, dissemination latency, and rollout throughput, yielding a practical provisioning rule for sustaining learner utilization. To mitigate dissemination bottlenecks and lower cost, ECHO-2 employs peer-assisted pipelined broadcast and cost-aware activation of heterogeneous workers. Experiments on GRPO post-training of 4B and 8B models under real wide-area bandwidth regimes show that ECHO-2 significantly improves cost efficiency while preserving RL reward comparable to strong baselines.
Authors: Mingda Liu, Zhenghan Zhu, Ze'an Miao, Katsuki Fujisawa
Abstract: Model editing has emerged as a practical approach for mitigating factual errors and outdated knowledge in large language models (LLMs). Among existing methods, the Locate-and-Edit (L&E) paradigm is the dominant framework: it locates MLP parameters implicated in expressing a target fact, and then performs a localized update to rewrite that fact. However, long sequences of edits often trigger abrupt model collapse in L&E beyond a critical point. We empirically identify a strong correlation between collapse and explosive growth of edited MLP weight norms, and formally prove that commonly used L&E update rules can induce exponential norm growth across sequential edits in the absence of explicit norm control. To address this issue, we propose Norm-Anchor Scaling NAS, a plug-and-play norm-constrained strategy. Across extensive experiments, NAS delays the collapse point of representative L&E algorithms by more than 4 times and yields a 72.2% average relative gain in editing performance, requiring only a single additional line of code and incurring negligible computational overhead.
Authors: Jihao Xin, Tian Lyu, David Keyes, Hatem Ltaief, Marco Canini
Abstract: Long-context inference in large language models is increasingly bottlenecked by the memory and compute cost of the KV-Cache. Low-rank factorization compresses KV projections by writing $W \approx A * B$, where A produces latent KV states and B can be absorbed into downstream weights. In modern RoPE-based LLMs, this absorption fails: RoPE forces latent KV states to be reconstructed to full dimension, reintroducing substantial memory and compute overhead. We propose RoPE-Aligned Pruning (RAP), which prunes entire RoPE-aligned column pairs to preserve RoPE's 2x2 rotation structure, restore B absorption, and eliminate reconstruction. Our evaluation on LLaMA-3-8B and Mistral-7B shows that RAP enables joint reduction of KV-Cache, attention parameters, and FLOPs by 20-30%, all at once, while maintaining strong accuracy. Notably, RAP reduces attention latency to 83% (prefill) and 77% (decode) of baseline.
Authors: Jack Michael Solomon, Rishi Leburu, Matthias Chung
Abstract: Inverse problems are fundamental to many scientific and engineering disciplines; they arise when one seeks to reconstruct hidden, underlying quantities from noisy measurements. Many applications demand not just point estimates but interpretable uncertainty. Providing fast inference alongside uncertainty estimates remains challenging yet desirable in numerous applications. We propose the Variational Sparse Paired Autoencoder (vsPAIR) to address this challenge. The architecture pairs a standard VAE encoding observations with a sparse VAE encoding quantities of interest, connected through a learned latent mapping. The variational structure enables uncertainty estimation, the paired architecture encourages interpretability by anchoring QoI representations to clean data, and sparse encodings provide structure by concentrating information into identifiable factors rather than diffusing across all dimensions. To validate the effectiveness of our proposed architecture, we conduct experiments on blind inpainting and computed tomography, demonstrating that vsPAIR is a capable inverse problem solver that can provide interpretable and structured uncertainty estimates.
Authors: Felix X. -F. Ye, Xingjie Li, An Yu, Ming-Ching Chang, Linsong Chu, Davis Wertheimer
Abstract: Entropic optimal transport (EOT) via Sinkhorn iterations is widely used in modern machine learning, yet GPU solvers remain inefficient at scale. Tensorized implementations suffer quadratic HBM traffic from dense $n\times m$ interactions, while existing online backends avoid storing dense matrices but still rely on generic tiled map-reduce reduction kernels with limited fusion. We present \textbf{FlashSinkhorn}, an IO-aware EOT solver for squared Euclidean cost that rewrites stabilized log-domain Sinkhorn updates as row-wise LogSumExp reductions of biased dot-product scores, the same normalization as transformer attention. This enables FlashAttention-style fusion and tiling: fused Triton kernels stream tiles through on-chip SRAM and update dual potentials in a single pass, substantially reducing HBM IO per iteration while retaining linear-memory operations. We further provide streaming kernels for transport application, enabling scalable first- and second-order optimization. On A100 GPUs, FlashSinkhorn achieves up to $32\times$ forward-pass and $161\times$ end-to-end speedups over state-of-the-art online baselines on point-cloud OT, improves scalability on OT-based downstream tasks. For reproducibility, we release an open-source implementation at https://github.com/ot-triton-lab/ot_triton.
Authors: Yao-Hui Li, Zeyu Wang, Xin Li, Wei Pang, Yingfang Yuan, Zhengkun Chen, Boya Zhang, Riashat Islam, Alex Lamb, Yonggang Zhang
Abstract: Model-based reinforcement learning (MBRL) achieves high sample efficiency by simulating future trajectories with learned dynamics and reward models. However, its effectiveness is severely compromised in sparse reward settings. The core limitation lies in the standard paradigm of regressing ground-truth scalar rewards: in sparse environments, this yields a flat, gradient-free landscape that fails to provide directional guidance for planning. To address this challenge, we propose Shaping Landscapes with Optimistic Potential Estimates (SLOPE), a novel framework that shifts reward modeling from predicting scalars to constructing informative potential landscapes. SLOPE employs optimistic distributional regression to estimate high-confidence upper bounds, which amplifies rare success signals and ensures sufficient exploration gradients. Evaluations on 30+ tasks across 5 benchmarks demonstrate that SLOPE consistently outperforms leading baselines in fully sparse, semi-sparse, and dense rewards.
Authors: Youngjae Cho, Jongsuk Kim, Ji-Hoon Kim
Abstract: Direct Preference Optimization (DPO) and related methods align large language models from pairwise preferences by regularizing updates against a fixed reference policy. As the policy drifts, a static reference, however, can become increasingly miscalibrated, leading to distributional mismatch and amplifying spurious preference signals under noisy supervision. Conversely, reference-free variants avoid mismatch but often suffer from unconstrained reward drift. We propose Geometric Anchor Preference Optimization (GAPO), which replaces the fixed reference with a dynamic, geometry-aware anchor: an adversarial local perturbation of the current policy within a small radius that serves as a pessimistic baseline. This anchor enables an adaptive reweighting mechanism, modulating the importance of each preference pair based on its local sensitivity. We further introduce the Anchor Gap, the reward discrepancy between the policy and its anchor, and show under smoothness conditions that it approximates worst-case local margin degradation. Optimizing a logistic objective weighted by this gap downweights geometrically brittle instances while emphasizing robust preference signals. Across diverse noise settings, GAPO consistently improves robustness while matching or improving performance on standard LLM alignment and reasoning benchmarks.
Authors: Emiliano Penaloza, Dheeraj Vattikonda, Nicolas Gontier, Alexandre Lacoste, Laurent Charlin, Massimo Caccia
Abstract: Training-time privileged information (PI) can enable language models to succeed on tasks they would otherwise fail, making it a powerful tool for reinforcement learning in hard, long-horizon settings. However, transferring capabilities learned with PI to policies that must act without it at inference time remains a fundamental challenge. We study this problem in the context of distilling frontier models for multi-turn agentic environments, which typically hide their internal reasoning and expose only action trajectories. This breaks standard distillation pipelines, since successful behavior is observable, but the reasoning process is not. For this, we introduce {\pi}-Distill, a joint teacher-student objective that trains a PI-conditioned teacher and an unconditioned student simultaneously using the same model. Additionally, we also introduce On-Policy Self-Distillation (OPSD), an alternative approach that trains using Reinforcement Learning (RL) with a reverse KL-penalty between the student and the PI-conditioned teacher. We show that both of these algorithms effectively distill frontier agents using action-only PI. Specifically, we find that {\pi}-Distill and, in some cases, OPSD, outperform industry standard practices (Supervised finetuning followed by RL) that assume access to full Chain-of-Thought supervision across multiple agentic benchmarks, models, and forms of PI. We complement our results with extensive analysis that characterizes the factors enabling effective learning with PI, focusing primarily on {\pi}-Distill and characterizing when OPSD is competitive.
Authors: Lixuan Guo, Yifei Wang, Tiansheng Wen, Yifan Wang, Aosong Feng, Bo Chen, Stefanie Jegelka, Chenyu You
Abstract: In the era of large foundation models, the quality of embeddings has become a central determinant of downstream task performance and overall system capability. Yet widely used dense embeddings are often extremely high-dimensional, incurring substantial costs in storage, memory, and inference latency. To address these, Contrastive Sparse Representation (CSR) is recently proposed as a promising direction, mapping dense embeddings into high-dimensional but k-sparse vectors, in contrast to compact dense embeddings such as Matryoshka Representation Learning (MRL). Despite its promise, CSR suffers severe degradation in the ultra-sparse regime, where over 80% of neurons remain inactive, leaving much of its efficiency potential unrealized. In this paper, we introduce CSRv2, a principled training approach designed to make ultra-sparse embeddings viable. CSRv2 stabilizes sparsity learning through progressive k-annealing, enhances representational quality via supervised contrastive objectives, and ensures end-to-end adaptability with full backbone finetuning. CSRv2 reduces dead neurons from 80% to 20% and delivers a 14% accuracy gain at k=2, bringing ultra-sparse embeddings on par with CSR at k=8 and MRL at 32 dimensions, all with only two active features. While maintaining comparable performance, CSRv2 delivers a 7x speedup over MRL, and yields up to 300x improvements in compute and memory efficiency relative to dense embeddings in text representation. Extensive experiments across text and vision demonstrate that CSRv2 makes ultra-sparse embeddings practical without compromising performance, where CSRv2 achieves 7%/4% improvement over CSR when k=4 and further increases this gap to 14%/6% when k=2 in text/vision representation. By making extreme sparsity viable, CSRv2 broadens the design space for real-time and edge-deployable AI systems where both embedding quality and efficiency are critical.
Authors: Han Li, Letian Zhu, Bohan Zhang, Rili Feng, Jiaming Wang, Yue Pan, Earl T. Barr, Sarro Federica, Zhaoyang Chu, He Ye
Abstract: LLM-based coding agents have shown strong performance on automated issue resolution benchmarks, yet existing evaluations largely focus on final task success, providing limited insight into how agents retrieve and use code context during problem solving. We introduce ContextBench, a process-oriented evaluation of context retrieval in coding agents. ContextBench consists of 1,136 issue-resolution tasks from 66 repositories across eight programming languages, each augmented with human-annotated gold contexts. We further implement an automated evaluation framework that tracks agent trajectories and measures context recall, precision, and efficiency throughout issue resolution. Using ContextBench, we evaluate four frontier LLMs and five coding agents. Our results show that sophisticated agent scaffolding yields only marginal gains in context retrieval ("The Bitter Lesson" of coding agents), LLMs consistently favor recall over precision, and substantial gaps exist between explored and utilized context. ContextBench augments existing end-to-end benchmarks with intermediate gold-context metrics that unbox the issue-resolution process. These contexts offer valuable intermediate signals for guiding LLM reasoning in software tasks.
Authors: Jorge L. Ruiz Williams
Abstract: We present the Condensate Theorem: attention sparsity is a learned topological property, not an architectural constraint. Through empirical analysis of trained language models, we find that attention mass concentrates on a distinct topological manifold -- and this manifold can be identified dynamically without checking every position. We prove a general result: for any query, projecting attention onto the Condensate Manifold (Anchor + Window + Dynamic Top-k) achieves 100% output equivalence with full $O(n^2)$ attention. This is not an approximation -- it is lossless parity. We validate this across GPT-2, Pythia, Qwen2, TinyLlama, and Mistral, demonstrating bit-exact token matching on 1,500+ generated tokens. By mapping this topology to hardware, our Topological Attention kernel achieves a 159x measured speedup at 131K tokens (3.94ms vs 628ms) and a projected >1,200x speedup at 1M tokens, reducing inference costs by >99.9% compared to Flash Attention. We conclude that the quadratic bottleneck is an artifact of naive implementation, not intelligence.
Authors: Thomas Frost, Hrisheekesh Vaidya, Steve Harris
Abstract: Offline reinforcement learning (ORL) has shown potential for improving decision-making in healthcare. However, contemporary research typically aggregates patient data into fixed time intervals, simplifying their mapping to standard ORL frameworks. The impact of these temporal manipulations on model safety and efficacy remains poorly understood. In this work, using both a gridworld navigation task and the UVA/Padova clinical diabetes simulator, we demonstrate that temporal resampling significantly degrades the performance of offline reinforcement learning algorithms during live deployment. We propose three mechanisms that drive this failure: (i) the generation of counterfactual trajectories, (ii) the distortion of temporal expectations, and (iii) the compounding of generalisation errors. Crucially, we find that standard off-policy evaluation metrics can fail to detect these drops in performance. Our findings reveal a fundamental risk in current healthcare ORL pipelines and emphasise the need for methods that explicitly handle the irregular timing of clinical decision-making.
Authors: Jiaming He, Fuming Luo, Hongwei Li, Wenbo Jiang, Wenshu Fan, Zhenbo Shi, Xudong Jiang, Yi Yu
Abstract: Unlearnable examples (UE) have emerged as a practical mechanism to prevent unauthorized model training on private vision data, while extending this protection to tabular data is nontrivial. Tabular data in finance and healthcare is highly sensitive, yet existing UE methods transfer poorly because tabular features mix numerical and categorical constraints and exhibit saliency sparsity, with learning dominated by a few dimensions. Under a Spectral Dominance condition, we show certified unlearnability is feasible when the poison spectrum overwhelms the clean semantic spectrum. Guided by this, we propose Unlearnable Tabular Data via DecOuPled Shortcut EmbeddIng (UTOPIA), which exploits feature redundancy to decouple optimization into two channels: high saliency features for semantic obfuscation and low saliency redundant features for embedding a hyper correlated shortcut, yielding constraint-aware dominant shortcuts while preserving tabular validity. Extensive experiments across tabular datasets and models show UTOPIA drives unauthorized training toward near random performance, outperforming strong UE baselines and transferring well across architectures.
Authors: Martin Fixman, Alessandro Abati, Juli\'an Jim\'enez Nimmo, Sean Lim, Esther Mondrag\'on
Abstract: Simulations are an indispensable step in the cycle of theory development and refinement, helping researchers formulate precise definitions, generate models, and make accurate predictions. This paper introduces the Pavlovian Associative Learning Models Simulator (PALMS), a Python environment to simulate Pavlovian conditioning experiments. In addition to the canonical Rescorla-Wagner model, PALMS incorporates several attentional learning approaches, including Pearce-Kaye-Hall, Mackintosh Extended, Le Pelley's Hybrid, and a novel extension of the Rescorla-Wagner model with a unified variable learning rate that integrates Mackintosh's and Pearce and Hall's opposing conceptualisations. The simulator's graphical interface allows for the input of entire experimental designs in an alphanumeric format, akin to that used by experimental neuroscientists. Moreover, it uniquely enables the simulation of experiments involving hundreds of stimuli, as well as the computation of configural cues and configural-cue compounds across all models, thereby considerably expanding their predictive capabilities. PALMS operates efficiently, providing instant visualisation of results, supporting rapid, precise comparisons of various models' predictions within a single architecture and environment. Furthermore, graphic displays can be easily saved, and simulated data can be exported to spreadsheets. To illustrate the simulator's capabilities and functionalities, we provide a detailed description of the software and examples of use, reproducing published experiments in the associative learning literature. PALMS is licensed under the open-source GNU Lesser General Public License 3.0. The simulator source code and the latest multiplatform release build are accessible as a GitHub repository at https://github.com/cal-r/PALMS-Simulator
Authors: Jianwen Chen, Xinyu Yang, Peng Xia, Arian Azarang, Yueh Z Lee, Gang Li, Hongtu Zhu, Yun Li, Beidi Chen, Huaxiu Yao
Abstract: Large language models (LLMs) have demonstrated strong performance and rapid progress in a wide range of medical reasoning tasks. However, their sequential autoregressive decoding forces inherently parallel clinical reasoning, such as differential diagnosis, into a single linear reasoning path, limiting both efficiency and reliability for complex medical problems. To address this, we propose MedVerse, a reasoning framework for complex medical inference that reformulates medical reasoning as a parallelizable directed acyclic graph (DAG) process based on Petri net theory. The framework adopts a full-stack design across data, model architecture, and system execution. For data creation, we introduce the MedVerse Curator, an automated pipeline that synthesizes knowledge-grounded medical reasoning paths and transforms them into Petri net-structured representations. At the architectural level, we propose a topology-aware attention mechanism with adaptive position indices that supports parallel reasoning while preserving logical consistency. Systematically, we develop a customized inference engine that supports parallel execution without additional overhead. Empirical evaluations show that MedVerse improves strong general-purpose LLMs by up to 8.9%. Compared to specialized medical LLMs, MedVerse achieves comparable performance while delivering a 1.3x reduction in inference latency and a 1.7x increase in generation throughput, enabled by its parallel decoding capability. Code is available at https://github.com/aiming-lab/MedVerse.
Authors: Yanlin Qi, Xinhang Chen, Huiqiang Jiang, Qitong Wang, Botao Peng, Themis Palpanas
Abstract: KV-cache retrieval is essential for long-context LLM inference, yet existing methods struggle with distribution drift and high latency at scale. We introduce ParisKV, a drift-robust, GPU-native KV-cache retrieval framework based on collision-based candidate selection, followed by a quantized inner-product reranking estimator. For million-token contexts, ParisKV supports CPU-offloaded KV caches via Unified Virtual Addressing (UVA), enabling on-demand top-$k$ fetching with minimal overhead. ParisKV matches or outperforms full attention quality on long-input and long-generation benchmarks. It achieves state-of-the-art long-context decoding efficiency: it matches or exceeds full attention speed even at batch size 1 for long contexts, delivers up to 2.8$\times$ higher throughput within full attention's runnable range, and scales to million-token contexts where full attention runs out of memory. At million-token scale, ParisKV reduces decode latency by 17$\times$ and 44$\times$ compared to MagicPIG and PQCache, respectively, two state-of-the-art KV-cache Top-$k$ retrieval baselines.
Authors: Alejandro Ruiz y Mesa, Guilherme Korol, Moritz Riesterer, Jo\~ao Paulo Cardoso de Lima, Jeronimo Castrillon
Abstract: LLM deployment on resource-constrained edge devices faces severe latency constraints, particularly in real-time applications where delayed responses can compromise safety or usability. Among many approaches to mitigate the inefficiencies of sequential token-by-token generation, Speculative Decoding (SD) has emerged as a promising technique. However, SD at the edge is hindered by two major challenges: (1) integrating SD into a compiler-based workflow without sacrificing performance or programmability, and (2) exploiting the heterogeneous compute resources of modern SoCs through carefully designed partitioning strategies. This work addresses these challenges by using an analytical cost model that explores heterogeneous hardware configurations and guides coarse-grained partitioning of LLM subgraphs, particularly with edge-typical short input sequence lengths. The cost model predicts when speculative sampling and heterogeneous execution are jointly beneficial and is validated on an edge device featuring a hexacore Cortex-A CPU and a Mali GPU, revealing up to 1.68$\times$ speedup for translation tasks, closely matching analytic expectations.
Authors: Tiwei Bie, Maosong Cao, Xiang Cao, Bingsen Chen, Fuyuan Chen, Kun Chen, Lun Du, Daozhuo Feng, Haibo Feng, Mingliang Gong, Zhuocheng Gong, Yanmei Gu, Jian Guan, Kaiyuan Guan, Hongliang He, Zenan Huang, Juyong Jiang, Zhonghui Jiang, Zhenzhong Lan, Chengxi Li, Jianguo Li, Zehuan Li, Huabin Liu, Lin Liu, Guoshan Lu, Yuan Lu, Yuxin Ma, Xingyu Mou, Zhenxuan Pan, Kaida Qiu, Yuji Ren, Jianfeng Tan, Yiding Tian, Zian Wang, Lanning Wei, Tao Wu, Yipeng Xing, Wentao Ye, Liangyu Zha, Tianze Zhang, Xiaolu Zhang, Junbo Zhao, Da Zheng, Hao Zhong, Wanli Zhong, Jun Zhou, Junlin Zhou, Liwang Zhu, Muzhi Zhu, Yihong Zhuang
Abstract: While LLaDA2.0 showcased the scaling potential of 100B-level block-diffusion models and their inherent parallelization, the delicate equilibrium between decoding speed and generation quality has remained an elusive frontier. Today, we unveil LLaDA2.1, a paradigm shift designed to transcend this trade-off. By seamlessly weaving Token-to-Token (T2T) editing into the conventional Mask-to-Token (M2T) scheme, we introduce a joint, configurable threshold-decoding scheme. This structural innovation gives rise to two distinct personas: the Speedy Mode (S Mode), which audaciously lowers the M2T threshold to bypass traditional constraints while relying on T2T to refine the output; and the Quality Mode (Q Mode), which leans into conservative thresholds to secure superior benchmark performances with manageable efficiency degrade. Furthering this evolution, underpinned by an expansive context window, we implement the first large-scale Reinforcement Learning (RL) framework specifically tailored for dLLMs, anchored by specialized techniques for stable gradient estimation. This alignment not only sharpens reasoning precision but also elevates instruction-following fidelity, bridging the chasm between diffusion dynamics and complex human intent. We culminate this work by releasing LLaDA2.1-Mini (16B) and LLaDA2.1-Flash (100B). Across 33 rigorous benchmarks, LLaDA2.1 delivers strong task performance and lightning-fast decoding speed. Despite its 100B volume, on coding tasks it attains an astounding 892 TPS on HumanEval+, 801 TPS on BigCodeBench, and 663 TPS on LiveCodeBench.
Authors: Leander Kurscheidt, Gabriele Masina, Roberto Sebastiani, Antonio Vergari
Abstract: In many safety-critical settings, probabilistic ML systems have to make predictions subject to algebraic constraints, e.g., predicting the most likely trajectory that does not cross obstacles. These real-world constraints are rarely convex, nor the densities considered are (log-)concave. This makes computing this constrained maximum a posteriori (MAP) prediction efficiently and reliably extremely challenging. In this paper, we first investigate under which conditions we can perform constrained MAP inference over continuous variables exactly and efficiently and devise a scalable message-passing algorithm for this tractable fragment. Then, we devise a general constrained MAP strategy that interleaves partitioning the domain into convex feasible regions with numerical constrained optimization. We evaluate both methods on synthetic and real-world benchmarks, showing our approaches outperform constraint-agnostic baselines, and scale to complex densities intractable for SoTA exact solvers.
Authors: Duc-Anh Nguyen, Nhien-An Le-Khac
Abstract: Multimodal multiview learning seeks to integrate information from diverse sources to enhance task performance. Existing approaches often struggle with flexible view configurations, including arbitrary view combinations, numbers of views, and heterogeneous modalities. Focusing on the context of human activity recognition, we propose RALIS, a model that combines multiview contrastive learning with a mixture-of-experts module to support arbitrary view availability during both training and inference. Instead of trying to reconstruct missing views, an adjusted center contrastive loss is used for self-supervised representation learning and view alignment, mitigating the impact of missing views on multiview fusion. This loss formulation allows for the integration of view weights to account for view quality. Additionally, it reduces computational complexity from $O(V^2)$ to $O(V)$, where $V$ is the number of views. To address residual discrepancies not captured by contrastive learning, we employ a mixture-of-experts module with a specialized load balancing strategy, tasked with adapting to arbitrary view combinations. We highlight the geometric relationship among components in our model and how they combine well in the latent space. RALIS is validated on four datasets encompassing inertial and human pose modalities, with the number of views ranging from three to nine, demonstrating its performance and flexibility.
Authors: Paul Saegert, Ullrich K\"othe
Abstract: Symbolic regression (SR) aims to discover interpretable analytical expressions that accurately describe observed data. Amortized SR promises to be much more efficient than the predominant genetic programming SR methods, but currently struggles to scale to realistic scientific complexity. We find that a key obstacle is the lack of a fast reduction of equivalent expressions to a concise normalized form. Amortized SR has addressed this by general-purpose Computer Algebra Systems (CAS) like SymPy, but the high computational cost severely limits training and inference speed. We propose SimpliPy, a rule-based simplification engine achieving a 100-fold speed-up over SymPy at comparable quality. This enables substantial improvements in amortized SR, including scalability to much larger training sets, more efficient use of the per-expression token budget, and systematic training set decontamination with respect to equivalent test expressions. We demonstrate these advantages in our Flash-ANSR framework, which achieves much better accuracy than amortized baselines (NeSymReS, E2E) on the FastSRB benchmark. Moreover, it performs on par with state-of-the-art direct optimization (PySR) while recovering more concise instead of more complex expressions with increasing inference budget.
Authors: Armando Angrisani, Brian Coyle, Elham Kashefi
Abstract: Understanding the power of quantum data in machine learning is central to many proposed applications of quantum technologies. While access to quantum data can offer exponential advantages for carefully designed learning tasks and often under strong assumptions on the data distribution, it remains an open question whether such advantages persist in less structured settings and under more realistic, naturally occurring distributions. Motivated by these practical concerns, we introduce a systematic framework based on quantum lossy data compression to bound the power of quantum data in the context of probably approximately correct (PAC) learning. Specifically, we provide lower bounds on the sample complexity of quantum learners for arbitrary functions when data is drawn from Zipf's distribution, a widely used model for the empirical distributions of real-world data. We also establish lower bounds on the size of quantum input data required to learn linear functions, thereby proving the optimality of previous positive results. Beyond learning theory, we show that our framework has applications in secure delegated quantum computation within the measurement-based quantum computation (MBQC) model. In particular, we constrain the amount of private information the server can infer, strengthening the security guarantees of the delegation protocol proposed in (Mantri et al., PRX, 2017).
Authors: Jared Edward Reser
Abstract: This article presents an artificial intelligence (AI) architecture intended to simulate the iterative updating of the human working memory system. It features several interconnected neural networks designed to emulate the specialized modules of the cerebral cortex. These are structured hierarchically and integrated into a global workspace. They are capable of temporarily maintaining high-level representational patterns akin to the psychological items maintained in working memory. This maintenance is made possible by persistent neural activity in the form of two modalities: sustained neural firing (resulting in a focus of attention) and synaptic potentiation (resulting in a short-term store). Representations held in persistent activity are recursively replaced resulting in incremental changes to the content of the working memory system. As this content gradually evolves, successive processing states overlap and are continuous with one another. The present article will explore how this architecture can lead to iterative shift in the distribution of coactive representations, ultimately leading to mental continuity between processing states, and thus to human-like thought and cognition. Taken together, these components outline a biologically motivated route toward synthetic consciousness or artificial sentience and subjectivity.
Authors: Jun Qi, Chao-Han Huck Yang, Pin-Yu Chen, Min-Hsiu Hsieh
Abstract: Data encoding remains a fundamental bottleneck in quantum machine learning, where amplitude encoding of high-dimensional classical vectors into quantum states incurs exponential cost. In this work, we propose a pre-trained tensor-train (TT) encoding network (Pre-TT-Encoder) that significantly reduces the computational complexity of amplitude encoding while preserving essential data structure. The Pre-TT-Encoder exploits low-rank TT decompositions learned from classical data, enabling polynomial-time state preparation in the number of qubits and TT-ranks. We provide a theoretical analysis of the encoding complexity and establish fidelity bounds that quantify the trade-off between TT-rank and approximation error. Empirical evaluations on classical (MNIST) and quantum-native (semiconductor quantum dot) datasets demonstrate that our approach achieves substantial gains in encoding efficiency over direct amplitude encoding and PCA-based dimensionality reduction, while maintaining competitive performance in downstream variational quantum circuit classification tasks. The proposed method highlights the role of tensor networks as scalable intermediaries between classical data and quantum processors.
Authors: Li Jiang, Xiao Li, Andre Milzarek, Junwen Qiu
Abstract: Chung's Lemma is a classical tool for establishing asymptotic convergence rates of (stochastic) optimization methods under strong convexity-type assumptions and appropriate polynomial diminishing step sizes. In this work, we develop a generalized version of Chung's Lemma, which provides a simple non-asymptotic convergence framework for a more general family of step size rules. We demonstrate broad applicability of the proposed generalized lemma by deriving tight non-asymptotic convergence rates for a large variety of stochastic methods. In particular, we obtain partially new non-asymptotic complexity results for stochastic optimization methods, such as Stochastic Gradient Descent (SGD) and Random Reshuffling (RR), under a general $(\theta,\mu)$-Polyak-Lojasiewicz (PL) condition and for various step sizes strategies, including polynomial, constant, exponential, and cosine step sizes rules. Notably, as a by-product of our analysis, we observe that exponential step sizes exhibit superior adaptivity to both landscape geometry and gradient noise; specifically, they achieve optimal convergence rates without requiring exact knowledge of the underlying landscape or separate parameter selection strategies for noisy and noise-free regimes. Our results demonstrate that the developed variant of Chung's Lemma offers a versatile, systematic, and streamlined approach to establish non-asymptotic convergence rates under general step size rules.
Authors: Xiwei Cheng, Xiangxin Zhou, Yuwei Yang, Yu Bao, Quanquan Gu
Abstract: Diffusion models have achieved promising results for Structure-Based Drug Design (SBDD). Nevertheless, high-quality protein subpocket and ligand data are relatively scarce, which hinders the models' generation capabilities. Recently, Direct Preference Optimization (DPO) has emerged as a pivotal tool for aligning generative models with human preferences. In this paper, we propose DecompDPO, a structure-based optimization method aligns diffusion models with pharmaceutical needs using multi-granularity preference pairs. DecompDPO introduces decomposition into the optimization objectives and obtains preference pairs at the molecule or decomposed substructure level based on each objective's decomposability. Additionally, DecompDPO introduces a physics-informed energy term to ensure reasonable molecular conformations in the optimization results. Notably, DecompDPO can be effectively used for two main purposes: (1) fine-tuning pretrained diffusion models for molecule generation across various protein families, and (2) molecular optimization given a specific protein subpocket after generation. Extensive experiments on the CrossDocked2020 benchmark show that DecompDPO significantly improves model performance, achieving up to 95.2% Med. High Affinity and a 36.2% success rate for molecule generation, and 100% Med. High Affinity and a 52.1% success rate for molecular optimization. Code is available at https://github.com/laviaf/DecompDPO.
Authors: Haoyuan Chen, Rui Tuo
Abstract: Gaussian process (GP) models have received increasing attention in recent years due to their superb prediction accuracy and modeling flexibility. To address the computational burdens of GP models for large-scale datasets, distributed learning for GPs are often adopted. Current aggregation models for distributed GPs is not time-efficient when incorporating correlations between GP experts. In this work, we propose a novel approach for aggregated prediction in distributed GPs. The technique is suitable for both the exact and sparse variational GPs. The proposed method incorporates correlations among experts, leading to better prediction accuracy with manageable computational requirements. As demonstrated by empirical studies, the proposed approach results in more stable predictions in less time than state-of-the-art consistent aggregation models.
Authors: Nam V. Nguyen, Thong T. Doan, Luong Tran, Van Nguyen, Quang Pham
Abstract: Mixture of experts (MoE) architectures have become a cornerstone for scaling up and are a key component in most large language models such as GPT-OSS, DeepSeek-V3, Llama-4, and Gemini-2.5. However, systematic research on MoE remains severely constrained by the prohibitive computational costs of training and evaluation, restricting large-scale studies accessible to most researchers. We introduce LibMoE, a unified framework for reproducible, efficient, and extensible MoE research that supports both pretraining and sparse-upcycling regimes. Beyond unified implementations, the framework provides transparent analytical tools for probing routing and expert dynamics. Leveraging this foundation, we conduct a comprehensive analysis along three dimensions: (i) routing dynamics, covering expert selection patterns, routing stability and optimality, and how routing entropy reveals task specialization and expert diversity; (ii) the effect of lightweight initialization on load balancing, demonstrating how subtle changes in router initialization shape early expert utilization; and (iii) training regime differences, revealing how sparse upcycling and full pretraining exhibit distinct routing patterns and stability profiles. By lowering the barrier to entry and standardizing evaluation, along with our comprehensive analysis, LibMoE broadens access to MoE research and establishes a reliable benchmark to guide future innovations. GitHub: \href{https://github.com/Fsoft-AIC/LibMoE}{https://github.com/Fsoft-AIC/LibMoE}.
URLs: https://github.com/Fsoft-AIC/LibMoE, https://github.com/Fsoft-AIC/LibMoE
Authors: Shuntaro Okada, Kenji Doi, Ryota Yoshihashi, Hirokatsu Kataoka, Tomohiro Tanaka
Abstract: We propose a general framework for optimizing noise schedules in diffusion models, applicable to both training and sampling. Our method enforces a constant rate of change in the probability distribution of diffused data throughout the diffusion process, where the rate of change is quantified using a user-defined discrepancy measure. We introduce three such measures, which can be flexibly selected or combined depending on the domain and model architecture. While our framework is inspired by theoretical insights, we do not aim to provide a complete theoretical justification of how distributional change affects sample quality. Instead, we focus on establishing a general-purpose scheduling framework and validating its empirical effectiveness. Through extensive experiments, we demonstrate that our approach consistently improves the performance of both pixel-space and latent-space diffusion models, across various datasets, samplers, and a wide range of number of function evaluations from 5 to 250. In particular, when applied to both training and sampling schedules, our method achieves a state-of-the-art FID score of 2.03 on LSUN Horse 256$\times$256, without compromising mode coverage.
Authors: Andre Kassis, Urs Hengartner, Yaoliang Yu
Abstract: Diffusion-based purification (DBP) has become a cornerstone defense against adversarial examples (AEs), regarded as robust due to its use of diffusion models (DMs) that project AEs onto the natural data manifold. We refute this core claim, theoretically proving that gradient-based attacks effectively target the DM rather than the classifier, causing DBP's outputs to align with adversarial distributions. This prompts a reassessment of DBP's robustness, attributing it to two critical flaws: incorrect gradients and inappropriate evaluation protocols that test only a single random purification of the AE. We show that with proper accounting for stochasticity and resubmission risk, DBP collapses. To support this, we introduce DiffBreak, the first reliable toolkit for differentiation through DBP, eliminating gradient flaws that previously further inflated robustness estimates. We also analyze the current defense scheme used for DBP where classification relies on a single purification, pinpointing its inherent invalidity. We provide a statistically grounded majority-vote (MV) alternative that aggregates predictions across multiple purified copies, showing partial but meaningful robustness gain. We then propose a novel adaptation of an optimization method against deepfake watermarking, crafting systemic perturbations that defeat DBP even under MV, challenging DBP's viability.
Authors: Dingzhi Yu, Wei Jiang, Hongyi Tao, Yuanyu Wan, Lijun Zhang
Abstract: Smoothness is crucial for attaining fast rates in first-order optimization. However, many optimization problems in modern machine learning involve non-smooth objectives. Recent studies relax the smoothness assumption by allowing the Lipschitz constant of the gradient to grow with respect to the gradient norm, which accommodates a broad range of objectives in practice. Despite this progress, existing generalizations of smoothness are restricted to Euclidean geometry with $\ell_2$-norm and only have theoretical guarantees for optimization in the Euclidean space. In this paper, we address this limitation by introducing a new $\ell*$-smoothness concept that measures the norm of Hessians in terms of a general norm and its dual, and establish convergence for mirror-descent-type algorithms, matching the rates under the classic smoothness. Notably, we propose a generalized self-bounding property that facilitates bounding the gradients via controlling suboptimality gaps, serving as a principal component for convergence analysis. Beyond deterministic optimization, we establish sharp convergence for stochastic mirror descent, matching state-of-the-art under classic smoothness. Our theory also extends to non-convex and composite optimization, which may shed light on practical usages of mirror descent, including pre-training and post-training of LLMs.
Authors: Laura Lewis, Dar Gilboa, Jarrod R. McClean
Abstract: Without large quantum computers to empirically evaluate performance, theoretical frameworks such as the quantum statistical query (QSQ) are a primary tool to study quantum algorithms for learning classical functions and search for quantum advantage in machine learning tasks. However, we only understand quantum advantage in this model at two extremes: either exponential advantages for uniform input distributions or no advantage for arbitrary distributions. Our work helps close the gap between these two regimes by designing an efficient quantum algorithm for learning periodic neurons in the QSQ model over a variety of non-uniform distributions and the first explicit treatment of real-valued functions. We prove that this problem is hard not only for classical gradient-based algorithms, which are the workhorses of machine learning, but also for a more general class of SQ algorithms, establishing an exponential quantum advantage.
Authors: Boshi Wang, Huan Sun
Abstract: Despite their impressive capabilities, LLMs exhibit a basic generalization failure known as the Reversal Curse, where they struggle to learn reversible factual associations. Understanding why this occurs could help identify weaknesses in current models and advance their generalization and robustness. In this paper, we conjecture that the Reversal Curse in LLMs is a manifestation of the long-standing binding problem in cognitive science, neuroscience and AI. Specifically, we hypothesize two primary causes of the Reversal Curse stemming from transformers' limitations in conceptual binding: the inconsistency and entanglements of concept representations. We perform a series of experiments that support these conjectures. Our exploration leads to a model design based on JEPA (Joint-Embedding Predictive Architecture) that for the first time breaks the Reversal Curse without side-stepping it with specialized data augmentation or non-causal masking, and moreover, generalization could be further improved by incorporating special memory layers that support disentangled concept representations. Our research opens up the broader fundamental challenge of designing models capable of learning systematic conceptual binding with less human scaffolding.
Authors: Liviu Aolaritei, Bart P. G. Van Parys, Henry Lam, Michael I. Jordan
Abstract: Importance Sampling (IS) is a widely used variance reduction technique for enhancing the efficiency of Monte Carlo methods, particularly in rare-event simulation and related applications. Despite its effectiveness, the performance of IS is highly sensitive to the choice of the proposal distribution and often requires stochastic calibration. While the design and analysis of IS have been extensively studied in estimation settings, applying IS within stochastic optimization introduces a lesser-known fundamental challenge: the decision variable and the importance sampling distribution are mutually dependent, creating a circular optimization structure. This interdependence complicates both convergence analysis and variance control. In this paper, we consider the generic setting of convex stochastic optimization with linear constraints. We propose a single-loop stochastic approximation algorithm, based on a variant of Nesterov's dual averaging, that jointly updates the decision variable and the importance sampling distribution, notably without time-scale separation or nested optimization. The method is globally convergent and achieves the minimal asymptotic variance among stochastic gradient schemes, which moreover matches the performance of an oracle sampler adapted to the optimal solution and thus effectively resolves the circular optimization challenge.
Authors: Satvik Golechha, Adri\`a Garriga-Alonso
Abstract: Prior studies on deception in language-based AI agents typically assess whether the agent produces a false statement about a topic, or makes a binary choice prompted by a goal, rather than allowing open-ended deceptive behavior to emerge in pursuit of a longer-term goal. To fix this, we introduce Among Us, a sandbox social deception game where LLM-agents exhibit long-term, open-ended deception as a consequence of the game objectives. While most benchmarks saturate quickly, Among Us can be expected to last much longer, because it is a multi-player game far from equilibrium. Using the sandbox, we evaluate 18 proprietary and open-weight LLMs and uncover a general trend: models trained with RL are comparatively much better at producing deception than detecting it. We evaluate the effectiveness of methods to detect lying and deception: logistic regression on the activations and sparse autoencoders (SAEs). We find that probes trained on a dataset of "pretend you're a dishonest model:.." generalize extremely well out-of-distribution, consistently obtaining AUROCs over 95% even when evaluated just on the deceptive statement, without the chain of thought. We also find two SAE features that work well at deception detection but are unable to steer the model to lie less. We hope our open-sourced sandbox, game logs, and probes serve to anticipate and mitigate deceptive behavior and capabilities in language-based agents.
Authors: Max Kirchner, Alexander C. Jenke, Sebastian Bodenstedt, Fiona R. Kolbinger, Oliver L. Saldanha, Jakob N. Kather, Martin Wagner, Stefanie Speidel
Abstract: Purpose: Data privacy regulations hinder the creation of generalizable foundation models (FMs) for surgery by preventing multi-institutional data aggregation. This study investigates federated learning (FL) as a privacy-preserving solution to collaboratively train robust surgical FMs. Methods: We introduce Federated EndoViT (FL-EndoViT), a federated framework that validates the Masked Autoencoder (MAE) pretraining strategy in a decentralized surgical setting. To ensure convergence under severe data heterogeneity, the architecture integrates adaptive Sharpness-Aware Minimization (FedSAM). Pretrained on the large-scale Endo700k dataset, FL-EndoViT is evaluated against a centralized baseline on different tasks including scene segmentation, action recognition, and phase recognition. Results: FedSAM is critical for successful pretraining, overcoming the convergence failures of standard federated methods. The resulting FL-EndoViT performs comparably to its centralized counterpart, with significant advantages in data-scarce, high-resolution segmentation and generalization to new surgical events. We also establish that full, end-to-end fine-tuning is necessary for optimal performance. Conclusion: This work validates FL with adaptive optimization as a viable paradigm for creating robust, privacy-preserving surgical FMs. Our findings provide a scalable framework for collaborative Surgical Data Science and underscore the optimizer's critical role in handling data heterogeneity. Future work should explore video-based models to incorporate spatiotemporal dynamics.
Authors: Luca Menicali, Andrew Grace, David H. Richter, Stefano Castruccio
Abstract: Fluid thermodynamics underpins atmospheric dynamics, climate science, industrial applications, and energy systems. However, direct numerical simulations (DNS) of such systems can be computationally prohibitive. To address this, we present a novel physics-informed spatiotemporal surrogate model for Rayleigh-Benard convection (RBC), a canonical example of convective fluid flow. Our approach combines convolutional neural networks, for spatial dimension reduction, with an innovative recurrent architecture, inspired by large language models, to model long-range temporal dynamics. Inference is penalized with respect to the governing partial differential equations to ensure physical interpretability. Since RBC exhibits turbulent behavior, we quantify uncertainty using a conformal prediction framework. This model replicates key physical features of RBC dynamics while significantly reducing computational cost, offering a scalable alternative to DNS for long-term simulations.
Authors: Feiyang Cai, Jiahui Bai, Tao Tang, Guijuan He, Joshua Luo, Tianyu Zhu, Srikanth Pilla, Gang Li, Ling Liu, Feng Luo
Abstract: Precise recognition, editing, and generation of molecules are essential prerequisites for both chemists and AI systems tackling various chemical tasks. We present MolLangBench, a comprehensive benchmark designed to evaluate fundamental molecule-language interface tasks: language-prompted molecular structure recognition, editing, and generation. To ensure high-quality, unambiguous, and deterministic outputs, we construct the recognition tasks using automated cheminformatics tools, and curate editing and generation tasks through rigorous expert annotation and validation. MolLangBench supports the evaluation of models that interface language with different molecular representations, including linear strings, molecular images, and molecular graphs. Evaluations of state-of-the-art models reveal significant limitations: the strongest model (GPT-5) achieves $86.2\%$ and $85.5\%$ accuracy on recognition and editing tasks, which are intuitively simple for humans, and performs even worse on the generation task, reaching only $43.0\%$ accuracy. These results highlight the shortcomings of current AI systems in handling even preliminary molecular recognition and manipulation tasks. We hope MolLangBench will catalyze further research toward more effective and reliable AI systems for chemical applications.The dataset and code can be accessed at https://huggingface.co/datasets/ChemFM/MolLangBench and https://github.com/TheLuoFengLab/MolLangBench, respectively.
URLs: https://huggingface.co/datasets/ChemFM/MolLangBench, https://github.com/TheLuoFengLab/MolLangBench,
Authors: Shuai Wang, Yizhou Sun, Judea Pearl, Ang Li
Abstract: Probabilities of causation play a central role in modern decision making. Tian and Pearl first introduced formal definitions and derived tight bounds for three binary probabilities of causation, such as the probability of necessity and sufficiency (PNS). However, estimating these probabilities requires both experimental and observational distributions specific to each subpopulation, which are often unreliable or impractical to obtain from limited population-level data. To solve this problem, we propose two machine learning models: Exact-MLP and Mask-MLP, which are trained on a small set of reliable subpopulations and are able to predict PNS bounds for all other subpopulations. We validate our models across four Structural Causal Models (SCMs), each evaluated on population-level data with sample sizes between 100k and 200k. Our models achieve average mean absolute errors (MAEs) of roughly 0.03 on main tasks, reducing MAE by about 80% relative to the corresponding baselines. These results demonstrate both the feasibility of machine learning models for learning probabilities of causation and the effectiveness of the proposed approach.
Authors: Thomas Deschatre, Xavier Warin
Abstract: This article presents an input convex neural network architecture using Kolmogorov-Arnold networks (ICKAN). Two specific networks are presented: the first is based on a low-order, linear-by-part, representation of functions, and a universal approximation theorem is provided. The second is based on cubic splines, for which only numerical results support convergence. We demonstrate on simple tests that these networks perform competitively with classical input convex neural networks (ICNNs). In a second part, we use the networks to solve some optimal transport problems needing a convex approximation of functions and demonstrate their effectiveness. Comparisons with ICNNs show that cubic ICKANs produce results similar to those of classical ICNNs.
Authors: George R. Nahass, Zhu Wang, Homa Rashidisabet, Won Hwa Kim, Sasha Hubschman, Jeffrey C. Peterson, Chad A. Purnell, Pete Setabutr, Ann Q. Tran, Darvin Yi, Sathya N. Ravi
Abstract: Machine unlearning aims to remove the influence of specific training samples from a trained model without full retraining. While prior work has largely focused on privacy-motivated settings, we recast unlearning as a general-purpose tool for post-deployment model revision. Specifically, we focus on utilizing unlearning in clinical contexts where data shifts, device deprecation, and policy changes are common. To this end, we propose a bilevel optimization formulation of boundary-based unlearning that can be solved using iterative algorithms. We provide convergence guarantees when first-order algorithms are used to unlearn. Our method introduces tunable loss design for controlling the forgetting-retention tradeoff and supports novel model composition strategies that merge the strengths of distinct unlearning runs. Across benchmark and real-world clinical imaging datasets, our approach outperforms baselines on both forgetting and retention metrics, including scenarios involving imaging devices and anatomical outliers. This work establishes machine unlearning as a modular, practical alternative to retraining for real-world model maintenance in clinical applications.
Authors: Steffen W. R. Werner, Benjamin Peherstorfer
Abstract: Learning stabilizing controllers from data is an important task in engineering applications; however, collecting informative data is challenging because unstable systems often lead to rapidly growing or erratic trajectories. In this work, we propose an adaptive sampling scheme that generates data while simultaneously stabilizing the system to avoid instabilities during the data collection. Under mild assumptions, the approach provably generates data sets that are informative for stabilization and have minimal size. The numerical experiments demonstrate that controller inference with the novel adaptive sampling approach learns controllers with up to one order of magnitude fewer data samples than unguided data generation. The results show that the proposed approach opens the door to stabilizing systems in edge cases and limit states where instabilities often occur and data collection is inherently difficult.
Authors: Xinxu Wei, Kanhao Zhao, Yong Jiao, Lifang He, Yu Zhang
Abstract: As large language models (LLMs) continue to revolutionize AI research, there is a growing interest in building large-scale brain foundation models to advance neuroscience. While most existing brain foundation models are pre-trained on time-series signals or connectome features, we propose a novel graph-based pre-training paradigm for constructing a brain graph foundation model. In this paper, we introduce the Brain Graph Foundation Model, termed BrainGFM, a unified framework that leverages graph contrastive learning and graph masked autoencoders for large-scale fMRI-based pre-training. BrainGFM is pre-trained on a diverse mixture of brain atlases with varying parcellations, significantly expanding the pre-training corpus and enhancing the model's ability to generalize across heterogeneous fMRI-derived brain representations. To support efficient and versatile downstream transfer, we integrate both graph prompts and language prompts into the model design, enabling BrainGFM to flexibly adapt to a wide range of atlases, neurological and psychiatric disorders, and task settings. Furthermore, we employ meta-learning to optimize the graph prompts, facilitating strong generalization to previously unseen disorders under both few-shot and zero-shot learning conditions via language-guided prompting. BrainGFM is pre-trained on 27 neuroimaging datasets spanning 25 common neurological and psychiatric disorders, encompassing 2 types of brain atlases (functional and anatomical) across 8 widely-used parcellations, and covering over 25,000 subjects, 60,000 fMRI scans, and a total of 400,000 graph samples aggregated across all atlases and parcellations.
Authors: Yuchen Zhang, Nikhil Keetha, Chenwei Lyu, Bhuvan Jhamb, Yutian Chen, Yuheng Qiu, Jay Karhade, Shreyas Jha, Yaoyu Hu, Deva Ramanan, Sebastian Scherer, Wenshan Wang
Abstract: Dense image correspondence is central to many applications, such as visual odometry, 3D reconstruction, object association, and re-identification. Historically, dense correspondence has been tackled separately for wide-baseline scenarios and optical flow estimation, despite the common goal of matching content between two images. In this paper, we develop a Unified Flow & Matching model (UFM), which is trained on unified data for pixels that are co-visible in both source and target images. UFM uses a simple, generic transformer architecture that directly regresses the (u,v) flow. It is easier to train and more accurate for large flows compared to the typical coarse-to-fine cost volumes in prior work. UFM is 28% more accurate than state-of-the-art flow methods (Unimatch), while also having 62% less error and 6.7x faster than dense wide-baseline matchers (RoMa). UFM is the first to demonstrate that unified training can outperform specialized approaches across both domains. This result enables fast, general-purpose correspondence and opens new directions for multi-modal, long-range, and real-time correspondence tasks.
Authors: Kasidit Srimahajariyapong, Supanut Thanasilp, Thiparat Chotibut
Abstract: Variational quantum algorithms (VQAs) promise near-term quantum advantage, yet parametrized quantum states commonly built from the digital gate-based approach often suffer from scalability issues such as barren plateaus, where the loss landscape becomes flat. We study an analog VQA ans\"atze composed of $M$ quenches of a disordered Ising chain, whose dynamics is native to several quantum simulation platforms. By tuning the disorder strength we place each quench in either a thermalized phase or a many-body-localized (MBL) phase and analyse (i) the ans\"atze's expressivity and (ii) the scaling of loss variance. Numerics shows that both phases reach maximal expressivity at large $M$, but barren plateaus emerge at far smaller $M$ in the thermalized phase than in the MBL phase. Exploiting this gap, we propose an MBL initialisation strategy: initialise the ans\"atze in the MBL regime at intermediate quench $M$, enabling an initial trainability while retaining sufficient expressivity for subsequent optimization. The results link quantum phases of matter and VQA trainability, and provide practical guidelines for scaling analog-hardware VQAs.
Authors: Xinyue Zeng, Tuo Wang, Adithya Kulkarni, Alexander Lu, Alexandra Ni, Phoebe Xing, Junhan Zhao, Siwei Chen, Dawei Zhou
Abstract: Intrinsically disordered regions (IDRs) play central roles in cellular function, yet remain poorly evaluated by existing protein structure prediction benchmarks. Current evaluations largely focus on well-folded domains, overlooking three fundamental challenges in realistic biological settings: the structural complexity of proteins, the resulting low availability of reliable ground truth, and prediction uncertainty that can propagate into high-risk downstream failures, such as in drug discovery, protein-protein interaction modeling, and functional annotation. We present DisProtBench, an IDR-centric benchmark that explicitly incorporates prediction uncertainty into the evaluation of protein structure prediction models (PSPMs). To address structural complexity and ground-truth scarcity, we curate and unify a large-scale, multi-modal dataset spanning disease-relevant IDRs, GPCR-ligand interactions, and multimeric protein complexes. To assess predictive uncertainty, we introduce Functional Uncertainty Sensitivity (FUS), a novel prediction uncertainty-stratified metric that quantifies downstream task performance under prediction uncertainty. Using this benchmark, we conduct a systematic evaluation of state-of-the-art PSPMs and reveal clear, task-dependent failure modes. Protein-protein interaction prediction degrades sharply in IDRs, while structure-based drug discovery remains comparatively robust. These effects are largely invisible to standard global accuracy metrics, which overestimate functional reliability under prediction uncertainty. We have open-sourced our benchmark and the codebase at https://github.com/Susan571/DisProtBench.
Authors: Aleksandar Armacki, Dragana Bajovic, Dusan Jakovetic, Soummya Kar
Abstract: We study convergence in high-probability of SGD-type methods in non-convex optimization and the presence of heavy-tailed noise. To combat the heavy-tailed noise, a general black-box nonlinear framework is considered, subsuming nonlinearities like sign, clipping, normalization and their smooth counterparts. Our first result shows that nonlinear SGD (N-SGD) achieves the rate $\widetilde{\mathcal{O}}(t^{-1/2})$, for any noise with unbounded moments and a symmetric probability density function (PDF). Crucially, N-SGD has exponentially decaying tails, matching the performance of linear SGD under light-tailed noise. To handle non-symmetric noise, we propose two novel estimators, based on the idea of noise symmetrization. The first, dubbed Symmetrized Gradient Estimator (SGE), assumes a noiseless gradient at any reference point is available at the start of training, while the second, dubbed Mini-batch SGE (MSGE), uses mini-batches to estimate the noiseless gradient. Combined with the nonlinear framework, we get N-SGE and N-MSGE methods, respectively, both achieving the same convergence rate and exponentially decaying tails as N-SGD, while allowing for non-symmetric noise with unbounded moments and PDF satisfying a mild technical condition, with N-MSGE additionally requiring bounded noise moment of order $p \in (1,2]$. Compared to works assuming noise with bounded $p$-th moment, our results: 1) are based on a novel symmetrization approach; 2) provide a unified framework and relaxed moment conditions; 3) imply optimal oracle complexity of N-SGD and N-SGE, strictly better than existing works when $p < 2$, while the complexity of N-MSGE is close to existing works. Compared to works assuming symmetric noise with unbounded moments, we: 1) provide a sharper analysis and improved rates; 2) facilitate state-dependent symmetric noise; 3) extend the strong guarantees to non-symmetric noise.
Authors: Sen Lu, Xiaoyu Zhang, Mingtao Hu, Eric Yeu-Jer Lee, Soohyeon Kim, Wei D. Lu
Abstract: A grand challenge in modern neuroscience is to bridge the gap between the detailed mapping of microscale neural circuits and mechanistic understanding of cognitive functions. While extensive knowledge exists about neuronal connectivity and biophysics, how these low-level phenomena eventually produce abstract behaviors remains largely unresolved. Here, we propose that a model based on State Space Models, an emerging class of deep learning architectures, can be a potential biological model for analysis. We suggest that the differential equations governing elements in a State Space Model are conceptually consistent with the dynamics of biophysical processes, while the model offers a scalable framework to build on the dynamics to produce emergent behaviors observed in experimental neuroscience. We test this model by training a network employing a diagonal state transition matrix on temporal discrimination tasks with reinforcement learning. Our results suggest that neural behaviors such as time cells naturally emerge from two fundamental principles: optimal pre-configuration and rotational dynamics. These features are shown mathematically to optimize history compression, and naturally generate structured temporal dynamics even prior to training, mirroring recent findings in biological circuits. We show that learning acts primarily as a selection mechanism that fine-tunes these pre-configured oscillatory modes, rather than constructing temporal codes de novo. The model can be readily scaled to abstract cognitive functions such as event counting, supporting the use of State Space Models as a computationally tractable framework for understanding neural activities.
Authors: Meixuan Wang, Yinyu Ye, Zijie Zhou
Abstract: We study offline scheduling for large language model (LLM) serving under a fixed KV-cache memory budget, where requests have heterogeneous prompt (prefill) and response (decode) lengths. Prompt tokens determine initial KV usage, and each generated token increases memory by one unit. Given a backlog of n requests arriving together, we schedule mixed prefill and decode batches to minimize total end-to-end latency. We show that heterogeneity in prompt lengths makes the problem computationally intractable and that widely used heuristics such as first-come-first-served and shortest-first can be arbitrarily suboptimal. We propose Sorted-F, which repeatedly forms feasible batches using a new selection metric that balances batch size against downstream decode cost, and prove it achieves a constant-factor guarantee on total latency. We further develop practical variants -- an exact solver for small instances and fast heuristics for larger ones -- and evaluate them on a public workload spanning short conversations and long-document summarization, where they consistently reduce average latency relative to standard baselines. Our results highlight that during peak-hour tidal backlogs, greedy GPU packing or short-request prioritization can perform poorly when prompt lengths vary widely, and provide a principled, tunable framework for designing production batch schedulers and planning capacity in memory-constrained LLM serving systems.
Authors: Ngoc N. Tran, Anwar Said, Waseem Abbas, Tyler Derr, Xenofon D. Koutsoukos
Abstract: While graph-based Android malware classifiers achieve over 94% accuracy on standard benchmarks, they exhibit a significant generalization gap under distribution shift, suffering up to 45% performance degradation when encountering unseen malware variants from known families. This work systematically investigates this critical yet overlooked challenge for real-world deployment by introducing a benchmarking suite designed to simulate two prevalent scenarios: MalNet-Tiny-Common for covariate shift, and MalNet-Tiny-Distinct for domain shift. Furthermore, we identify an inherent limitation in existing benchmarks where the inputs are structure-only function call graphs, which fails to capture the latent semantic patterns necessary for robust generalization. To verify this, we construct a semantic enrichment framework that augments the original topology with function-level attributes, including lightweight metadata and LLM-based code embeddings. By providing this expanded feature set, we aim to equip future research with richer behavioral information to facilitate the development of more sophisticated detection techniques. Empirical evaluations confirm the effectiveness of our data-centric methodology, with which classification performs better under distribution shift compared to model-based approaches, and consistently further enhances robustness when used in conjunction. We release our precomputed datasets, along with an extensible implementation of our comprehensive pipeline, to lay the groundwork for building resilient malware detection systems for evolving threat environments.
Authors: Kexin Chu, Zecheng Lin, Dawei Xiang, Zixu Shen, Jianchang Su, Cheng Chu, Yiwei Yang, Wenhui Zhang, Wenfei Wu, Wei Zhang
Abstract: Global KV-cache sharing is an effective optimization for accelerating large language model (LLM) inference, yet it introduces an API-visible timing side channel that lets adversaries infer sensitive user inputs from shared entries, leading to cross-tenant privacy risks. To address this problem, we introduce SafeKV (Secure and Flexible KV-cache Sharing), a system-level co-design of privacy enforcement and KV-cache management. SafeKV integrates lightweight detection and isolation directly into the serving runtime to eliminate cross-tenant reuse of sensitive KV-cache blocks under our threat model, while recovering most of the performance benefits of global sharing. Our key contributions are: (1) a three-tier asynchronous detection pipeline that decouples privacy classification from inference and supports streaming workloads, (2) a unified radix-tree-based memory manager with path compression and sensitivity-aware eviction for scalable selective isolation, and (3) an RDR-guided (Reuse Diversity Ratio) runtime safeguard that detects and bounds residual leakage. On large LLM backends, SafeKV reduces the time-to-first-token (TTFT) overhead compared to full isolation by up to 40.58% and raises throughput by up to 2.66x. Overall, SafeKV restores the efficiency of KV reuse while enforcing strong, practical privacy for multi-tenant LLM inference.
Authors: Akshay Mhatre, Noujoud Nader, Patrick Diehl, Deepti Gupta
Abstract: Large Language Models (LLMs) such as ChatGPT-4, Claude 3, and LLaMA 4 are increasingly embedded in software/application development, supporting tasks from code generation to debugging. Yet, their real-world effectiveness in detecting diverse software bugs, particularly complex, security-relevant vulnerabilities, remains underexplored. This study presents a systematic, empirical evaluation of these three leading LLMs using a benchmark of foundational programming errors, classic security flaws, and advanced, production-grade bugs in C++ and Python. The dataset integrates real code from SEED Labs, OpenSSL (via the Suresoft GLaDOS database), and PyBugHive, validated through local compilation and testing pipelines. A novel multi-stage, context-aware prompting protocol simulates realistic debugging scenarios, while a graded rubric measures detection accuracy, reasoning depth, and remediation quality. Our results show that all models excel at identifying syntactic and semantic issues in well-scoped code, making them promising for educational use and as first-pass reviewers in automated code auditing. Performance diminishes in scenarios involving complex security vulnerabilities and large-scale production code, with ChatGPT-4 and Claude 3 generally providing more nuanced contextual analyses than LLaMA 4. This highlights both the promise and the present constraints of LLMs in serving as reliable code analysis tools.
Authors: Payal Varshney, Adriano Lucieri, Christoph Balada, Sheraz Ahmed, Andreas Dengel
Abstract: Video-based AI systems are increasingly adopted in safety-critical domains such as autonomous driving and healthcare. However, interpreting their decisions remains challenging due to the inherent spatiotemporal complexity of video data and the opacity of deep learning models. Existing explanation techniques often suffer from limited temporal coherence and a lack of actionable causal insights. Current counterfactual explanation methods typically do not incorporate guidance from the target model, reducing semantic fidelity and practical utility. We introduce Latent Diffusion for Video Counterfactual Explanations (LD-ViCE), a novel framework designed to explain the behavior of video-based AI models. Compared to previous approaches, LD-ViCE reduces the computational costs of generating explanations by operating in latent space using a state-of-the-art diffusion model, while producing realistic and interpretable counterfactuals through an additional refinement step. Experiments on three diverse video datasets - EchoNet-Dynamic (cardiac ultrasound), FERV39k (facial expression), and Something-Something V2 (action recognition) with multiple target models covering both classification and regression tasks, demonstrate that LD-ViCE generalizes well and achieves state-of-the-art performance. On the EchoNet-Dynamic dataset, LD-ViCE achieves significantly higher regression accuracy than prior methods and exhibits high temporal consistency, while the refinement stage further improves perceptual quality. Qualitative analyses confirm that LD-ViCE produces semantically meaningful and temporally coherent explanations, providing actionable insights into model behavior. LD-ViCE advances the trustworthiness and interpretability of video-based AI systems through visually coherent counterfactual explanations.
Authors: Savini Kashmira, Jayanaka Dantanarayana, Thamirawaran Sathiyalogeswaran, Yichao Yuan, Nishil Talati, Krisztian Flautner, Lingjia Tang, Jason Mars
Abstract: This paper presents GRAPHMEND, a high-level compiler technique that eliminates FX graph breaks in PyTorch 2 programs. Although PyTorch 2 introduced TorchDynamo and TorchInductor to enable just-in-time graph compilation, unresolved dynamic control flow and unsupported Python constructs often fragment models into multiple FX graphs. These fragments force frequent fallbacks to eager mode, introduce costly CPU-to-GPU synchronizations, and reduce optimization opportunities. GRAPHMEND addresses this limitation by analyzing and transforming source code before execution. Built on the Jaseci compilation framework, GRAPHMEND introduces two code transformations that remove graph breaks due to dynamic control flow and Python side effects. This design allows PyTorch's compilation pipeline to capture larger, uninterrupted FX graphs without requiring manual refactoring by developers. Evaluation across eight Hugging Face models shows that GRAPHMEND removes graph breaks due to dynamic control flow and Python side effects, reducing the break count to 0 in 6 models and reducing it from 5 to 2 in another model. On NVIDIA RTX 3090 and A40 GPUs, GRAPHMEND achieves up to 75% latency reductions and up to 8% higher end-to-end throughput. These results demonstrate that high-level code transformation is an effective complement to PyTorch's dynamic JIT compilation pipeline, substantially improving both usability and performance.
Authors: Trinnhallen Brisley, Gordon Ross, Daniel Paulin
Abstract: Hawkes process models are used in settings where past events increase the likelihood of future events occurring. Many applications record events as counts on a regular grid, yet discrete-time Hawkes models remain comparatively underused and are often constrained by fixed-form baselines and excitation kernels. In particular, there is a lack of flexible, nonparametric treatments of both the baseline and the excitation in discrete time. To this end, we propose the Gaussian Process Discrete Hawkes Process (GP-DHP), a nonparametric framework that places Gaussian process priors on both the baseline and the excitation and performs inference through a collapsed latent representation. This yields smooth, data-adaptive structure without prespecifying trends, periodicities, or decay shapes, and enables maximum a posteriori (MAP) estimation with near-linear-time \(O(T\log T)\) complexity. A closed-form projection recovers interpretable baseline and excitation functions from the optimized latent trajectory. In simulations, GP-DHP recovers diverse excitation shapes and evolving baselines. In case studies on U.S. terrorism incidents and weekly Cryptosporidiosis counts, it improves test predictive log-likelihood over standard parametric discrete Hawkes baselines while capturing bursts, delays, and seasonal background variation. The results indicate that flexible discrete-time self-excitation can be achieved without sacrificing scalability or interpretability.
Authors: Carly Sutter, Kara J. Sulia, Nick P. Bassill, Christopher D. Wirz, Christopher D. Thorncroft, Jay C. Rothenberger, Vanessa Przybylo, Mariana G. Cains, Jacob Radford, David Aaron Evans
Abstract: Transportation agencies make critical operational decisions during hazardous weather events, including assessment of road conditions and resource allocation. In this study, machine learning models are developed to provide additional support for the New York State Department of Transportation (NYSDOT) by automatically classifying current road conditions across the state. Convolutional neural networks and random forests are trained on NYSDOT roadside camera images and weather data to predict road surface conditions. This task draws critically on a robust hand-labeled dataset of ~22,000 camera images containing six road surface conditions: severe snow, snow, wet, dry, poor visibility, or obstructed. Model generalizability is prioritized to meet the operational needs of the NYSDOT decision makers, including integration of operational datasets and use of representative and realistic images. The weather-related road surface condition model in this study achieves an accuracy of 81.5% on completely unseen cameras. With operational deployment, this model has the potential to improve spatial and temporal awareness of road surface conditions, which can strengthen decision-making for operations, roadway maintenance, and traveler safety, particularly during winter weather events.
Authors: Haomiao Chen, Keith W Jamison, Mert R. Sabuncu, Amy Kuceyeski
Abstract: Neural encoding models aim to predict fMRI-measured brain responses to natural images. fMRI data is acquired as a 3D volume of voxels, where each voxel has a defined spatial location in the brain. However, conventional encoding models often flatten this volume into a 1D vector and treat voxel responses as independent outputs. This removes spatial context, discards anatomical information, and ties each model to a subject-specific voxel grid. We introduce the Neural Response Function (NRF), a framework that models fMRI activity as a continuous function over anatomical space rather than a flat vector of voxels. NRF represents brain activity as a continuous implicit function: given an image and a spatial coordinate (x, y, z) in standardized MNI space, the model predicts the response at that location. This formulation decouples predictions from the training grid, supports querying at arbitrary spatial resolutions, and enables resolution-agnostic analyses. By grounding the model in anatomical space, NRF exploits two key properties of brain responses: (1) local smoothness -- neighboring voxels exhibit similar response patterns; modeling responses continuously captures these correlations and improves data efficiency, and (2) cross-subject alignment -- MNI coordinates unify data across individuals, allowing a model pretrained on one subject to be fine-tuned on new subjects. In experiments, NRF outperformed baseline models in both intrasubject encoding and cross-subject adaptation, achieving high performance while reducing the data size needed by orders of magnitude. To our knowledge, NRF is the first anatomically aware encoding model to move beyond flattened voxels, learning a continuous mapping from images to brain responses in 3D space.
Authors: Jannis Chemseddine, Gregor Kornhardt, Richard Duong, Gabriele Steidl
Abstract: The default Gaussian latent in flow-based generative models poses challenges when learning certain distributions such as heavy-tailed ones. We introduce a general framework for learning data-adaptive latent distributions using one-dimensional quantile functions, optimized via the Wasserstein distance between noise and data. The quantile-based parameterization naturally adapts to both heavy-tailed and compactly supported distributions and shortens transport paths. Numerical results confirm the method's flexibility and effectiveness achieved with negligible computational overhead.
Authors: Alistair Brash, Junyi Lu, Bruce Stephen, Blair Brown, Robert Atkinson, Craig Michie, Fraser MacIntyre, Christos Tachtatzis
Abstract: Limited visibility of distribution network power flows at the low voltage level presents challenges to both distribution network operators from a planning perspective and distribution system operators from a congestion management perspective. More representative loads are required to support meaningful analysis of LV substations; otherwise, such analysis risks misinforming future decisions. Traditional load profiling relies on typical profiles, oversimplifying substation-level complexity. Generative models have attempted to address this through synthesising representative loads from historical exemplars; however, while these approaches can approximate load shapes to a convincing degree of fidelity, analysis of the co-behaviour between substations is limited, which ultimately impacts higher voltage level network operation. This limitation will become even more pronounced with the increasing integration of low-carbon technologies, as estimates of base loads fail to capture load diversity. To address this gap, Conditional Diffusion models for synthesising daily active and reactive power profiles at the low voltage distribution substation level are proposed. The evaluation of fidelity is demonstrated through conventional metrics capturing temporal and statistical realism, as well as power flow modelling. Multiple models are proposed to handle varying levels of data availability, ranging from unconditional synthesis to an informed generation driven by metadata and daily statistics. The results show synthesised load profiles are plausible both independently and as a cohort in a wider power systems context. The Conditional Diffusion model is benchmarked against both naive and state-of-the-art models to demonstrate its effectiveness in producing realistic scenarios on which to base sub-regional power distribution network planning and operations.
Authors: Zhengshen Zhang, Hao Li, Yalun Dai, Zhengbang Zhu, Lei Zhou, Chenchen Liu, Dong Wang, Francis E. H. Tay, Sijin Chen, Ziwei Liu, Yuxiao Liu, Xinghang Li, Pan Zhou
Abstract: Existing vision-language-action (VLA) models act in 3D real-world but are typically built on 2D encoders, leaving a spatial reasoning gap that limits generalization and adaptability. Recent 3D integration techniques for VLAs either require specialized sensors and transfer poorly across modalities, or inject weak cues that lack geometry and degrade vision-language alignment. In this work, we introduce FALCON (From Spatial to Action), a novel paradigm that injects rich 3D spatial tokens into the action head. FALCON leverages spatial foundation models to deliver strong geometric priors from RGB alone, and includes an Embodied Spatial Model that can optionally fuse depth, or pose for higher fidelity when available, without retraining or architectural changes. To preserve language reasoning, spatial tokens are consumed by a Spatial-Enhanced Action Head rather than being concatenated into the vision-language backbone. These designs enable FALCON to address limitations in spatial representation, modality transferability, and alignment. In comprehensive evaluations across three simulation benchmarks and eleven real-world tasks, our proposed FALCON achieves state-of-the-art performance, consistently surpasses competitive baselines, and remains robust under clutter, spatial-prompt conditioning, and variations in object scale and height.
Authors: Brennen A. Hill
Abstract: Traditional neural networks, while powerful, rely on biologically implausible learning mechanisms such as global backpropagation. This paper introduces the Structurally Adaptive Predictive Inference Network (SAPIN), a novel computational model inspired by the principles of active inference and the morphological plasticity observed in biological neural cultures. SAPIN operates on a 2D grid where processing units, or cells, learn by minimizing local prediction errors. The model features two primary, concurrent learning mechanisms: a local, Hebbian-like synaptic plasticity rule based on the temporal difference between a cell's actual activation and its learned expectation, and a structural plasticity mechanism where cells physically migrate across the grid to optimize their information-receptive fields. This dual approach allows the network to learn both how to process information (synaptic weights) and also where to position its computational resources (network topology). We validated the SAPIN model on the classic Cart Pole reinforcement learning benchmark. Our results demonstrate that the architecture can successfully solve the CartPole task, achieving robust performance. The network's intrinsic drive to minimize prediction error and maintain homeostasis was sufficient to discover a stable balancing policy. We also found that while continual learning led to instability, locking the network's parameters after achieving success resulted in a stable policy. When evaluated for 100 episodes post-locking (repeated over 100 successful agents), the locked networks maintained an average 82% success rate.
Authors: Shashanka B R, Mohith Charan R, Seema Banu F
Abstract: Chunking has emerged as a critical technique that enhances generative models by grounding their responses in efficiently segmented knowledge [1]. While initially developed for unimodal (primarily textual) domains, recent advances in multimodal foundation models have extended chunking approaches to incorporate diverse data types, including images, audio, and video [2]. A critical component underpinning the success of these systems is the chunking strategy how large, continuous streams of multimodal data are segmented into semantically meaningful units suitable for processing [3]. Despite its importance, chunking remains an under-explored area, especially in the context of multimodal systems where modality-specific constraints, semantic preservation, and alignment across modalities introduce unique challenges. Our goal is to consolidating the landscape of multimodal chunking strategies, providing researchers and practitioners with a technical foundation and design space for developing more effective and efficient multimodal AI systems. This survey paves the way for innovations in robust chunking pipelines that scale with modality complexity, enhance processing accuracy, and improve generative coherence in real-world applications. This survey provides a comprehensive taxonomy and technical analysis of chunking strategies tailored for each modality: text, images, audio, video, and cross-modal data. We examine classical and modern approaches such as fixed-size token windowing, recursive text splitting, object-centric visual chunking, silence-based audio segmentation, and scene detection in videos. Each approach is analyzed in terms of its underlying methodology, supporting tools (e.g., LangChain, Detectron2, PySceneDetect), benefits, and challenges, particularly those related to granularity-context trade-offs and multimodal alignment. Furthermore, we explore emerging cross-modal chunking strategies that aim to preserve alignment and semantic consistency across disparate data types [4]. We also include comparative insights, highlight open problems such as asynchronous information density and noisy alignment signals, and identify opportunities for future research in adaptive, learning-based, and task-specific chunking.
Authors: Xabier de Zuazo, Ibon Saratxaga, Eva Navas
Abstract: Decoding speech-related information from non-invasive MEG is a key step toward scalable brain-computer interfaces. We present compact Conformer-based decoders on the LibriBrain 2025 PNPL benchmark for two core tasks: Speech Detection and Phoneme Classification. Our approach adapts a compact Conformer to raw 306-channel MEG signals, with a lightweight convolutional projection layer and task-specific heads. For Speech Detection, a MEG-oriented SpecAugment provided a first exploration of MEG-specific augmentation. For Phoneme Classification, we used inverse-square-root class weighting and a dynamic grouping loader to handle 100-sample averaged examples. In addition, a simple instance-level normalization proved critical to mitigate distribution shifts on the holdout split. Using the official Standard track splits and F1-macro for model selection, our best systems achieved 88.9% (Speech) and 65.8% (Phoneme) on the leaderboard, winning the Phoneme Classification Standard track. For further implementation details, the technical documentation, source code, and checkpoints are available at https://github.com/neural2speech/libribrain-experiments.
URLs: https://github.com/neural2speech/libribrain-experiments.
Authors: Kaiyuan Xu
Abstract: In recent years, the proliferation of misinformation and fake news has posed serious threats to individuals and society, spurring intense research into automated detection methods. Previous work showed that integrating content, user preferences, and propagation structure achieves strong performance, but leaves all graph-level representation learning entirely to the GNN, hiding any explicit topological cues. To close this gap, we introduce a lightweight enhancement: for each node, we append two classical graph-theoretic metrics, degree centrality and local clustering coefficient, to its original BERT and profile embeddings, thus explicitly flagging the roles of hub and community. In the UPFD Politifact subset, this simple modification boosts macro F1 from 0.7753 to 0.8344 over the original baseline. Our study not only demonstrates the practical value of explicit topology features in fake-news detection but also provides an interpretable, easily reproducible template for fusing graph metrics in other information-diffusion tasks.
Authors: Liviu Aolaritei, Michael I. Jordan
Abstract: Deciding when to stop stochastic gradient descent (SGD) has long remained unresolved in a statistically rigorous sense. While SGD is routinely monitored as it runs, the classical theory of SGD provides guarantees only at pre-specified iteration horizons and offers no valid way to decide, based on the observed trajectory, when further computation is justified. We address this gap by developing anytime-valid confidence sequences for stochastic gradient methods, which remain valid under continuous monitoring and directly induce statistically valid, trajectory-dependent stopping rules: stop as soon as the current upper confidence bound on an appropriate performance measure falls below a user-specified tolerance. The confidence sequences are constructed using nonnegative supermartingales, are time-uniform, and depend only on observable quantities along the SGD trajectory, without requiring prior knowledge of the optimization horizon. In convex optimization, this yields anytime-valid certificates for weighted suboptimality of projected SGD under general stepsize schedules, without assuming smoothness or strong convexity. In nonconvex optimization, it yields time-uniform certificates for weighted first-order stationarity under smoothness assumptions. We further characterize the stopping-time complexity of the resulting stopping rules under standard stepsize schedules. To the best of our knowledge, this is the first framework that provides statistically valid, time-uniform stopping rules for SGD across both convex and nonconvex settings based solely on its observed trajectory.
Authors: Sirui Chen, Zi-ang Cao, Zhengyi Luo, Fernando Casta\~neda, Chenran Li, Tingwu Wang, Ye Yuan, Linxi "Jim" Fan, C. Karen Liu, Yuke Zhu
Abstract: Recent progress in humanoid robots has unlocked agile locomotion skills, including backflipping, running, and crawling. Yet it remains challenging for a humanoid robot to perform forceful manipulation tasks such as moving objects, wiping, and pushing a cart. We propose adaptive Compliance Humanoid control through hIsight Perturbation (CHIP), a plug-and-play module that enables controllable end-effector stiffness while preserving agile tracking of dynamic reference motions. CHIP is easy to implement and requires neither data augmentation nor additional reward tuning. We show that a generalist motion-tracking controller trained with CHIP can perform a diverse set of forceful manipulation tasks that require different end-effector compliance, such as multi-robot collaboration, wiping, box delivery, and door opening.
Authors: Mozes Jacobs, Thomas Fel, Richard Hakim, Alessandra Brondetta, Demba Ba, T. Andy Keller
Abstract: As Vision Transformers (ViTs) become standard vision backbones, a mechanistic account of their computational phenomenology is essential. Despite architectural cues that hint at dynamical structure, there is no settled framework that interprets Transformer depth as a well-characterized flow. In this work, we introduce the Block-Recurrent Hypothesis (BRH), arguing that trained ViTs admit a block-recurrent depth structure such that the computation of the original $L$ blocks can be accurately rewritten using only $k \ll L$ distinct blocks applied recurrently. Across diverse ViTs, between-layer representational similarity matrices suggest few contiguous phases. To determine whether these phases reflect genuinely reusable computation, we train block-recurrent surrogates of pretrained ViTs: Recurrent Approximations to Phase-structured TransfORmers (Raptor). In small-scale, we demonstrate that stochastic depth and training promote recurrent structure and subsequently correlate with our ability to accurately fit Raptor. We then provide an empirical existence proof for BRH by training a Raptor model to recover $96\%$ of DINOv2 ImageNet-1k linear probe accuracy in only 2 blocks at equivalent runtime. Finally, we leverage our hypothesis to develop a program of Dynamical Interpretability. We find i) directional convergence into class-dependent angular basins with self-correcting trajectories under small perturbations, ii) token-specific dynamics, where cls executes sharp late reorientations while patch tokens exhibit strong late-stage coherence toward their mean direction, and iii) a collapse to low rank updates in late depth, consistent with convergence to low-dimensional attractors. Altogether, we find a compact recurrent program emerges along ViT depth, pointing to a low-complexity normative solution that enables these models to be studied through principled dynamical systems analysis.
Authors: Kaiwen Tang, Jiaqi Zheng, Yuze Jin, Yupeng Qiu, Guangda Sun, Zhanglu Yan, Weng-Fai Wong
Abstract: Time-series forecasting in domains like traffic management and industrial monitoring often requires real-time, energy-efficient processing on edge devices with limited resources. Spiking neural networks (SNNs) offer event-driven computation and ultra-low power and have been proposed for use in this space. Unfortunately, existing SNN-based time-series forecasters often use complex transformer blocks. To address this issue, we propose SpikySpace, a spiking state-space model (SSM) that reduces the quadratic cost in the attention block to linear time via spiking selective scanning. Further, we introduce PTsoftplus and PTSiLU, two efficient approximations of SiLU and Softplus that replace costly exponential and division operations with simple bit-shifts. Evaluated on four multivariate time-series benchmarks, SpikySpace outperforms the leading SNN in terms of accuracy by up to 3.0% while reducing energy consumption by over 96.1%. As the first fully spiking state-space model, SpikySpace bridges neuromorphic efficiency with modern sequence modeling, opening a practical path toward efficient time series forecasting systems. Our code is available at https://anonymous.4open.science/r/SpikySpace.
Authors: Benjamin Rozonoyer, Chong You, Michael Boratko, Himanshu Jain, Nilesh Gupta, Srinadh Bhojanapalli, Andrew McCallum, Felix Yu
Abstract: The success of Large Language Models (LLMs) has motivated a shift toward generative approaches to retrieval and ranking, aiming to supersede classical Dual Encoders (DEs) and Cross Encoders (CEs). A prominent paradigm is pointwise Autoregressive Ranking (ARR), where an LLM generates document identifiers (docIDs) token-by-token to enable ranking via beam search. ARR offers the promise of superior expressivity compared to DEs while avoiding the prohibitive computational cost of CEs. However, a formal theoretical foundation for this expressive power has been missing. Moreover, the standard next-token prediction loss is rank-agnostic and inappropriate for finetuning an LLM for ranking tasks. In this paper, we first prove that the expressive capacity of ARR is strictly superior to DEs. While a DE requires an embedding dimension that grows linearly with corpus size to achieve arbitrary rankings, ARR can solve it with a constant hidden dimension. We then propose SToICaL (Simple Token-Item Calibrated Loss), a generalized rank-aware training loss for LLM finetuning. By using item-level reweighting and prefix-tree marginalization, we distribute probability mass over valid docID tokens based on their ground-truth relevance. Experiments on WordNet and ESCI datasets verify that our loss suppresses invalid docID generations and significantly improves ranking metrics beyond top-1 retrieval.
Authors: Masahiro Kato
Abstract: Estimating the Riesz representer is central to debiased machine learning for causal and structural parameter estimation. We propose generalized Riesz regression, a unified framework for estimating the Riesz representer by fitting a representer model via Bregman divergence minimization. This framework includes various divergences as special cases, such as the squared distance and the Kullback--Leibler (KL) divergence, where the former recovers Riesz regression and the latter recovers tailored loss minimization. Under suitable pairs of divergence and model specifications (link functions), the dual problems of the Riesz representer fitting problem correspond to covariate balancing, which we call automatic covariate balancing. Moreover, under the same specifications, the sample average of outcomes weighted by the estimated Riesz representer satisfies Neyman orthogonality even without estimating the regression function, a property we call automatic Neyman orthogonalization. This property not only reduces the estimation error of Neyman orthogonal scores but also clarifies a key distinction between debiased machine learning and targeted maximum likelihood estimation (TMLE). Our framework can also be viewed as a generalization of density ratio fitting under Bregman divergences to Riesz representer estimation, and it applies beyond density ratio estimation. We provide convergence analyses for both reproducing kernel Hilbert space (RKHS) and neural network model classes. A Python package for generalized Riesz regression is released as genriesz and is available at https://github.com/MasaKat0/genriesz.
Authors: Zeyu Bian, Lan Wang, Chengchun Shi, Zhengling Qi
Abstract: Fairness is a central pillar of trustworthy machine learning, especially in domains where accuracy- or profit-driven optimization is insufficient. While most fairness research focuses on supervised learning, fairness in policy learning remains less explored. Because policy learning is interventional, it induces two distinct fairness targets: action fairness (equitable action assignments) and outcome fairness (equitable downstream consequences). Crucially, equalizing actions does not generally equalize outcomes when groups face different constraints or respond differently to the same action. We propose a novel double fairness learning (DFL) framework that explicitly manages the trade-off among three objectives: action fairness, outcome fairness, and value maximization. We integrate fairness directly into a multi-objective optimization problem for policy learning and employ a lexicographic weighted Tchebyshev method that recovers Pareto solutions beyond convex settings, with theoretical guarantees on the regret bounds. Our framework is flexible and accommodates various commonly used fairness notions. Extensive simulations demonstrate improved performance relative to competing methods. In applications to a motor third-party liability insurance dataset and an entrepreneurship training dataset, DFL substantially improves both action and outcome fairness while incurring only a modest reduction in overall value.
Authors: Xiaocheng Fang, Zhengyao Ding, Guangkun Nie, Jieyi Cai, Yujie Xiao, Bo Liu, Jiarui Jin, Haoyu Wang, Shun Huang, Ting Chen, Hongyan Li, Shenda Hong
Abstract: Cardiac Magnetic Resonance (CMR) imaging provides a comprehensive assessment of cardiac structure and function but remains constrained by high acquisition costs and reliance on expert annotations, limiting the availability of large-scale labeled datasets. In contrast, electrocardiograms (ECGs) are inexpensive, widely accessible, and offer a promising modality for conditioning the generative synthesis of cine CMR. To this end, we propose ECGFlowCMR, a novel ECG-to-CMR generative framework that integrates a Phase-Aware Masked Autoencoder (PA-MAE) and an Anatomy-Motion Disentangled Flow (AMDF) to address two fundamental challenges: (1) the cross-modal temporal mismatch between multi-beat ECG recordings and single-cycle CMR sequences, and (2) the anatomical observability gap due to the limited structural information inherent in ECGs. Extensive experiments on the UK Biobank and a proprietary clinical dataset demonstrate that ECGFlowCMR can generate realistic cine CMR sequences from ECG inputs, enabling scalable pretraining and improving performance on downstream cardiac disease classification and phenotype prediction tasks.
Authors: Enrico Ventura, Beatrice Achilli, Luca Ambrogioni, Carlo Lucibello
Abstract: Classifier-free guidance (CFG) is the de facto standard for conditional sampling in diffusion models, yet it often leads to a loss of diversity in generated samples. We formalize this phenomenon as generative distortion, defined as the mismatch between the CFG-induced sampling distribution and the true conditional distribution. Considering Gaussian mixtures and their exact scores, and leveraging tools from statistical physics, we characterize the onset of distortion in a high-dimensional regime as a function of the number of classes. Our analysis reveals that distortions emerge through a phase transition in the effective potential governing the guided dynamics. In particular, our dynamical mean-field analysis shows that distortion persists when the number of modes grows exponentially with dimension, but vanishes in the sub-exponential regime. Consistent with prior finite-dimensional results, we further demonstrate that vanilla CFG shifts the mean and shrinks the variance of the conditional distribution. We show that standard CFG schedules are fundamentally incapable of preventing variance shrinkage. Finally, we propose a theoretically motivated guidance schedule featuring a negative-guidance window, which mitigates loss of diversity while preserving class separability.
Authors: Min Cai, Yu Liang, Longzheng Wang, Yan Wang, Yueyang Zhang, Long Xia, Zhiyuan Sun, Xi Ye, Daiting Shi
Abstract: Reinforcement learning (RL) has played a central role in recent advances in large reasoning models (LRMs), yielding strong gains in verifiable and open-ended reasoning. However, training a single general-purpose LRM across diverse domains remains challenging due to pronounced domain heterogeneity. Through a systematic study of two widely used strategies, Sequential RL and Mixed RL, we find that both incur substantial cross-domain interference at the behavioral and gradient levels, resulting in limited overall gains. To address these challenges, we introduce **M**odular **G**radient **S**urgery (**MGS**), which resolves gradient conflicts at the module level within the transformer. When applied to Llama and Qwen models, MGS achieves average improvements of 4.3 (16.6\%) and 4.5 (11.1\%) points, respectively, over standard multi-task RL across three representative domains (math, general chat, and instruction following). Further analysis demonstrates that MGS remains effective under prolonged training. Overall, our study clarifies the sources of interference in multi-domain RL and presents an effective solution for training general-purpose LRMs.
Authors: Mario Franco, Carlos Gershenson
Abstract: Nowadays, neural networks act as a synonym for artificial intelligence. Present neural network models, although remarkably powerful, are inefficient both in terms of data and energy. Several alternative forms of neural networks have been proposed to address some of these problems. Specifically, spiking neural networks are suitable for efficient hardware implementations. However, effective learning algorithms for spiking networks remain elusive, although it is suspected that effective plasticity mechanisms could alleviate the problem of data efficiency. Here, we present a new framework for spiking neural networks - Spark - built upon the idea of modular design, from simple components to entire models. The aim of this framework is to provide an efficient and streamlined pipeline for spiking neural networks. We showcase this framework by solving the sparse-reward cartpole problem with simple plasticity mechanisms. We hope that a framework compatible with traditional ML pipelines may accelerate research in the area, specifically for continuous and unbatched learning, akin to the one animals exhibit.
Authors: Peter L. Chen, Xiaopeng Li, Xi Chen, Tianyi Lin
Abstract: Direct alignment methods are increasingly used to align large language models (LLMs) with human preferences. However, many real-world alignment problems involve multiple conflicting objectives, where naive aggregation of preferences can lead to unstable training and poor trade-offs. In particular, weighted loss methods may fail to identify update directions that simultaneously improve all objectives, and existing multi-objective approaches often rely on explicit reward models, introducing additional complexity and distorting user-specified preferences. The contributions of this paper are two-fold. First, we propose a Reward-free Alignment framework for Conflicted Objectives (RACO) that directly leverages pairwise preference data and resolves gradient conflicts via a novel clipped variant of conflict-averse gradient descent. We provide convergence guarantees to Pareto-critical points that respect user-specified objective weights, and further show that clipping can strictly improve convergence rate in the two-objective setting. Second, we improve our method using some heuristics and conduct experiments to demonstrate the compatibility of the proposed framework for LLM alignment. Both qualitative and quantitative evaluations on multi-objective summarization and safety alignment tasks across multiple LLM families (Qwen 3, Llama 3, Gemma 3) show that our method consistently achieves better Pareto trade-offs compared to existing multi-objective alignment baselines.
Authors: Qiuming Luo, Tao Zeng, Feng Li, Heming Liu, Rui Mao, Chang Kong
Abstract: Zero-shot Handwritten Chinese Character Recognition (HCCR) aims to recognize unseen characters by leveraging radical-based semantic compositions. However, existing approaches often treat characters as flat radical sequences, neglecting the hierarchical topology and the uneven information density of different components. To address these limitations, we propose an Entropy-Aware Structural Alignment Network that bridges the visual-semantic gap through information-theoretic modeling. First, we introduce an Information Entropy Prior to dynamically modulate positional embeddings via multiplicative interaction, acting as a saliency detector that prioritizes discriminative roots over ubiquitous components. Second, we construct a Dual-View Radical Tree to extract multi-granularity structural features, which are integrated via an adaptive Sigmoid-based gating network to encode both global layout and local spatial roles. Finally, a Top-K Semantic Feature Fusion mechanism is devised to augment the decoding process by utilizing the centroid of semantic neighbors, effectively rectifying visual ambiguities through feature-level consensus. Extensive experiments demonstrate that our method establishes new state-of-the-art performance, achieving an accuracy of 55.04\% on the ICDAR 2013 dataset ($m=1500$), significantly outperforming existing CLIP-based baselines in the challenging zero-shot setting. Furthermore, the framework exhibits exceptional data efficiency, demonstrating rapid adaptability with minimal support samples, achieving 92.41\% accuracy with only one support sample per class.
Authors: Anastasis Kratsios, Giulia Livieri, A. Martina Neuman
Abstract: We study the statistical behaviour of reasoning probes in a stylized model of looped reasoning, given by Boolean circuits whose computational graph is a perfect $\nu$-ary tree ($\nu\ge 2$) and whose output is appended to the input and fed back iteratively for subsequent computation rounds. A reasoning probe has access to a sampled subset of internal computation nodes, possibly without covering the entire graph, and seeks to infer which $\nu$-ary Boolean gate is executed at each queried node, representing uncertainty via a probability distribution over a fixed collection of $\mathtt{m}$ admissible $\nu$-ary gates. This partial observability induces a generalization problem, which we analyze in a realizable, transductive setting. We show that, when the reasoning probe is parameterized by a graph convolutional network (GCN)-based hypothesis class and queries $N$ nodes, the worst-case generalization error attains the optimal rate $\mathcal{O}(\sqrt{\log(2/\delta)}/\sqrt{N})$ with probability at least $1-\delta$, for $\delta\in (0,1)$. Our analysis combines snowflake metric embedding techniques with tools from statistical optimal transport. A key insight is that this optimal rate is achievable independently of graph size, owing to the existence of a low-distortion one-dimensional snowflake embedding of the induced graph metric. As a consequence, our results provide a sharp characterization of how structural properties of the computational graph govern the statistical efficiency of reasoning under partial access.
Authors: Gr\'egoire Dhimo\"ila, Thomas Fel, Victor Boutin, Agustin Picard
Abstract: Vision-language models (VLMs) align images and text with remarkable success, yet the geometry of their shared embedding space remains poorly understood. To probe this geometry, we begin from the Iso-Energy Assumption, which exploits cross-modal redundancy: a concept that is truly shared should exhibit the same average energy across modalities. We operationalize this assumption with an Aligned Sparse Autoencoder (SAE) that encourages energy consistency during training while preserving reconstruction. We find that this inductive bias changes the SAE solution without harming reconstruction, giving us a representation that serves as a tool for geometric analysis. Sanity checks on controlled data with known ground truth confirm that alignment improves when Iso-Energy holds and remains neutral when it does not. Applied to foundational VLMs, our framework reveals a clear structure with practical consequences: (i) sparse bimodal atoms carry the entire cross-modal alignment signal; (ii) unimodal atoms act as modality-specific biases and fully explain the modality gap; (iii) removing unimodal atoms collapses the gap without harming performance; (iv) restricting vector arithmetic to the bimodal subspace yields in-distribution edits and improved retrieval. These findings suggest that the right inductive bias can both preserve model fidelity and render the latent geometry interpretable and actionable.
Authors: Shanlei Mu, Yuchen Jiang, Shikang Wu, Shiyong Hong, Tianmu Sha, Junjie Zhang, Jie Zhu, Zhe Chen, Zhe Wang, Jingjian Lin
Abstract: Industrial recommender systems increasingly adopt multi-scenario learning (MSL) and multi-task learning (MTL) to handle diverse user interactions and contexts, but existing approaches suffer from two critical drawbacks: (1) underutilization of large-scale model parameters due to limited interaction with complex feature modules, and (2) difficulty in jointly modeling scenario and task information in a unified framework. To address these challenges, we propose a unified \textbf{M}ulti-\textbf{D}istribution \textbf{L}earning (MDL) framework, inspired by the "prompting" paradigm in large language models (LLMs). MDL treats scenario and task information as specialized tokens rather than auxiliary inputs or gating signals. Specifically, we introduce a unified information tokenization module that transforms features, scenarios, and tasks into a unified tokenized format. To facilitate deep interaction, we design three synergistic mechanisms: (1) feature token self-attention for rich feature interactions, (2) domain-feature attention for scenario/task-adaptive feature activation, and (3) domain-fused aggregation for joint distribution prediction. By stacking these interactions, MDL enables scenario and task information to "prompt" and activate the model's vast parameter space in a bottom-up, layer-wise manner. Extensive experiments on real-world industrial datasets demonstrate that MDL significantly outperforms state-of-the-art MSL and MTL baselines. Online A/B testing on Douyin Search platform over one month yields +0.0626\% improvement in LT30 and -0.3267\% reduction in change query rate. MDL has been fully deployed in production, serving hundreds of millions of users daily.
Authors: Jinhua Lyu, Tianmin Yu, Ying Ma, Naichen Shi
Abstract: In this work, we investigate the large-scale mean-field variational inference (MFVI) problem from a mini-batch primal-dual perspective. By reformulating MFVI as a constrained finite-sum problem, we develop a novel primal-dual algorithm based on an augmented Lagrangian formulation, termed primal-dual variational inference (PD-VI). PD-VI jointly updates global and local variational parameters in the evidence lower bound in a scalable manner. To further account for heterogeneous loss geometry across different variational parameter blocks, we introduce a block-preconditioned extension, P$^2$D-VI, which adapts the primal-dual updates to the geometry of each parameter block and improves both numerical robustness and practical efficiency. We establish convergence guarantees for both PD-VI and P$^2$D-VI under properly chosen constant step size, without relying on conjugacy assumptions or explicit bounded-variance conditions. In particular, we prove $O(1/T)$ convergence to a stationary point in general settings and linear convergence under strong convexity. Numerical experiments on synthetic data and a real large-scale spatial transcriptomics dataset demonstrate that our methods consistently outperform existing stochastic variational inference approaches in terms of convergence speed and solution quality.
Authors: Yanxiao Liu, Yijun Fan, Deniz G\"und\"uz
Abstract: In this paper, we propose a novel class of change of measure inequalities via a unified framework based on the data processing inequality for $f$-divergences, which is surprisingly elementary yet powerful enough to yield tighter inequalities. We provide change of measure inequalities in terms of a broad family of information measures, including $f$-divergences (with Kullback-Leibler divergence and $\chi^2$-divergence as special cases), R\'enyi divergence, and $\alpha$-mutual information (with maximal leakage as a special case). We then embed these inequalities into the analysis of generalization error for stochastic learning algorithms, yielding novel and tighter high-probability information-theoretic generalization bounds, while also recovering several best-known results via simplified analyses. A key advantage of our framework is its flexibility: it readily adapts to a range of settings, including the conditional mutual information framework, PAC-Bayesian theory, and differential privacy mechanisms, for which we derive new generalization bounds.
Authors: Keonvin Park, Aditya Pal, Jin Hong Mok
Abstract: Food segmentation models trained on static images have achieved strong performance on benchmark datasets; however, their reliability in video settings remains poorly understood. In real-world applications such as food monitoring and instance counting, segmentation outputs must be temporally consistent, yet image-trained models often break down when deployed on videos. In this work, we analyze this failure through an instance segmentation and tracking perspective, focusing on apples as a representative food category. Models are trained solely on image-level food segmentation data and evaluated on video sequences using an instance segmentation with tracking-by-matching framework, enabling object-level temporal analysis. Our results reveal that high frame-wise segmentation accuracy does not translate to stable instance identities over time. Temporal appearance variations, particularly illumination changes, specular reflections, and texture ambiguity, lead to mask flickering and identity fragmentation, resulting in significant errors in apple counting. These failures are largely overlooked by conventional image-based metrics, which substantially overestimate real-world video performance. Beyond diagnosing the problem, we examine practical remedies that do not require full video supervision, including post-hoc temporal regularization and self-supervised temporal consistency objectives. Our findings suggest that the root cause of failure lies in image-centric training objectives that ignore temporal coherence, rather than model capacity. This study highlights a critical evaluation gap in food segmentation research and motivates temporally-aware learning and evaluation protocols for video-based food analysis.
Authors: Scott Thornton
Abstract: Hybrid Retrieval-Augmented Generation (RAG) pipelines combine vector similarity search with knowledge graph expansion for multi-hop reasoning. We show that this composition introduces a distinct security failure mode: a vector-retrieved "seed" chunk can pivot via entity links into sensitive graph neighborhoods, causing cross-tenant data leakage that does not occur in vector-only retrieval. We formalize this risk as Retrieval Pivot Risk (RPR) and introduce companion metrics Leakage@k, Amplification Factor, and Pivot Depth (PD) to quantify leakage magnitude and traversal structure. We present seven Retrieval Pivot Attacks that exploit the vector-to-graph boundary and show that adversarial injection is not required: naturally shared entities create cross-tenant pivot paths organically. Across a synthetic multi-tenant enterprise corpus and the Enron email corpus, the undefended hybrid pipeline exhibits high pivot risk (RPR up to 0.95) with multiple unauthorized items returned per query. Leakage consistently appears at PD=2, which we attribute to the bipartite chunk-entity topology and formalize as a proposition. We then show that enforcing authorization at a single location, the graph expansion boundary, eliminates measured leakage (RPR near 0) across both corpora, all attack variants, and label forgery rates up to 10 percent, with minimal overhead. Our results indicate the root cause is boundary enforcement, not inherently complex defenses: two individually secure retrieval components can compose into an insecure system unless authorization is re-checked at the transition point.