Authors: Alex Echeverria, S\'avio Salvarino Teles de Oliveira, Fernando Marques Federson
Abstract: The adaptation of Large-Scale Language Models (LLMs) to specific domains depends on high-quality fine-tuning datasets, particularly in instructional format (e.g., Question-Answer - Q&A). However, generating these datasets, particularly from unstructured sources such as call center audio recordings, poses a significant challenge due to the noisy and disorganized nature of the data. This paper presents a solution to this challenge by offering an end-to-end automated pipeline for generating Q&A instructional datasets from such recordings. The methodology developed comprises sequential steps of audio processing (including diarization, noise removal and automatic transcription), textual processing (cleaning, normalization, and anonymization), semantic extraction of customer demands and attendant responses using vector embeddings, and matching via semantic search to form the final Q&A pairs. As a result, the complete pipeline was successfully implemented, generating a dataset specifically formatted for Instruct Fine Tuning. The practical value and feasibility of the generated dataset were substantiated and functionally demonstrated through the successful fine-tuning of an LLM model (based on Llama 2 7B). The conclusion of the paper states that the proposed approach is viable for converting unstructured conversational data from call centers into valuable resources for training LLMs. This development has the potential to open up avenues for creating more effective AI systems for Q&A tasks in the customer service domain. The developed codes have been made publicly available to promote reproducibility and future research.
Authors: Ruben Neyroud, Sam Corley
Abstract: While most LLMs are autoregressive, diffusion-based LLMs have recently emerged as an alternative method for generation. Greedy Coordinate Gradient (GCG) attacks have proven effective against autoregressive models, but their applicability to diffusion language models remains largely unexplored. In this work, we present an exploratory study of GCG-style adversarial prompt attacks on LLaDA (Large Language Diffusion with mAsking), an open-source diffusion LLM. We evaluate multiple attack variants, including prefix perturbations and suffix-based adversarial generation, on harmful prompts drawn from the AdvBench dataset. Our study provides initial insights into the robustness and attack surface of diffusion language models and motivates the development of alternative optimization and evaluation strategies for adversarial analysis in this setting.
Authors: Anh-Tuan Mai, Cam-Van Thi Nguyen, Duc-Trong Le
Abstract: Multimodal emotion recognition in conversation (MERC) requires representations that effectively integrate signals from multiple modalities. These signals include modality-specific cues, information shared across modalities, and interactions that emerge only when modalities are combined. In information-theoretic terms, these correspond to \emph{unique}, \emph{redundant}, and \emph{synergistic} contributions. An ideal representation should leverage all three, yet achieving such balance remains challenging. Recent advances in contrastive learning and augmentation-based methods have made progress, but they often overlook the role of data preparation in preserving these components. In particular, applying augmentations directly to raw inputs or fused embeddings can blur the boundaries between modality-unique and cross-modal signals. To address this challenge, we propose a two-phase framework \emph{\textbf{D}ivide and \textbf{R}efine} (\textbf{DnR}). In the \textbf{Divide} phase, each modality is explicitly decomposed into uniqueness, pairwise redundancy, and synergy. In the \textbf{Refine} phase, tailored objectives enhance the informativeness of these components while maintaining their distinct roles. The refined representations are plug-and-play compatible with diverse multimodal pipelines. Extensive experiments on IEMOCAP and MELD demonstrate consistent improvements across multiple MERC backbones. These results highlight the effectiveness of explicitly dividing, refining, and recombining multimodal representations as a principled strategy for advancing emotion recognition. Our implementation is available at https://github.com/mattam301/DnR-WACV2026
Authors: Zewen Yang, Xiaobing Dai, Jiajun Cheng, Yulong Huang, Peng Shi
Abstract: Effective cooperation is pivotal in distributed learning for multi-agent systems, where the interplay between the quantity and quality of the machine learning models is crucial. This paper reveals the irrationality of indiscriminate inclusion of all models on agents for joint prediction, highlighting the imperative to prioritize quality over quantity in cooperative learning. Specifically, we present the first selective online learning framework for distributed Gaussian process (GP) regression, namely distributed error-informed GP (EIGP), that enables each agent to assess its neighboring collaborators, using the proposed selection function to choose the higher quality GP models with less prediction errors. Moreover, algorithmic enhancements are embedded within the EIGP, including a greedy algorithm (gEIGP) for accelerating prediction and an adaptive algorithm (aEIGP) for improving prediction accuracy. In addition, approaches for fast prediction and model update are introduced in conjunction with the error-informed quantification term iteration and a data deletion strategy to achieve real-time learning operations. Numerical simulations are performed to demonstrate the effectiveness of the developed methodology, showcasing its superiority over the state-of-the-art distributed GP methods with different benchmarks.
Authors: Uygar Kurt
Abstract: Quantization is a practical technique for making large language models easier to deploy by reducing the precision used to store and operate on model weights. This can lower memory use and improve runtime feasibility on constrained hardware, which is especially relevant for users running models locally. Quantization in llama.cpp enables large language models to run on commodity hardware, but available formats are often evaluated inconsistently, making it hard to choose among schemes. We present a unified empirical study of the llama.cpp quantization on a single modern model, Llama-3.1-8B-Instruct (FP16, GGUF), covering 3-8 bit K-quant and legacy formats. We evaluate downstream task performance across standard reasoning, knowledge, instruction-following, and truthfulness benchmarks, and also measure perplexity and CPU throughput (prefill/decoding) alongside model size, compression, and quantization time. Ultimately, this work is a practical guide for choosing a llama.cpp quantization scheme, helping readers make informed, context-aware decisions for their intended use and resource budget.
Authors: Brady Steele
Abstract: We investigate learned KV cache compression through Speculative Importance Prediction (SIP), a 1.7M parameter non-query-aware scorer that predicts token importance from KV representations alone. Despite architectural sophistication (multi-horizon lookahead, cross-attention), SIP does not outperform simple baselines, including random selection, across 5 seeds, 4 retention levels, and 3 tasks. Key findings: (1) position-based heuristics (keep first 4 + last N tokens) match or exceed learned approaches; (2) prefill attention provides equivalent signal to complex learned scorers; (3) marginal information in KV representations beyond position and prefill attention appears limited for importance prediction. We hypothesize that circular dependence between future queries and generation trajectories contributes to this difficulty.
Authors: Kangyu Zheng, Kai Zhang, Jiale Tan, Xuehan Chen, Yingzhou Lu, Zaixi Zhang, Lichao Sun, Marinka Zitnik, Tianfan Fu, Zhiding Liang
Abstract: Currently, the field of structure-based drug design is dominated by three main types of algorithms: search-based algorithms, deep generative models, and reinforcement learning. While existing works have typically focused on comparing models within a single algorithmic category, cross-algorithm comparisons remain scarce. In this paper, to fill the gap, we establish a benchmark to evaluate the performance of fifteen models across these different algorithmic foundations by assessing the pharmaceutical properties of the generated molecules and their docking affinities and poses with specified target proteins. We highlight the unique advantages of each algorithmic approach and offer recommendations for the design of future SBDD models. We emphasize that 1D/2D ligand-centric drug design methods can be used in SBDD by treating the docking function as a black-box oracle, which is typically neglected. Our evaluation reveals distinct patterns across model categories. 3D structure-based models excel in binding affinities but show inconsistencies in chemical validity and pose quality. 1D models demonstrate reliable performance in standard molecular metrics but rarely achieve optimal binding affinities. 2D models offer balanced performance, maintaining high chemical validity while achieving moderate binding scores. Through detailed analysis across multiple protein targets, we identify key improvement areas for each model category, providing insights for researchers to combine strengths of different approaches while addressing their limitations. All the code that are used for benchmarking is available in https://github.com/zkysfls/2025-sbdd-benchmark
Authors: Pengyu Liu, Mariel V\'azquez, Nata\v{s}a Jonoska
Abstract: Tree structures appear in many fields of the life sciences, including phylogenetics, developmental biology and nucleic acid structures. Trees can be used to represent RNA secondary structures, which directly relate to the function of non-coding RNAs. Recent developments in sequencing technology and artificial intelligence have yielded numerous biological data that can be represented with tree structures. This requires novel methods for tree structure data analytics. Tree polynomials provide a computationally efficient, interpretable and comprehensive way to encode tree structures as matrices, which are compatible with most data analytics tools. Machine learning methods based on the Canberra distance between tree polynomials have been introduced to analyze phylogenies and nucleic acid structures. In this paper, we compare the performance of different distances in tree clustering methods based on a tree distinguishing polynomial. We also implement two basic autoencoder models for clustering trees using the polynomial. We find that the distance based methods with entry-level normalized distances have the highest clustering accuracy among the compared methods.
Authors: Xiucheng Xu, Bingbing Xu, Xueyun Tian, Zihe Huang, Rongxin Chen, Yunfan Li, Huawei Shen
Abstract: External memory systems are pivotal for enabling Large Language Model (LLM) agents to maintain persistent knowledge and perform long-horizon decision-making. Existing paradigms typically follow a two-stage process: computationally expensive memory construction (e.g., structuring data into graphs) followed by naive retrieval-augmented generation. However, our empirical analysis reveals two fundamental limitations: complex construction incurs high costs with marginal performance gains, and simple context concatenation fails to bridge the gap between retrieval recall and reasoning accuracy. To address these challenges, we propose CoM (Chain-of-Memory), a novel framework that advocates for a paradigm shift toward lightweight construction paired with sophisticated utilization. CoM introduces a Chain-of-Memory mechanism that organizes retrieved fragments into coherent inference paths through dynamic evolution, utilizing adaptive truncation to prune irrelevant noise. Extensive experiments on the LongMemEval and LoCoMo benchmarks demonstrate that CoM outperforms strong baselines with accuracy gains of 7.5%-10.4%, while drastically reducing computational overhead to approximately 2.7% of token consumption and 6.0% of latency compared to complex memory architectures.
Authors: Jun Liu, Leo Yu Zhang, Fengpeng Li, Isao Echizen, Jiantao Zhou
Abstract: Hard-label black-box settings, where only top-1 predicted labels are observable, pose a fundamentally constrained yet practically important feedback model for understanding model behavior. A central challenge in this regime is whether meaningful gradient information can be recovered from such discrete responses. In this work, we develop a unified theoretical perspective showing that a wide range of existing sign-flipping hard-label attacks can be interpreted as implicitly approximating the sign of the true loss gradient. This observation reframes hard-label attacks from heuristic search procedures into instances of gradient sign recovery under extremely limited feedback. Motivated by this first-principles understanding, we propose a new attack framework that combines a zero-query frequency-domain initialization with a Pattern-Driven Optimization (PDO) strategy. We establish theoretical guarantees demonstrating that, under mild assumptions, our initialization achieves higher expected cosine similarity to the true gradient sign compared to random baselines, while the proposed PDO procedure attains substantially lower query complexity than existing structured search approaches. We empirically validate our framework through extensive experiments on CIFAR-10, ImageNet, and ObjectNet, covering standard and adversarially trained models, commercial APIs, and CLIP-based models. The results show that our method consistently surpasses SOTA hard-label attacks in both attack success rate and query efficiency, particularly in low-query regimes. Beyond image classification, our approach generalizes effectively to corrupted data, biomedical datasets, and dense prediction tasks. Notably, it also successfully circumvents Blacklight, a SOTA stateful defense, resulting in a $0\%$ detection rate. Our code will be released publicly soon at https://github.com/csjunjun/DPAttack.git.
Authors: YuanLab. ai, Shawn Wu, Jiangang Luo, Tong Yu, Darcy Chen, Sean Wang, Xudong Zhao, Louie Li, Claire Wang, Hunter He, Carol Wang, Allen Wang
Abstract: Although Mixture-of-Experts (MoE) Large Language Models (LLMs) deliver superior accuracy with a reduced number of active parameters, their pre-training represents a significant computationally bottleneck due to underutilized experts and limited training efficiency. This work introduces a Layer-Adaptive Expert Pruning (LAEP) algorithm designed for the pre-training stage of MoE LLMs. In contrast to previous expert pruning approaches that operate primarily in the post-training phase, the proposed algorithm enhances training efficiency by selectively pruning underutilized experts and reorganizing experts across computing devices according to token distribution statistics. Comprehensive experiments demonstrate that LAEP effectively reduces model size and substantially improves pre-training efficiency. In particular, when pre-training the 1010B Base model from scratch, LAEP achieves a 48.3\% improvement in training efficiency alongside a 33.3% parameter reduction, while still delivering excellent performance across multiple domains.
Authors: Anupam Agrawal, Rajesh Mohanty, Shamik Bhattacharjee, Abhimanyu Mittal
Abstract: Contextual Bandit (CB) algorithms are widely adopted for personalized recommendations but often struggle in dynamic environments typical of fantasy sports, where rapid changes in user behavior and dramatic shifts in reward distributions due to external influences necessitate frequent retraining. To address these challenges, we propose a Hierarchical Contextual Uplift Bandit framework. Our framework dynamically adjusts contextual granularity from broad, system-wide insights to detailed, user-specific contexts, using contextual similarity to facilitate effective policy transfer and mitigate cold-start issues. Additionally, we integrate uplift modeling principles into our approach. Results from large-scale A/B testing on the Dream11 fantasy sports platform show that our method significantly enhances recommendation quality, achieving a 0.4% revenue improvement while also improving user satisfaction metrics compared to the current production system. We subsequently deployed this system to production as the default catalog personalization system in May 2025 and observed a further 0.5% revenue improvement.
Authors: Krishna Sharma, Vivek Yelleti
Abstract: Log anomaly detection is essential for system reliability, but it is extremely challenging to do considering it involves class imbalance. Additionally, the models trained in one domain are not applicable to other domains, necessitating the need for cross-domain adaptation (such as HDFS and Linux). Traditional detection models often fail to generalize due to significant data drift and the inherent absence of labeled anomalies in new target domains. To handle the above challenges, we proposed a new end-to-end framework based on a meta-learning approach. Our methodology first gets the data ready by combining a Drain3 log parsing mechanism with a dynamic drift-based labeling technique that uses semantic and fuzzy matching to move existing anomaly knowledge from one source to another. BERT-based semantic embeddings are obtained, and the feature selection is invoked to reduce the dimensionality. Later, Model Agnostic Meta-Learning (MAML) and Prototypical Networks models are trained to adapt quickly and effectively. The SMOTE oversampling method is employed to handle imbalances in the data. All the results are obtained by employing the leave-one-out source method, and the corresponding mean F1 scores are reported. Our empirical findings validate that the proposed meta-learning-driven approach yielded the highest mean F1 score and proved to be effective for cross-domain settings.
Authors: Yewon Han, Sunghyun Kim, Eunyi Jeong, Sungkyung Lee, Seokwoo Yun, Sangsoo Lim
Abstract: Accurate prediction of drug response in precision medicine requires models that capture how specific chemical substructures interact with cellular pathway states. However, most existing deep learning approaches treat chemical and transcriptomic modalities independently or combine them only at late stages, limiting their ability to model fine-grained, context-dependent mechanisms of drug action. In addition, standard attention mechanisms are often sensitive to noise and sparsity in high-dimensional biological networks, hindering both generalization and interpretability. We present DiSPA, a representation learning framework that explicitly disentangles structure-driven and context-driven mechanisms of drug response through bidirectional conditioning between chemical substructures and pathway-level gene expression. DiSPA introduces a differential cross-attention module that suppresses spurious pathway-substructure associations while amplifying contextually relevant interactions. Across multiple evaluation settings on the GDSC benchmark, DiSPA achieves state-of-the-art performance, with particularly strong improvements in the disjoint-set setting, which assesses generalization to unseen drug-cell combinations. Beyond predictive accuracy, DiSPA yields mechanistically informative representations: learned attention patterns recover known pharmacophores, distinguish structure-driven from context-dependent compounds, and exhibit coherent organization across biological pathways. Furthermore, we demonstrate that DiSPA trained solely on bulk RNA-seq data enables zero-shot transfer to spatial transcriptomics, revealing region-specific drug sensitivity patterns without retraining. Together, these results establish DiSPA as a robust and interpretable framework for integrative pharmacogenomic modeling, enabling principled analysis of drug response mechanisms beyond post hoc interpretation.
Authors: Yongchao Huang
Abstract: Joint Embedding Predictive Architectures (JEPA) offer a scalable paradigm for self-supervised learning by predicting latent representations rather than reconstructing high-entropy observations. However, existing formulations rely on \textit{deterministic} regression objectives, which mask probabilistic semantics and limit its applicability in stochastic control. In this work, we introduce \emph{Variational JEPA (VJEPA)}, a \textit{probabilistic} generalization that learns a predictive distribution over future latent states via a variational objective. We show that VJEPA unifies representation learning with Predictive State Representations (PSRs) and Bayesian filtering, establishing that sequential modeling does not require autoregressive observation likelihoods. Theoretically, we prove that VJEPA representations can serve as sufficient information states for optimal control without pixel reconstruction, while providing formal guarantees for collapse avoidance. We further propose \emph{Bayesian JEPA (BJEPA)}, an extension that factorizes the predictive belief into a learned dynamics expert and a modular prior expert, enabling zero-shot task transfer and constraint (e.g. goal, physics) satisfaction via a Product of Experts. Empirically, through a noisy environment experiment, we demonstrate that VJEPA and BJEPA successfully filter out high-variance nuisance distractors that cause representation collapse in generative baselines. By enabling principled uncertainty estimation (e.g. constructing credible intervals via sampling) while remaining likelihood-free regarding observations, VJEPA provides a foundational framework for scalable, robust, uncertainty-aware planning in high-dimensional, noisy environments.
Authors: Danny Butvinik, Nana Boateng, Achi Hackmon
Abstract: We study the problem of converting a stream of risk scores into one or more review queues under explicit intake constraints[cite: 6]. Instead of top-$K$ or manually tuned cutoffs, we fit an online adaptive kernel density to the score stream, transform the density into a tail-mass curve to meet capacity, and ``snap'' the resulting cut to a persistent density valley detected across bandwidths[cite: 7]. The procedure is label-free, supports multi-queue routing, and operates in real time with sliding windows or exponential forgetting[cite: 8]. On synthetic, drifting, multimodal streams, the method achieves competitive capacity adherence while reducing threshold jitter[cite: 9]. Updates cost $O(G)$ per event with constant memory per activity
Authors: Naoya Onizawa, Takahiro Hanyu
Abstract: Probabilistic computing using probabilistic bits (p-bits) presents an efficient alternative to traditional CMOS logic for complex problem-solving, including simulated annealing and machine learning. Realizing p-bits with emerging devices such as magnetic tunnel junctions (MTJs) introduces device variability, which was expected to negatively impact computational performance. However, this study reveals an unexpected finding: device variability can not only degrade but also enhance algorithm performance, particularly by leveraging timing variability. This paper introduces a GPU-accelerated, open-source simulated annealing framework based on p-bits that models key device variability factors -- timing, intensity, and offset -- to reflect real-world device behavior. Through CUDA-based simulations, our approach achieves a two-order magnitude speedup over CPU implementations on the MAX-CUT benchmark with problem sizes ranging from 800 to 20,000 nodes. By providing a scalable and accessible tool, this framework aims to advance research in probabilistic computing, enabling optimization applications in diverse fields.
Authors: Mrigank Dhingra, Omer San
Abstract: Long-horizon autoregressive forecasting of chaotic dynamical systems remains challenging due to rapid error amplification and distribution shift: small one-step inaccuracies compound into physically inconsistent rollouts and collapse of large-scale statistics. We introduce MSR-HINE, a hierarchical implicit forecaster that augments multiscale latent priors with multi-rate recurrent modules operating at distinct temporal scales. At each step, coarse-to-fine recurrent states generate latent priors, an implicit one-step predictor refines the state with multiscale latent injections, and a gated fusion with posterior latents enforces scale-consistent updates; a lightweight hidden-state correction further aligns recurrent memories with fused latents. The resulting architecture maintains long-term context on slow manifolds while preserving fast-scale variability, mitigating error accumulation in chaotic rollouts. Across two canonical benchmarks, MSR-HINE yields substantial gains over a U-Net autoregressive baseline: on Kuramoto-Sivashinsky it reduces end-horizon RMSE by 62.8% at H=400 and improves end-horizon ACC by +0.983 (from -0.155 to 0.828), extending the ACC >= 0.5 predictability horizon from 241 to 400 steps; on Lorenz-96 it reduces RMSE by 27.0% at H=100 and improves end horizon ACC by +0.402 (from 0.144 to 0.545), extending the ACC >= 0.5 horizon from 58 to 100 steps.
Authors: Yilong Dai, Shengyu Chen, Ziyi Wang, Xiaowei Jia, Yiqun Xie, Vipin Kumar, Runlong Yu
Abstract: Partial differential equations (PDEs) are central to scientific modeling. Modern workflows increasingly rely on learning-based components to support model reuse, inference, and integration across large computational processes. Despite the emergence of various physics-aware data-driven approaches, the field still lacks a unified perspective to uncover their relationships, limitations, and appropriate roles in scientific workflows. To this end, we propose a unifying perspective to place two dominant paradigms: Physics-Informed Neural Networks (PINNs) and Neural Operators (NOs), within a shared design space. We organize existing methods from three fundamental dimensions: what is learned, how physical structures are integrated into the learning process, and how the computational load is amortized across problem instances. In this way, many challenges can be best understood as consequences of these structural properties of learning PDEs. By analyzing advances through this unifying view, our survey aims to facilitate the development of reliable learning-based PDE solvers and catalyze a synthesis of physics and data.
Authors: Giulio Rossolini
Abstract: Adversarial attacks are widely used to evaluate model robustness, yet their validity as proxies for robustness to random perturbations remains debated. We ask whether an adversarial perturbation provides a representative estimate of robustness under random noise of the same magnitude, or instead reflects an atypical worst-case event. To this end, we introduce a probabilistic metric that quantifies noisy risk with respect to directionally biased perturbation distributions, parameterized by a concentration factor $\kappa$ that interpolates between isotropic noise and adversarial direction. Using this framework, we study the limits of adversarial perturbations as estimators of noisy risk by proposing an attack strategy designed to operate in regimes statistically closer to uniform noise. Experiments on ImageNet and CIFAR-10 systematically benchmark widely used attacks, highlighting when adversarial success meaningfully reflects noisy risk and when it fails, thereby informing their use in safety-oriented evaluation.
Authors: Hunjae Lee, Corey Clark
Abstract: In decoder-only (causal) transformers, the computation graph created by causal masking routes information through both direct-path attention and indirect paths formed by intermediate tokens. We denote these indirect paths between token pairs as their runways. We argue that certain failure modes of causal transformers as observed by a growing body of recent works are likely exacerbated by a misalignment between these two information propagation modes. We formalize runway cascade as a phenomenon whereby this misalignment results in redundancies and irrelevant information cascading to token representations despite adequately learned attention patterns. As a solution, we propose runway-aware rewiring as a more explicit way of incorporating runway context directly into each token's direct-path attention. This mechanism re-wires the attention pattern for each token based on a summary of its runway landscape, enabling awareness of accumulating representational influences and allowing for more balanced information propagation. Our proposed methodology introduces no additional parameters and can seamlessly be integrated into standard attention mechanism. Empirically, our rewired transformer results in steady improvements in general language modeling as well as noticeably stronger information retrieval and extrapolation abilities compared to standard transformers.
Authors: Alistair Cheong, Haolin Cong, Tyler Yang, Dustin Miao
Abstract: Many LLM-based open-ended search systems freeze the foundation model that proposes improvements to existing solutions, which may bottleneck long-run progress. Recent work has explored updating the proposal model at test time [arXiv:2511.23473], but the update strategy is still typically hand-specified. Therefore, this study investigated whether an LLM can use task feedback to decide how it should update its weights. For tractability, we focused on the simpler case where there is only one round of self-improvement, and restricted the update operator to self-supervised next token prediction (NTP), leaving the model freedom in choosing its training data and key NTP hyperparameters. Using the Self-Adapting Language Models (SEAL) [arXiv:2506.10943] framework as a testbed, we relaxed its fixed human template constraint and allowed the model to generate its own self-edit templates, thereby giving it more control over its training data and hyperparameters. Two variants were studied, differing in whether template generation was conditioned on a lightweight archive of past templates. In SEAL's Single-Passage Knowledge Incorporation setting with Qwen3-8B on SQuAD [arXiv:1606.05250], the no-archive variant performed comparably to the weaker "Implications" baseline, while the archive variant outperformed "Implications" and approached the strongest human-designed "Rewrite" baseline without surpassing it. Further analysis of collapse in the model's exploration revealed that a naive archive can confer some short-term robustness but can also accelerate homogenization, suggesting that explicit novelty pressure may be required to consistently advance beyond carefully optimized human strategies. Our code is available at https://github.com/cheongalc/search-self-edit-strategies .
URLs: https://github.com/cheongalc/search-self-edit-strategies
Authors: Tiantian Yang, Yuxuan Wang, Zhenwei Zhou, Ching-Ti Liu
Abstract: Omics data, such as transcriptomics, proteomics, and metabolomics, provide critical insights into disease mechanisms and clinical outcomes. However, their high dimensionality, small sample sizes, and intricate biological networks pose major challenges for reliable prediction and meaningful interpretation. Graph Neural Networks (GNNs) offer a promising way to integrate prior knowledge by encoding feature relationships as graphs. Yet, existing methods typically rely solely on either an externally curated feature graph or a data-driven generated one, which limits their ability to capture complementary information. To address this, we propose the external and generated Graph Neural Network (engGNN), a dual-graph framework that jointly leverages both external known biological networks and data-driven generated graphs. Specifically, engGNN constructs a biologically informed undirected feature graph from established network databases and complements it with a directed feature graph derived from tree-ensemble models. This dual-graph design produces more comprehensive embeddings, thereby improving predictive performance and interpretability. Through extensive simulations and real-world applications to gene expression data, engGNN consistently outperforms state-of-the-art baselines. Beyond classification, engGNN provides interpretable feature importance scores that facilitate biologically meaningful discoveries, such as pathway enrichment analysis. Taken together, these results highlight engGNN as a robust, flexible, and interpretable framework for disease classification and biomarker discovery in high-dimensional omics contexts.
Authors: Deming Chen (Celine), Vijay Ganesh (Celine), Weikai Li (Celine), Yingyan (Celine), Lin, Yong Liu, Subhasish Mitra, David Z. Pan, Ruchir Puri, Jason Cong, Yizhou Sun
Abstract: This report distills the discussions and recommendations from the NSF Workshop on AI for Electronic Design Automation (EDA), held on December 10, 2024 in Vancouver alongside NeurIPS 2024. Bringing together experts across machine learning and EDA, the workshop examined how AI-spanning large language models (LLMs), graph neural networks (GNNs), reinforcement learning (RL), neurosymbolic methods, etc.-can facilitate EDA and shorten design turnaround. The workshop includes four themes: (1) AI for physical synthesis and design for manufacturing (DFM), discussing challenges in physical manufacturing process and potential AI applications; (2) AI for high-level and logic-level synthesis (HLS/LLS), covering pragma insertion, program transformation, RTL code generation, etc.; (3) AI toolbox for optimization and design, discussing frontier AI developments that could potentially be applied to EDA tasks; and (4) AI for test and verification, including LLM-assisted verification tools, ML-augmented SAT solving, security/reliability challenges, etc. The report recommends NSF to foster AI/EDA collaboration, invest in foundational AI for EDA, develop robust data infrastructures, promote scalable compute infrastructure, and invest in workforce development to democratize hardware design and enable next-generation hardware systems. The workshop information can be found on the website https://ai4eda-workshop.github.io/.
Authors: Nilesh Prasad Pandey, Jangseon Park, Onat Gungor, Flavio Ponzina, Tajana Rosing
Abstract: Deploying Small Language Models (SLMs) on edge platforms is critical for real-time, privacy-sensitive generative AI, yet constrained by memory, latency, and energy budgets. Quantization reduces model size and cost but suffers from device noise in emerging non-volatile memories, while conventional memory hierarchies further limit efficiency. SRAM provides fast access but has low density, DRAM must simultaneously accommodate static weights and dynamic KV caches, which creates bandwidth contention, and Flash, although dense, is primarily used for initialization and remains inactive during inference. These limitations highlight the need for hybrid memory organizations tailored to LLM inference. We propose Outlier-aware Quantization with Memory Co-design (QMC), a retraining-free quantization with a novel heterogeneous memory architecture. QMC identifies inlier and outlier weights in SLMs, storing inlier weights in compact multi-level Resistive-RAM (ReRAM) while preserving critical outliers in high-precision on-chip Magnetoresistive-RAM (MRAM), mitigating noise-induced degradation. On language modeling and reasoning benchmarks, QMC outperforms and matches state-of-the-art quantization methods using advanced algorithms and hybrid data formats, while achieving greater compression under both algorithm-only evaluation and realistic deployment settings. Specifically, compared against SoTA quantization methods on the latest edge AI platform, QMC reduces memory usage by 6.3x-7.3x, external data transfers by 7.6x, energy by 11.7x, and latency by 12.5x when compared to FP16, establishing QMC as a scalable, deployment-ready co-design for efficient on-device inference.
Authors: Andrew Crossman, Jonah Dodd, Viralam Ramamurthy Chaithanya Kumar, Riyaz Mohammed, Andrew R. Plummer, Chandra Sekharudu, Deepak Warrier, Mohammad Yekrangian
Abstract: MITRE ATT&CK is a cybersecurity knowledge base that organizes threat actor and cyber-attack information into a set of tactics describing the reasons and goals threat actors have for carrying out attacks, with each tactic having a set of techniques that describe the potential methods used in these attacks. One major application of ATT&CK is the use of its tactic and technique hierarchy by security specialists as a framework for annotating cyber-threat intelligence reports, vulnerability descriptions, threat scenarios, inter alia, to facilitate downstream analyses. To date, the tagging process is still largely done manually. In this technical note, we provide a stratified "task space" characterization of the MITRE ATT&CK text tagging task for organizing previous efforts toward automation using AIML methods, while also clarifying pathways for constructing new methods. To illustrate one of the pathways, we use the task space strata to stage-wise construct our own multi-label hierarchical classification models for the text tagging task via experimentation over general cyber-threat intelligence text -- using shareable computational tools and publicly releasing the models to the security community (via https://github.com/jpmorganchase/MITRE_models). Our multi-label hierarchical approach yields accuracy scores of roughly 94% at the tactic level, as well as accuracy scores of roughly 82% at the technique level. The models also meet or surpass state-of-the-art performance while relying only on classical machine learning methods -- removing any dependence on LLMs, RAG, agents, or more complex hierarchical approaches. Moreover, we show that GPT-4o model performance at the tactic level is significantly lower (roughly 60% accuracy) than our own approach. We also extend our baseline model to a corpus of threat scenarios for financial applications produced by subject matter experts.
Authors: Xiaojie Yang, Dizhi Huang, Hangli Ge, Masahiro Sano, Takeaki Ohdake, Kazuma Hatano, Noboru Koshizuka
Abstract: Accurate forecasting of daily attendance is vital for managing transportation, crowd flows, and services at large-scale international events such as Expo 2025 Osaka, Kansai, Japan. However, existing approaches often rely on multi-source external data (such as weather, traffic, and social media) to improve accuracy, which can lead to unreliable results when historical data are insufficient. To address these challenges, we propose a Transformer-based framework that leverages reservation dynamics, i.e., ticket bookings and subsequent updates within a time window, as a proxy for visitors' attendance intentions, under the assumption that such intentions are eventually reflected in reservation patterns. This design avoids the complexity of multi-source integration while still capturing external influences like weather and promotions implicitly embedded in reservation dynamics. We construct a dataset combining entrance records and reservation dynamics and evaluate the model under both single-channel (total attendance) and two-channel (separated by East and West gates) settings. Results show that separately modeling East and West gates consistently improves accuracy, particularly for short- and medium-term horizons. Ablation studies further confirm the importance of the encoder-decoder structure, inverse-style embedding, and adaptive fusion module. Overall, our findings indicate that reservation dynamics offer a practical and informative foundation for attendance forecasting in large-scale international events.
Authors: Shovito Barua Soumma, Asiful Arefeen, Stephanie M. Carpenter, Melanie Hingle, Hassan Ghasemzadeh
Abstract: Counterfactual explanations (CFEs) provide human-centric interpretability by identifying the minimal, actionable changes required to alter a machine learning model's prediction. Therefore, CFs can be used as (i) interventions for abnormality prevention and (ii) augmented data for training robust models. We conduct a comprehensive evaluation of CF generation using large language models (LLMs), including GPT-4 (zero-shot and few-shot) and two open-source models-BioMistral-7B and LLaMA-3.1-8B, in both pretrained and fine-tuned configurations. Using the multimodal AI-READI clinical dataset, we assess CFs across three dimensions: intervention quality, feature diversity, and augmentation effectiveness. Fine-tuned LLMs, particularly LLaMA-3.1-8B, produce CFs with high plausibility (up to 99%), strong validity (up to 0.99), and realistic, behaviorally modifiable feature adjustments. When used for data augmentation under controlled label-scarcity settings, LLM-generated CFs substantially restore classifier performance, yielding an average 20% F1 recovery across three scarcity scenarios. Compared with optimization-based baselines such as DiCE, CFNOW, and NICE, LLMs offer a flexible, model-agnostic approach that generates more clinically actionable and semantically coherent counterfactuals. Overall, this work demonstrates the promise of LLM-driven counterfactuals for both interpretable intervention design and data-efficient model training in sensor-based digital health. Impact: SenseCF fine-tunes an LLM to generate valid, representative counterfactual explanations and supplement minority class in an imbalanced dataset for improving model training and boosting model robustness and predictive performance
Authors: Xiao Hu, Hong Xie, Tao Tan, Defu Lian, Jianyu Han
Abstract: A large number of heuristics have been proposed to optimize the reinforcement fine-tuning of LLMs. However, inconsistent claims are made from time to time, making this area elusive. Reflecting on this situation, two fundamental questions still lack a clear understanding: 1) what is the role of each optimizing choice? 2) which ones are the bottlenecks? This paper aims to shed light on them, and it faces the challenge of several entangled confounding factors in the fine-tuning process. To tackle this challenge, we propose a bottom-up experiment pipeline. The bottom layer is composed of a minimalist configuration: one training data, one rollout per round and the reward directly serve as the learning signal without advantage function design. This minimalist configuration connects to multi-armed bandit learning with extremely large discrete action space, which offers theories to corroborate the experiment findings. The up procedure of the experiment pipeline expanding the minimalist configuration layer by layer, examining the role of each design choice. Experimental results on three LLMs and two reasoning datasets not only reveal new understanding of the design choice but also yield essential insights to shape the area.
Authors: Jingru Li, Yibo Fan, Huan Li
Abstract: Large Language Models (LLMs) achieve competitive performance across diverse natural language processing (NLP) tasks, yet pretraining is computationally demanding, making optimizer efficiency an important practical consideration. Muon accelerates LLM pretraining via orthogonal momentum updates that serve as a matrix analogue of the element-wise sign operator. Motivated by the recent perspective that Adam is a variance-adaptive sign update algorithm, we propose two variants of Muon, Muon-NSR and Muon-VS, which apply variance-adaptive normalization to momentum before orthogonalization. Muon-NSR applies noise-to-signal ratio (NSR) modulation, while Muon-VS performs variance-based scaling without introducing additional hyperparameters. Experiments on GPT-2 and LLaMA pretraining demonstrate that our proposed methods accelerate convergence and consistently achieve lower validation loss than both competitive, well-tuned AdamW and Muon baselines. For example, on the LLaMA-1.2B model, Muon-NSR and Muon-VS reduce the iterations required to reach the target validation loss by $1.36\times$ relative to the well-tuned Muon following the recent benchmark.
Authors: Yvonne Yang, Eranki Vasistha
Abstract: Credit default risk arises from complex interactions among borrowers, financial institutions, and transaction-level behaviors. While strong tabular models remain highly competitive in credit scoring, they may fail to explicitly capture cross-entity dependencies embedded in multi-table financial histories. In this work, we construct a massive-scale heterogeneous graph containing over 31 million nodes and more than 50 million edges, integrating borrower attributes with granular transaction-level entities such as installment payments, POS cash balances, and credit card histories. We evaluate heterogeneous graph neural networks (GNNs), including heterogeneous GraphSAGE and a relation-aware attentive heterogeneous GNN, against strong tabular baselines. We find that standalone GNNs provide limited lift over a competitive gradient-boosted tree baseline, while a hybrid ensemble that augments tabular features with GNN-derived customer embeddings achieves the best overall performance, improving both ROC-AUC and PR-AUC. We further observe that contrastive pretraining can improve optimization stability but yields limited downstream gains under generic graph augmentations. Finally, we conduct structured explainability and fairness analyses to characterize how relational signals affect subgroup behavior and screening-oriented outcomes.
Authors: Yuyu Liu, Jiannan Yang, Ziyang Yu, Weishen Pan, Fei Wang, Tengfei Ma
Abstract: Missing data in single-cell sequencing datasets poses significant challenges for extracting meaningful biological insights. However, existing imputation approaches, which often assume uniformity and data completeness, struggle to address cases with large patches of missing data. In this paper, we present CROT, an optimal transport-based imputation algorithm designed to handle patch-based missing data in tabular formats. Our approach effectively captures the underlying data structure in the presence of significant missingness. Notably, it achieves superior imputation accuracy while significantly reducing runtime, demonstrating its scalability and efficiency for large-scale datasets. This work introduces a robust solution for imputation in heterogeneous, high-dimensional datasets with structured data absence, addressing critical challenges in both biological and clinical data analysis. Our code is available at Anomalous Github.
Authors: Zhihao Chen, Zirui Gong, Jianting Ning, Yanjun Zhang, Leo Yu Zhang
Abstract: Federated Rank Learning (FRL) is a promising Federated Learning (FL) paradigm designed to be resilient against model poisoning attacks due to its discrete, ranking-based update mechanism. Unlike traditional FL methods that rely on model updates, FRL leverages discrete rankings as a communication parameter between clients and the server. This approach significantly reduces communication costs and limits an adversary's ability to scale or optimize malicious updates in the continuous space, thereby enhancing its robustness. This makes FRL particularly appealing for applications where system security and data privacy are crucial, such as web-based auction and bidding platforms. While FRL substantially reduces the attack surface, we demonstrate that it remains vulnerable to a new class of local model poisoning attack, i.e., fine-grained control attacks. We introduce the Edge Control Attack (ECA), the first fine-grained control attack tailored to ranking-based FL frameworks. Unlike conventional denial-of-service (DoS) attacks that cause conspicuous disruptions, ECA enables an adversary to precisely degrade a competitor's accuracy to any target level while maintaining a normal-looking convergence trajectory, thereby avoiding detection. ECA operates in two stages: (i) identifying and manipulating Ascending and Descending Edges to align the global model with the target model, and (ii) widening the selection boundary gap to stabilize the global model at the target accuracy. Extensive experiments across seven benchmark datasets and nine Byzantine-robust aggregation rules (AGRs) show that ECA achieves fine-grained accuracy control with an average error of only 0.224%, outperforming the baseline by up to 17x. Our findings highlight the need for stronger defenses against advanced poisoning attacks. Our code is available at: https://github.com/Chenzh0205/ECA
Authors: Jianwen Sun, Xinrui Li, Fuqing Li, Xiaoxuan Shen
Abstract: Symbolic Regression aims to automatically identify compact and interpretable mathematical expressions that model the functional relationship between input and output variables. Most existing search-based symbolic regression methods typically rely on the fitting error to inform the search process. However, in the vast expression space, numerous candidate expressions may exhibit similar error values while differing substantially in structure, leading to ambiguous search directions and hindering convergence to the underlying true function. To address this challenge, we propose a novel framework named EGRL-SR (Experience-driven Goal-conditioned Reinforcement Learning for Symbolic Regression). In contrast to traditional error-driven approaches, EGRL-SR introduces a new perspective: leveraging precise historical trajectories and optimizing the action-value network to proactively guide the search process, thereby achieving a more robust expression search. Specifically, we formulate symbolic regression as a goal-conditioned reinforcement learning problem and incorporate hindsight experience replay, allowing the action-value network to generalize common mapping patterns from diverse input-output pairs. Moreover, we design an all-point satisfaction binary reward function that encourages the action-value network to focus on structural patterns rather than low-error expressions, and concurrently propose a structure-guided heuristic exploration strategy to enhance search diversity and space coverage. Experiments on public benchmarks show that EGRL-SR consistently outperforms state-of-the-art methods in recovery rate and robustness, and can recover more complex expressions under the same search budget. Ablation results validate that the action-value network effectively guides the search, with both the reward function and the exploration strategy playing critical roles.
Authors: Pengfei Ding, Yan Wang, Guanfeng Liu
Abstract: Graph unlearning (GU), which removes nodes, edges, or features from trained graph neural networks (GNNs), is crucial in Web applications where graph data may contain sensitive, mislabeled, or malicious information. However, existing GU methods lack a clear understanding of the key factors that determine unlearning effectiveness, leading to three fundamental limitations: (1) impractical and inaccurate GU difficulty assessment due to test-access requirements and invalid assumptions, (2) ineffectiveness on hard-to-unlearn tasks, and (3) misaligned evaluation protocols that overemphasize easy tasks and fail to capture true forgetting capability. To address these issues, we establish GNN memorization as a new perspective for understanding graph unlearning and propose MGU, a Memorization-guided Graph Unlearning framework. MGU achieves three key advances: it provides accurate and practical difficulty assessment across different GU tasks, develops an adaptive strategy that dynamically adjusts unlearning objectives based on difficulty levels, and establishes a comprehensive evaluation protocol that aligns with practical requirements. Extensive experiments on ten real-world graphs demonstrate that MGU consistently outperforms state-of-the-art baselines in forgetting quality, computational efficiency, and utility preservation.
Authors: Yutong Chen, Jiandong Gao, Ji Wu
Abstract: Training Large Reasoning Model (LRM) is usually unstable and unpredictable, especially on hard problems or weak foundation models. We found that the current post-training scaling strategy can still improve on these cases. We propose CoScale-RL, a novel scaling strategy with better data and computational efficiency. We first scale up solutions to make problems solvable. The core idea is to collect multiple solutions for each problem, rather than simply enlarging the dataset. Then, we scale up rollout computation to stabilize Reinforcement Learning. We further leverage a model merge technique called Re-distillation to sustain or even improve computational efficiency when scaling up. Our method significantly improves data and computational efficiency, with an average 3.76$\times$ accuracy improvement on four benchmarks. CoScale-RL is able to improve an LRM's ability boundary without an extensive SFT dataset. Our method provides a new scaling direction to further improve LRM's reasoning ability.
Authors: Tianchi Chen, Jan Bima, Sean L. Wu, Otto Ritter, Bingjia Yang, Xiang Yu
Abstract: Optimally sequencing experimental assays in drug discovery is a high-stakes planning problem under severe uncertainty and resource constraints. A primary obstacle for standard reinforcement learning (RL) is the absence of an explicit environment simulator or transition data $(s, a, s')$; planning must rely solely on a static database of historical outcomes. We introduce the Implicit Bayesian Markov Decision Process (IBMDP), a model-based RL framework designed for such simulator-free settings. IBMDP constructs a case-guided implicit model of transition dynamics by forming a nonparametric belief distribution using similar historical outcomes. This mechanism enables Bayesian belief updating as evidence accumulates and employs ensemble MCTS planning to generate stable policies that balance information gain toward desired outcomes with resource efficiency. We validate IBMDP through comprehensive experiments. On a real-world central nervous system (CNS) drug discovery task, IBMDP reduced resource consumption by up to 92\% compared to established heuristics while maintaining decision confidence. To rigorously assess decision quality, we also benchmarked IBMDP in a synthetic environment with a computable optimal policy. Our framework achieves significantly higher alignment with this optimal policy than a deterministic value iteration alternative that uses the same similarity-based model, demonstrating the superiority of our ensemble planner. IBMDP offers a practical solution for sequential experimental design in data-rich but simulator-poor domains.
Authors: Yao Lu, Dengdong Fan, Jianzheng Nie, Fan Xu, Jie Chen, Bin Zhou, Yonghong Tian
Abstract: We present PCL-Reasoner-V1.5, a 32-billion-parameter large language model (LLM) for mathematical reasoning. The model is built upon Qwen2.5-32B and refined via supervised fine-tuning (SFT) followed by reinforcement learning (RL). A central innovation is our proposed offline RL method, which provides superior training stability and efficiency over standard online RL methods such as GRPO. Our model achieves state-of-the-art performance among models post-trained on Qwen2.5-32B, attaining average accuracies of 90.9% on AIME 2024 and 85.6% on AIME 2025. Our work demonstrates offline RL as a stable and efficient paradigm for advancing reasoning in LLMs. All experiments were conducted on Huawei Ascend 910C NPUs.
Authors: Bizu Feng, Zhimu Yang, Shaode Yu, Zixin Hu
Abstract: Despite the widespread success of Graph Neural Networks (GNNs), understanding the reasons behind their specific predictions remains challenging. Existing explainability methods face a trade-off that gradient-based approaches are computationally efficient but often ignore structural interactions, while game-theoretic techniques capture interactions at the cost of high computational overhead and potential deviation from the model's true reasoning path. To address this gap, we propose FSX (Message Flow Sensitivity Enhanced Structural Explainer), a novel hybrid framework that synergistically combines the internal message flows of the model with a cooperative game approach applied to the external graph data. FSX first identifies critical message flows via a novel flow-sensitivity analysis: during a single forward pass, it simulates localized node perturbations and measures the resulting changes in message flow intensities. These sensitivity-ranked flows are then projected onto the input graph to define compact, semantically meaningful subgraphs. Within each subgraph, a flow-aware cooperative game is conducted, where node contributions are evaluated fairly through a Shapley-like value that incorporates both node-feature importance and their roles in sustaining or destabilizing the identified critical flows. Extensive evaluation across multiple datasets and GNN architectures demonstrates that FSX achieves superior explanation fidelity with significantly reduced runtime, while providing unprecedented insights into the structural logic underlying model predictions--specifically, how important sub-structures exert influence by governing the stability of key internal computational pathways.
Authors: Hongyue Wu, Hangyu Li, Guodong Fan, Haoran Zhu, Shizhan Chen, Zhiyong Feng
Abstract: Federated learning (FL) enables collaborative model training without sharing raw data in edge environments, but is constrained by limited communication bandwidth and heterogeneous client data distributions. Prototype-based FL mitigates this issue by exchanging class-wise feature prototypes instead of full model parameters; however, existing methods still suffer from suboptimal generalization under severe communication constraints. In this paper, we propose RefProtoFL, a communication-efficient FL framework that integrates External-Referenced Prototype Alignment (ERPA) for representation consistency with Adaptive Probabilistic Update Dropping (APUD) for communication efficiency. Specifically, we decompose the model into a private backbone and a lightweight shared adapter, and restrict federated communication to the adapter parameters only. To further reduce uplink cost, APUD performs magnitude-aware Top-K sparsification, transmitting only the most significant adapter updates for server-side aggregation. To address representation inconsistency across heterogeneous clients, ERPA leverages a small server-held public dataset to construct external reference prototypes that serve as shared semantic anchors. For classes covered by public data, clients directly align local representations to public-induced prototypes, whereas for uncovered classes, alignment relies on server-aggregated global reference prototypes via weighted averaging. Extensive experiments on standard benchmarks demonstrate that RefProtoFL attains higher classification accuracy than state-of-the-art prototype-based FL baselines.
Authors: Injin Kong, Hyoungjoon Lee, Yohan Jo
Abstract: Post-training pretrained Autoregressive models (ARMs) into Masked Diffusion models (MDMs) has emerged as a cost-effective strategy to overcome the limitations of sequential generation. However, the internal algorithmic transformations induced by this paradigm shift remain unexplored, leaving it unclear whether post-trained MDMs acquire genuine bidirectional reasoning capabilities or merely repackage autoregressive heuristics. In this work, we address this question by conducting a comparative circuit analysis of ARMs and their MDM counterparts. Our analysis reveals a systematic "mechanism shift" dependent on the structural nature of the task. Structurally, we observe a distinct divergence: while MDMs largely retain autoregressive circuitry for tasks dominated by local causal dependencies, they abandon initialized pathways for global planning tasks, exhibiting distinct rewiring characterized by increased early-layer processing. Semantically, we identify a transition from sharp, localized specialization in ARMs to distributed integration in MDMs. Through these findings, we conclude that diffusion post-training does not merely adapt model parameters but fundamentally reorganizes internal computation to support non-sequential global planning.
Authors: Harold Kiossou, Pierre Schaus, Siegfried Nijssen
Abstract: In recent years, significant progress has been made on algorithms for learning optimal decision trees, primarily in the context of binary features. Extending these methods to continuous features remains substantially more challenging due to the large number of potential splits for each feature. Recently, an elegant exact algorithm was proposed for learning optimal decision trees with continuous features; however, the rapidly increasing computational time limits its practical applicability to shallow depths (typically 3 or 4). It relies on a depth-first search optimization strategy that fully optimizes the left subtree of each split before exploring the corresponding right subtree. While effective in finding optimal solutions given sufficient time, this strategy can lead to poor anytime behavior: when interrupted early, the best-found tree is often highly unbalanced and suboptimal. In such cases, purely greedy methods such as C4.5 may, paradoxically, yield better solutions. To address this limitation, we propose an anytime, yet complete approach leveraging limited discrepancy search, distributing the computational effort more evenly across the entire tree structure, and thus ensuring that a high-quality decision tree is available at any interruption point. Experimental results show that our approach outperforms the existing one in terms of anytime performance.
Authors: Dong Sun, Rahul Nittala, Rebekka Burkholz
Abstract: Despite their practical success, it remains unclear why Mixture of Experts (MoE) models can outperform dense networks beyond sheer parameter scaling. We study an iso-parameter regime where inputs exhibit latent modular structure but are corrupted by feature noise, a proxy for noisy internal activations. We show that sparse expert activation acts as a noise filter: compared to a dense estimator, MoEs achieve lower generalization error under feature noise, improved robustness to perturbations, and faster convergence speed. Empirical results on synthetic data and real-world language tasks corroborate the theoretical insights, demonstrating consistent robustness and efficiency gains from sparse modular computation.
Authors: Ond\v{r}ej Holub (Czech Technical University in Prague), Essi Ryymin (H\"ame University of Applied Sciences), Rodrigo Alves (Czech Technical University in Prague)
Abstract: Designing good reflection questions is pedagogically important but time-consuming and unevenly supported across teachers. This paper introduces a reflection-in-reflection framework for automated generation of reflection questions with large language models (LLMs). Our approach coordinates two role-specialized agents, a Student-Teacher and a Teacher-Educator, that engage in a Socratic multi-turn dialogue to iteratively refine a single question given a teacher-specified topic, key concepts, student level, and optional instructional materials. The Student-Teacher proposes candidate questions with brief rationales, while the Teacher-Educator evaluates them along clarity, depth, relevance, engagement, and conceptual interconnections, responding only with targeted coaching questions or a fixed signal to stop the dialogue. We evaluate the framework in an authentic lower-secondary ICT setting on the topic, using GPT-4o-mini as the backbone model and a stronger GPT- 4-class LLM as an external evaluator in pairwise comparisons of clarity, relevance, depth, and overall quality. First, we study how interaction design and context (dynamic vs.fixed iteration counts; presence or absence of student level and materials) affect question quality. Dynamic stopping combined with contextual information consistently outperforms fixed 5- or 10-step refinement, with very long dialogues prone to drift or over-complication. Second, we show that our two-agent protocol produces questions that are judged substantially more relevant and deeper, and better overall, than a one-shot baseline using the same backbone model.
Authors: Christian Fiedler
Abstract: In supervised learning with distributional inputs in the two-stage sampling setup, relevant to applications like learning-based medical screening or causal learning, the inputs (which are probability distributions) are not accessible in the learning phase, but only samples thereof. This problem is particularly amenable to kernel-based learning methods, where the distributions or samples are first embedded into a Hilbert space, often using kernel mean embeddings (KMEs), and then a standard kernel method like Support Vector Machines (SVMs) is applied, using a kernel defined on the embedding Hilbert space. In this work, we contribute to the theoretical analysis of this latter approach, with a particular focus on classification with distributional inputs using SVMs. We establish a new oracle inequality and derive consistency and learning rate results. Furthermore, for SVMs using the hinge loss and Gaussian kernels, we formulate a novel variant of an established noise assumption from the binary classification literature, under which we can establish learning rates. Finally, some of our technical tools like a new feature space for Gaussian kernels on Hilbert spaces are of independent interest.
Authors: Mohamed Abouras, Catherine M. Elias
Abstract: On and off-ramps are understudied road sections even though they introduce a higher level of variation in highway interactions. Predicting vehicles' behavior in these areas can decrease the impact of uncertainty and increase road safety. In this paper, the difference between this Area of Interest (AoI) and a straight highway section is studied. Multi-layered LSTM architecture to train the AoI model with ExiD drone dataset is utilized. In the process, different prediction horizons and different models' workflow are tested. The results show great promise on horizons up to 4 seconds with prediction accuracy starting from about 76% for the AoI and 94% for the general highway scenarios on the maximum horizon.
Authors: Baojun Che, Yifan Chen, Daniel Zhengyu Huang, Xinying Mao, Weijie Wang
Abstract: Black-box variational inference (BBVI) with Gaussian mixture families offers a flexible approach for approximating complex posterior distributions without requiring gradients of the target density. However, standard numerical optimization methods often suffer from instability and inefficiency. We develop a stable and efficient framework that combines three key components: (1) affine-invariant preconditioning via natural gradient formulations, (2) an exponential integrator that unconditionally preserves the positive definiteness of covariance matrices, and (3) adaptive time stepping to ensure stability and to accommodate distinct warm-up and convergence phases. The proposed approach has natural connections to manifold optimization and mirror descent. For Gaussian posteriors, we prove exponential convergence in the noise-free setting and almost-sure convergence under Monte Carlo estimation, rigorously justifying the necessity of adaptive time stepping. Numerical experiments on multimodal distributions, Neal's multiscale funnel, and a PDE-based Bayesian inverse problem for Darcy flow demonstrate the effectiveness of the proposed method.
Authors: Olaf Yunus Laitinen Imanov, Taner Yilmaz, Derya Umut Kulali
Abstract: We introduce Strategic Doctrine Language Models (sdLM), a learning-system framework for multi-document strategic reasoning with doctrinal consistency constraints and calibrated uncertainty. The approach combines multi-document attention, temporal encoding, and a doctrine-consistency layer to improve long-horizon forecasting and plan plausibility while reducing severe doctrinal violations. We evaluate sdLM using (i) expert-panel scoring of strategic scenarios (N=47), (ii) doctrine consistency on 336 doctrine publications (12,847 statements), and (iii) geopolitical forecasting on 127 historical counterfactuals (1945-2020) across 12-60 month horizons. Across these benchmarks, sdLM achieves higher strategic quality and better calibration than strong general-purpose LLM baselines, and remains competitive with human experts on long-horizon judgments. We further report ablations, scaling trends, and deployment-oriented performance/latency characteristics to clarify which components drive improvements and how they translate to operational settings.
Authors: Keyu Lv, Manyi Zhang, Xiaobo Xia, Jingchen Ni, Shannan Yan, Xianzhi Yu, Lu Hou, Chun Yuan, Haoli Bai
Abstract: Reasoning models excel at complex tasks such as coding and mathematics, yet their inference is often slow and token-inefficient. To improve the inference efficiency, post-training quantization (PTQ) usually comes with the cost of large accuracy drops, especially for reasoning tasks under low-bit settings. In this study, we present a systematic empirical study of quantization-aware training (QAT) for reasoning models. Our key findings include: (1) Knowledge distillation is a robust objective for reasoning models trained via either supervised fine-tuning or reinforcement learning; (2) PTQ provides a strong initialization for QAT, improving accuracy while reducing training cost; (3) Reinforcement learning remains feasible for quantized models given a viable cold start and yields additional gains; and (4) Aligning the PTQ calibration domain with the QAT training domain accelerates convergence and often improves the final accuracy. Finally, we consolidate these findings into an optimized workflow (Reasoning-QAT), and show that it consistently outperforms state-of-the-art PTQ methods across multiple LLM backbones and reasoning datasets. For instance, on Qwen3-0.6B, it surpasses GPTQ by 44.53% on MATH-500 and consistently recovers performance in the 2-bit regime.
Authors: Giorgia Rigamonti, Mirko Paolo Barbato, Davide Marelli, Paolo Napoletano
Abstract: Effective management of Type 1 Diabetes requires continuous glucose monitoring and precise insulin adjustments to prevent hyperglycemia and hypoglycemia. With the growing adoption of wearable glucose monitors and mobile health applications, accurate blood glucose prediction is essential for enhancing automated insulin delivery and decision-support systems. This paper presents a deep learning-based approach for personalized blood glucose prediction, leveraging patient-specific data to improve prediction accuracy and responsiveness in real-world scenarios. Unlike traditional generalized models, our method accounts for individual variability, enabling more effective subject-specific predictions. We compare Leave-One-Subject-Out Cross-Validation with a fine-tuning strategy to evaluate their ability to model patient-specific dynamics. Results show that personalized models significantly improve the prediction of adverse events, enabling more precise and timely interventions in real-world scenarios. To assess the impact of patient-specific data, we conduct experiments comparing a multimodal, patient-specific approach against traditional CGM-only methods. Additionally, we perform an ablation study to investigate model performance with progressively smaller training sets, identifying the minimum data required for effective personalization-an essential consideration for real-world applications where extensive data collection is often challenging. Our findings underscore the potential of adaptive, personalized glucose prediction models for advancing next-generation diabetes management, particularly in wearable and mobile health platforms, enhancing consumer-oriented diabetes care solutions.
Authors: Hang Zhao, Hongru Li, Dongfang Xu, Shenghui Song, Khaled B. Letaief
Abstract: Semantic communication is emerging as a key enabler for distributed edge intelligence due to its capability to convey task-relevant meaning. However, achieving communication-efficient training and robust inference over wireless links remains challenging. This challenge is further exacerbated for multi-modal edge inference (MMEI) by two factors: 1) prohibitive communication overhead for distributed learning over bandwidth-limited wireless links, due to the \emph{multi-modal} nature of the system; and 2) limited robustness under varying channels and noisy multi-modal inputs. In this paper, we propose a three-stage communication-aware distributed learning framework to improve training and inference efficiency while maintaining robustness over wireless channels. In Stage~I, devices perform local multi-modal self-supervised learning to obtain shared and modality-specific encoders without device--server exchange, thereby reducing the communication cost. In Stage~II, distributed fine-tuning with centralized evidential fusion calibrates per-modality uncertainty and reliably aggregates features distorted by noise or channel fading. In Stage~III, an uncertainty-guided feedback mechanism selectively requests additional features for uncertain samples, optimizing the communication--accuracy tradeoff in the distributed setting. Experiments on RGB--depth indoor scene classification show that the proposed framework attains higher accuracy with far fewer training communication rounds and remains robust to modality degradation or channel variation, outperforming existing self-supervised and fully supervised baselines.
Authors: Han Li, Hua Sun
Abstract: Social media increasingly disseminates information through mixed image text posts, but rumors often exploit subtle inconsistencies and forged content, making detection based solely on post content difficult. Deep semantic mismatch rumors, which superficially align images and texts, pose particular challenges and threaten online public opinion. Existing multimodal rumor detection methods improve cross modal modeling but suffer from limited feature extraction, noisy alignment, and inflexible fusion strategies, while ignoring external factual evidence necessary for verifying complex rumors. To address these limitations, we propose a multimodal rumor detection model enhanced with external evidence and forgery features. The model uses a ResNet34 visual encoder, a BERT text encoder, and a forgery feature module extracting frequency-domain traces and compression artifacts via Fourier transformation. BLIP-generated image descriptions bridge image and text semantic spaces. A dual contrastive learning module computes contrastive losses between text image and text description pairs, improving detection of semantic inconsistencies. A gated adaptive feature-scaling fusion mechanism dynamically adjusts multimodal fusion and reduces redundancy. Experiments on Weibo and Twitter datasets demonstrate that our model outperforms mainstream baselines in macro accuracy, recall, and F1 score.
Authors: Harry Mead, Bruno Lacerda, Jakob Foerster, Nick Hawes
Abstract: Unsupervised Environment Design (UED) seeks to automatically generate training curricula for reinforcement learning (RL) agents, with the goal of improving generalisation and zero-shot performance. However, designing effective curricula remains a difficult problem, particularly in settings where small subsets of environment parameterisations result in significant increases in the complexity of the required policy. Current methods struggle with a difficult credit assignment problem and rely on regret approximations that fail to identify challenging levels, both of which are compounded as the size of the environment grows. We propose Dynamic Environment Generation for UED (DEGen) to enable a denser level generator reward signal, reducing the difficulty of credit assignment and allowing for UED to scale to larger environment sizes. We also introduce a new regret approximation, Maximised Negative Advantage (MNA), as a significantly improved metric to optimise for, that better identifies more challenging levels. We show empirically that MNA outperforms current regret approximations and when combined with DEGen, consistently outperforms existing methods, especially as the size of the environment grows. We have made all our code available here: https://github.com/HarryMJMead/Dynamic-Environment-Generation-for-UED.
URLs: https://github.com/HarryMJMead/Dynamic-Environment-Generation-for-UED.
Authors: Mingyue Cheng, Xiaoyu Tao, Huajian Zhang, Qi Liu, Enhong Chen
Abstract: Most existing time series classification methods adopt a discriminative paradigm that maps input sequences directly to one-hot encoded class labels. While effective, this paradigm struggles to incorporate contextual features and fails to capture semantic relationships among classes. To address these limitations, we propose InstructTime, a novel framework that reformulates time series classification as a multimodal generative task. Specifically, continuous numerical sequences, contextual textual features, and task instructions are treated as multimodal inputs, while class labels are generated as textual outputs by tuned language models. To bridge the modality gap, InstructTime introduces a time series discretization module that converts continuous sequences into discrete temporal tokens, together with an alignment projection layer and a generative self-supervised pre-training strategy to enhance cross-modal representation alignment. Building upon this framework, we further propose InstructTime++, which extends InstructTime by incorporating implicit feature modeling to compensate for the limited inductive bias of language models. InstructTime++ leverages specialized toolkits to mine informative implicit patterns from raw time series and contextual inputs, including statistical feature extraction and vision-language-based image captioning, and translates them into textual descriptions for seamless integration. Extensive experiments on multiple benchmark datasets demonstrate the superior performance of InstructTime++.
Authors: Liping Chen, Mujie Liu, Haytham Fayek
Abstract: Modern machine learning systems are increasingly realised as multistage pipelines, yet existing transparency mechanisms typically operate at a model level: they describe what a system is and why it behaves as it does, but not how individual data samples are operationally recorded, tracked, and verified as they traverse the pipeline. This absence of verifiable, sample-level traceability leaves practitioners and users unable to determine whether a specific sample was used, when it was processed, or whether the corresponding records remain intact over time. We introduce FG-Trac, a model-agnostic framework that establishes verifiable, fine-grained sample-level traceability throughout machine learning pipelines. FG-Trac defines an explicit mechanism for capturing and verifying sample lifecycle events across preprocessing and training, computes contribution scores explicitly grounded in training checkpoints, and anchors these traces to tamper-evident cryptographic commitments. The framework integrates without modifying model architectures or training objectives, reconstructing complete and auditable data-usage histories with practical computational overhead. Experiments on a canonical convolutional neural network and a multimodal graph learning pipeline demonstrate that FG-Trac preserves predictive performance while enabling machine learning systems to furnish verifiable evidence of how individual samples were used and propagated during model execution.
Authors: Christos Petridis, Konstantinos Pelechrinis
Abstract: Identifying combinations of players (that is, lineups) in basketball - and other sports - that perform well when they play together is one of the most important tasks in sports analytics. One of the main challenges associated with this task is the frequent substitutions that occur during a game, which results in highly sparse data. In particular, a National Basketball Association (NBA) team will use more than 600 lineups during a season, which translates to an average lineup having seen the court in approximately 25-30 possessions. Inevitably, any statistics that one collects for these lineups are going to be noisy, with low predictive value. Yet, there is no existing work (in the public at least) that addresses this problem. In this work, we propose a regression-based approach that controls for the opposition faced by each lineup, while it also utilizes information about the players making up the lineups. Our experiments show that L-RAPM provides improved predictive power than the currently used baseline, and this improvement increases as the sample size for the lineups gets smaller.
Authors: Michael Feil, Julius Lipp
Abstract: Batch inference workloads for causal transformer models frequently process sequences that share common prefixes, such as system prompts, few-shot examples, or shared queries. Standard inference engines treat each sequence independently, redundantly recomputing identical MLP activations for every copy of the shared prefix. We introduce RadixMLP, a technique that exploits the position-wise nature of MLPs, LayerNorms, linear projections, and embeddings to eliminate this redundancy. RadixMLP dynamically maps batches to a prefix trie, gathering shared segments into a compressed representation for position-wise computation and scattering results back only at attention boundaries. RadixMLP is stateless and operates within a single forward pass. In end-to-end serving benchmarks on MS~MARCO v1.1 with Qwen3 models (0.6B to 8B parameters), RadixMLP achieves 1.44-1.59$\times$ speedups in realistic reranking workloads, with up to $5\times$ speedups on synthetic benchmarks with longer shared prefixes. Our code is available at https://github.com/michaelfeil/radix-mlp.
Authors: Jannis Becktepe, Aleksandra Franz, Nils Thuerey, Sebastian Peitz
Abstract: Reinforcement learning (RL) has shown promising results in active flow control (AFC), yet progress in the field remains difficult to assess as existing studies rely on heterogeneous observation and actuation schemes, numerical setups, and evaluation protocols. Current AFC benchmarks attempt to address these issues but heavily rely on external computational fluid dynamics (CFD) solvers, are not fully differentiable, and provide limited 3D and multi-agent support. To overcome these limitations, we introduce FluidGym, the first standalone, fully differentiable benchmark suite for RL in AFC. Built entirely in PyTorch on top of the GPU-accelerated PICT solver, FluidGym runs in a single Python stack, requires no external CFD software, and provides standardized evaluation protocols. We present baseline results with PPO and SAC and release all environments, datasets, and trained models as public resources. FluidGym enables systematic comparison of control methods, establishes a scalable foundation for future research in learning-based flow control, and is available at https://github.com/safe-autonomous-systems/fluidgym.
Authors: Adam Rokah, Daniel Veress, Caleb Caulk, Sourav Sharan
Abstract: Mixture-of-Experts (MoE) architectures enable conditional computation by routing inputs to multiple expert subnetworks and are often motivated as a mechanism for scaling large language models. In this project, we instead study MoE behavior in an image classification setting, focusing on predictive performance, expert utilization, and generalization. We compare dense, SoftMoE, and SparseMoE classifier heads on the CIFAR10 dataset under comparable model capacity. Both MoE variants achieve slightly higher validation accuracy than the dense baseline while maintaining balanced expert utilization through regularization, avoiding expert collapse. To analyze generalization, we compute Hessian-based sharpness metrics at convergence, including the largest eigenvalue and trace of the loss Hessian, evaluated on both training and test data. We find that SoftMoE exhibits higher sharpness by these metrics, while Dense and SparseMoE lie in a similar curvature regime, despite all models achieving comparable generalization performance. Complementary loss surface perturbation analyses reveal qualitative differences in non-local behavior under finite parameter perturbations between dense and MoE models, which help contextualize curvature-based measurements without directly explaining validation accuracy. We further evaluate empirical inference efficiency and show that naively implemented conditional routing does not yield inference speedups on modern hardware at this scale, highlighting the gap between theoretical and realized efficiency in sparse MoE models.
Authors: Dirk Tasche
Abstract: Factorizable joint shift (FJS) was proposed as a type of distribution shift (or dataset shift) that comprises both covariate and label shift. Recently, it has been observed that FJS actually arises from consecutive label and covariate (or vice versa) shifts. Research into FJS so far has been confined to the case of categorical label spaces. We propose a framework for analysing distribution shift in the case of general label spaces, thus covering both classification and regression models. Based on the framework, we generalise existing results on FJS to general label spaces and propose a related extension of the expectation maximisation (EM) algorithm for class prior probabilities. We also take a fresh look at generalized label shift (GLS) in the case of general label spaces.
Authors: Mertcan Daysalilar, Fuat Uyguroglu, Gabriel Nicolosi, Adam Meyers
Abstract: The electric vehicle routing problem with time windows (EVRPTW) is a complex optimization problem in sustainable logistics, where routing decisions must minimize total travel distance, fleet size, and battery usage while satisfying strict customer time constraints. Although deep reinforcement learning (DRL) has shown great potential as an alternative to classical heuristics and exact solvers, existing DRL models often struggle to maintain training stability-failing to converge or generalize when constraints are dense. In this study, we propose a curriculum-based deep reinforcement learning (CB-DRL) framework designed to resolve this instability. The framework utilizes a structured three-phase curriculum that gradually increases problem complexity: the agent first learns distance and fleet optimization (Phase A), then battery management (Phase B), and finally the full EVRPTW (Phase C). To ensure stable learning across phases, the framework employs a modified proximal policy optimization algorithm with phase-specific hyperparameters, value and advantage clipping, and adaptive learning-rate scheduling. The policy network is built upon a heterogeneous graph attention encoder enhanced by global-local attention and feature-wise linear modulation. This specialized architecture explicitly captures the distinct properties of depots, customers, and charging stations. Trained exclusively on small instances with N=10 customers, the model demonstrates robust generalization to unseen instances ranging from N=5 to N=100, significantly outperforming standard baselines on medium-scale problems. Experimental results confirm that this curriculum-guided approach achieves high feasibility rates and competitive solution quality on out-of-distribution instances where standard DRL baselines fail, effectively bridging the gap between neural speed and operational reliability.
Authors: Oliver Wei{\ss}l, Vincenzo Riccio, Severin Kacianka, Andrea Stocco
Abstract: The increasing deployment of deep learning systems requires systematic evaluation of their reliability in real-world scenarios. Traditional gradient-based adversarial attacks introduce small perturbations that rarely correspond to realistic failures and mainly assess robustness rather than functional behavior. Generative test generation methods offer an alternative but are often limited to simple datasets or constrained input domains. Although diffusion models enable high-fidelity image synthesis, their computational cost and limited controllability restrict their applicability to large-scale testing. We present HyNeA, a generative testing method that enables direct and efficient control over diffusion-based generation. HyNeA provides dataset-free controllability through hypernetworks, allowing targeted manipulation of the generative process without relying on architecture-specific conditioning mechanisms or dataset-driven adaptations such as fine-tuning. HyNeA employs a distinct training strategy that supports instance-level tuning to identify failure-inducing test cases without requiring datasets that explicitly contain examples of similar failures. This approach enables the targeted generation of realistic failure cases at substantially lower computational cost than search-based methods. Experimental results show that HyNeA improves controllability and test diversity compared to existing generative test generators and generalizes to domains where failure-labeled training data is unavailable.
Authors: Chenyu Liu, Haige Li, Luca Rossi
Abstract: Graph Neural Networks (GNNs) are neural networks that aim to process graph data, capturing the relationships and interactions between nodes using the message-passing mechanism. GNN quantization has emerged as a promising approach for reducing model size and accelerating inference in resource-constrained environments. Compared to quantization in LLMs, quantizing graph features is more emphasized in GNNs. Inspired by the above, we propose to leverage prompt learning, which manipulates the input data, to improve the performance of quantization-aware training (QAT) for GNNs. To mitigate the issue that prompting the node features alone can only make part of the quantized aggregation result optimal, we introduce Low-Rank Aggregation Prompting (LoRAP), which injects lightweight, input-dependent prompts into each aggregated feature to optimize the results of quantized aggregations. Extensive evaluations on 4 leading QAT frameworks over 9 graph datasets demonstrate that LoRAP consistently enhances the performance of low-bit quantized GNNs while introducing a minimal computational overhead.
Authors: Oleg Shchendrigin, Egor Cherepanov, Alexey K. Kovalev, Aleksandr I. Panov
Abstract: Effective decision-making in the real world depends on memory that is both stable and adaptive: environments change over time, and agents must retain relevant information over long horizons while also updating or overwriting outdated content when circumstances shift. Existing Reinforcement Learning (RL) benchmarks and memory-augmented agents focus primarily on retention, leaving the equally critical ability of memory rewriting largely unexplored. To address this gap, we introduce a benchmark that explicitly tests continual memory updating under partial observability, i.e. the natural setting where an agent must rely on memory rather than current observations, and use it to compare recurrent, transformer-based, and structured memory architectures. Our experiments reveal that classic recurrent models, despite their simplicity, demonstrate greater flexibility and robustness in memory rewriting tasks than modern structured memories, which succeed only under narrow conditions, and transformer-based agents, which often fail beyond trivial retention cases. These findings expose a fundamental limitation of current approaches and emphasize the necessity of memory mechanisms that balance stable retention with adaptive updating. Our work highlights this overlooked challenge, introduces benchmarks to evaluate it, and offers insights for designing future RL agents with explicit and trainable forgetting mechanisms. Code: https://quartz-admirer.github.io/Memory-Rewriting/
Authors: Johannes Meuer, Maximilian Witte, \'Eti\'enne Pl\'esiat, Thomas Ludwig, Christopher Kadow
Abstract: Kilometer-scale Earth system models are essential for capturing local climate change. However, these models are computationally expensive and produce petabyte-scale outputs, which limits their utility for applications such as probabilistic risk assessment. Here, we present the Field-Space Autoencoder, a scalable climate emulation framework based on a spherical compression model that overcomes these challenges. By utilizing Field-Space Attention, the model efficiently operates on native climate model output and therefore avoids geometric distortions caused by forcing spherical data onto Euclidean grids. This approach preserves physical structures significantly better than convolutional baselines. By producing a structured compressed field, it serves as a good baseline for downstream generative emulation. In addition, the model can perform zero-shot super-resolution that maps low-resolution large ensembles and scarce high-resolution data into a shared representation. We train a generative diffusion model on these compressed fields. The model can simultaneously learn internal variability from abundant low-resolution data and fine-scale physics from sparse high-resolution data. Our work bridges the gap between the high volume of low-resolution ensemble statistics and the scarcity of high-resolution physical detail.
Authors: Anmol Goel, Alan Ritter, Iryna Gurevych
Abstract: We expose a critical limitation in current approaches to machine unlearning in language models: despite the apparent success of unlearning algorithms, information about the forgotten data remains linearly decodable from internal representations. To systematically assess this discrepancy, we introduce an interpretable, information-theoretic framework for auditing unlearning using Partial Information Decomposition (PID). By comparing model representations before and after unlearning, we decompose the mutual information with the forgotten data into distinct components, formalizing the notions of unlearned and residual knowledge. Our analysis reveals that redundant information, shared across both models, constitutes residual knowledge that persists post-unlearning and correlates with susceptibility to known adversarial reconstruction attacks. Leveraging these insights, we propose a representation-based risk score that can guide abstention on sensitive inputs at inference time, providing a practical mechanism to mitigate privacy leakage. Our work introduces a principled, representation-level audit for unlearning, offering theoretical insight and actionable tools for safer deployment of language models.
Authors: Haonan Yuan, Qingyun Sun, Jiacheng Tao, Xingcheng Fu, Jianxin Li
Abstract: Graph Foundation Models (GFMs) have emerged as a frontier in graph learning, which are expected to deliver transferable representations across diverse tasks. However, GFMs remain constrained by in-memory bottlenecks: they attempt to encode knowledge into model parameters, which limits semantic capacity, introduces heavy lossy compression with conflicts, and entangles graph representation with the knowledge in ways that hinder efficient adaptation, undermining scalability and interpretability. In this work,we propose RAG-GFM, a Retrieval-Augmented Generation aided Graph Foundation Model that offloads knowledge from parameters and complements parameterized learning. To externalize graph knowledge, we build a dual-modal unified retrieval module, where a semantic store from prefix-structured text and a structural store from centrality-based motif. To preserve heterogeneous information, we design a dual-view alignment objective that contrasts both modalities to capture both content and relational patterns. To enable efficient downstream adaptation, we perform in-context augmentation to enrich supporting instances with retrieved texts and motifs as contextual evidence. Extensive experiments on five benchmark graph datasets demonstrate that RAG-GFM consistently outperforms 13 state-of-the-art baselines in both cross-domain node and graph classification, achieving superior effectiveness and efficiency.
Authors: Bostan Khan, Masoud Daneshtalab
Abstract: Federated Neural Architecture Search (FedNAS) aims to automate model design for privacy-preserving Federated Learning (FL) but currently faces two critical bottlenecks: unguided supernet training that yields suboptimal models, and costly multi-hour pipelines for post-training subnet discovery. We introduce DeepFedNAS, a novel, two-phase framework underpinned by a principled, multi-objective fitness function that synthesizes mathematical network design with architectural heuristics. Enabled by a re-engineered supernet, DeepFedNAS introduces Federated Pareto Optimal Supernet Training, which leverages a pre-computed Pareto-optimal cache of high-fitness architectures as an intelligent curriculum to optimize shared supernet weights. Subsequently, its Predictor-Free Search Method eliminates the need for costly accuracy surrogates by utilizing this fitness function as a direct, zero-cost proxy for accuracy, enabling on-demand subnet discovery in mere seconds. DeepFedNAS achieves state-of-the-art accuracy (e.g., up to 1.21% absolute improvement on CIFAR-100), superior parameter and communication efficiency, and a substantial ~61x speedup in total post-training search pipeline time. By reducing the pipeline from over 20 hours to approximately 20 minutes (including initial cache generation) and enabling 20-second individual subnet searches, DeepFedNAS makes hardware-aware FL deployments instantaneous and practical. The complete source code and experimental scripts are available at: https://github.com/bostankhan6/DeepFedNAS
Authors: Tianshi Xu, Yuteng Chen, Meng Li
Abstract: Agentic Reinforcement Learning (RL) has empowered Large Language Models (LLMs) to utilize tools like Python interpreters for complex problem-solving. However, for parameter-constrained models (e.g., 4B--7B), the exploration phase is often plagued by frequent execution failures, creating noisy trajectories that hinder policy optimization. Under standard outcome-based reward settings, this noise leads to a critical credit assignment issue, where erroneous actions are inadvertently reinforced alongside successful outcomes. Existing mitigations face a dilemma: dense rewards often trigger reward hacking, while supersampling incurs prohibitive computational costs. To address these challenges, we propose CLEANER. Distinct from external filtering methods, CLEANER exploits the model's intrinsic self-correction capabilities to eliminate error-contaminated context directly during data collection. At its core, the Similarity-Aware Adaptive Rollback (SAAR) mechanism autonomously constructs clean, purified trajectories by retrospectively replacing failures with successful self-corrections. Based on semantic similarity, SAAR adaptively regulates replacement granularity from shallow execution repairs to deep reasoning substitutions. By training on these self-purified paths, the model internalizes correct reasoning patterns rather than error-recovery loops. Empirical results on AIME24/25, GPQA, and LiveCodeBench show average accuracy gains of 6%, 3%, and 5% over baselines. Notably, CLEANER matches state-of-the-art performance using only one-third of the training steps, highlighting trajectory purification as a scalable solution for efficient agentic RL. Our models and code are available at GitHub
Authors: Yuval Ran-Milo, Yotam Alexander, Shahar Mendel, Nadav Cohen
Abstract: Transformers trained via Reinforcement Learning (RL) with outcome-based supervision can spontaneously develop the ability to generate intermediate reasoning steps (Chain-of-Thought). Yet the mechanism by which sparse rewards drive gradient descent to discover such systematic reasoning remains poorly understood. We address this by analyzing the gradient flow dynamics of single-layer Transformers on a synthetic graph traversal task that cannot be solved without Chain-of-Thought (CoT) but admits a simple iterative solution. We prove that despite training solely on final-answer correctness, gradient flow drives the model to converge to a structured, interpretable algorithm that iteratively traverses the graph vertex-by-vertex. We characterize the distributional properties required for this emergence, identifying the critical role of "simple examples": instances requiring fewer reasoning steps. When the training distribution places sufficient mass on these simpler instances, the model learns a generalizable traversal strategy that extrapolates to longer chains; when this mass vanishes, gradient-based learning becomes infeasible. We corroborate our theoretical results through experiments on synthetic data and with real-world language models on mathematical reasoning tasks, validating that our theoretical findings carry over to practical settings.
Authors: Dhrubo Saha
Abstract: Training deep computer vision models requires manual oversight or hyperparameter tuning of the learning rate (LR) schedule. While existing adaptive optimizers schedule the LR automatically, they suffer from computational and memory overhead, incompatibility with regularization, and suboptimal LR choices. In this work, we introduce the ZENITH (Zero-overhead Evolution using Norm-Informed Training History) optimizer, which adapts the LR using the temporal evolution of the gradient norm. Image classification experiments spanning 6 CNN architectures and 6 benchmarks demonstrate that ZENITH achieves higher test accuracy in lower wall-clock time than baselines. It also yielded superior mAP in object detection, keypoint detection, and instance segmentation on MS COCO using the R-CNN family of models. Furthermore, its compatibility with regularization enables even better generalization.
Authors: Garrett G. Wen, Buxin Su, Natalie Collina, Zhun Deng, Weijie Su
Abstract: Machine learning and artificial intelligence conferences such as NeurIPS and ICML now regularly receive tens of thousands of submissions, posing significant challenges to maintaining the quality and consistency of the peer review process. This challenge is particularly acute for best paper awards, which are an important part of the peer review process, yet whose selection has increasingly become a subject of debate in recent years. In this paper, we introduce an author-assisted mechanism to facilitate the selection of best paper awards. Our method employs the Isotonic Mechanism for eliciting authors' assessments of their own submissions in the form of a ranking, which is subsequently utilized to adjust the raw review scores for optimal estimation of the submissions' ground-truth quality. We demonstrate that authors are incentivized to report truthfully when their utility is a convex additive function of the adjusted scores, and we validate this convexity assumption for best paper awards using publicly accessible review data of ICLR from 2019 to 2023 and NeurIPS from 2021 to 2023. Crucially, in the special case where an author has a single quota -- that is, may nominate only one paper -- we prove that truthfulness holds even when the utility function is merely nondecreasing and additive. This finding represents a substantial relaxation of the assumptions required in prior work. For practical implementation, we extend our mechanism to accommodate the common scenario of overlapping authorship. Finally, simulation results demonstrate that our mechanism significantly improves the quality of papers selected for awards.
Authors: Christoph Bartmann, Johannes Schimunek, Mykyta Ielanskyi, Philipp Seidl, G\"unter Klambauer, Sohvi Luukkonen
Abstract: A molecule's properties are fundamentally determined by its composition and structure encoded in its molecular graph. Thus, reasoning about molecular properties requires the ability to parse and understand the molecular graph. Large Language Models (LLMs) are increasingly applied to chemistry, tackling tasks such as molecular name conversion, captioning, text-guided generation, and property or reaction prediction. Most existing benchmarks emphasize general chemical knowledge, rely on literature or surrogate labels that risk leakage or bias, or reduce evaluation to multiple-choice questions. We introduce MolecularIQ, a molecular structure reasoning benchmark focused exclusively on symbolically verifiable tasks. MolecularIQ enables fine-grained evaluation of reasoning over molecular graphs and reveals capability patterns that localize model failures to specific tasks and molecular structures. This provides actionable insights into the strengths and limitations of current chemistry LLMs and guides the development of models that reason faithfully over molecular structure.
Authors: Ziwen Zhong, Zhitao Shu, Yue Zhao
Abstract: Emotion recognition is a fundamental component of next-generation human-computer interaction (HCI), enabling machines to perceive, understand, and respond to users' affective states. However, existing systems often rely on single-modality analysis such as facial expressions, speech tone, or textual sentiment, resulting in limited robustness and poor generalization in real-world environments. To address these challenges, this study proposes a Cloud-Based Cross-Modal Transformer (CMT) framework for multimodal emotion recognition and adaptive human-computer interaction. The proposed model integrates visual, auditory, and textual signals using pretrained encoders (Vision Transformer, Wav2Vec2, and BERT) and employs a cross-modal attention mechanism to capture complex interdependencies among heterogeneous features. By leveraging cloud computing infrastructure with distributed training on Kubernetes and TensorFlow Serving, the system enables scalable, low-latency emotion recognition for large-scale user interactions. Experiments conducted on benchmark datasets including IEMOCAP, MELD, and AffectNet demonstrate that the CMT achieves state-of-the-art performance, improving the F1-score by 3.0 percent and reducing cross-entropy loss by 12.9 percent compared to strong multimodal baselines. Additionally, cloud deployment evaluations show an average response latency of 128 ms, representing a 35 percent reduction compared with conventional transformer-based fusion systems. These results confirm that the proposed framework enables efficient, real-time emotion recognition and adaptive feedback in applications such as intelligent customer service, virtual tutoring systems, and affective computing interfaces, marking an important step toward cloud-native affective computing and emotionally intelligent interactive systems.
Authors: Xiaoxuan Yang, Peilin Chen, Tergel Molom-Ochir, Yiran Chen
Abstract: Transformers have become central to natural language processing and large language models, but their deployment at scale faces three major challenges. First, the attention mechanism requires massive matrix multiplications and frequent movement of intermediate results between memory and compute units, leading to high latency and energy costs. Second, in long-context inference, the key-value cache (KV cache) can grow unpredictably and even surpass the model's weight size, creating severe memory and bandwidth bottlenecks. Third, the quadratic complexity of attention with respect to sequence length amplifies both data movement and compute overhead, making large-scale inference inefficient. To address these issues, this work introduces processing-in-memory solutions that restructure attention and feed-forward computation to minimize off-chip data transfers, dynamically compress and prune the KV cache to manage memory growth, and reinterpret attention as an associative memory operation to reduce complexity and hardware footprint. Moreover, we evaluate our processing-in-memory design against state-of-the-art accelerators and general-purpose GPUs, demonstrating significant improvements in energy efficiency and latency. Together, these approaches address computation overhead, memory scalability, and attention complexity, further enabling efficient, end-to-end acceleration of Transformer models.
Authors: Taoliang Tan, Chengwei Ma, Zhen Tian, Zhao Lin, Dongdong Li, Si Shi
Abstract: The intelligent review of power grid engineering design drawings is crucial for power system safety. However, current automated systems struggle with ultra-high-resolution drawings due to high computational demands, information loss, and a lack of holistic semantic understanding for design error identification. This paper proposes a novel three-stage framework for intelligent power grid drawing review, driven by pre-trained Multimodal Large Language Models (MLLMs) through advanced prompt engineering. Mimicking the human expert review process, the first stage leverages an MLLM for global semantic understanding to intelligently propose domain-specific semantic regions from a low-resolution overview. The second stage then performs high-resolution, fine-grained recognition within these proposed regions, acquiring detailed information with associated confidence scores. In the final stage, a comprehensive decision-making module integrates these confidence-aware results to accurately diagnose design errors and provide a reliability assessment. Preliminary results on real-world power grid drawings demonstrate our approach significantly enhances MLLM's ability to grasp macroscopic semantic information and pinpoint design errors, showing improved defect discovery accuracy and greater reliability in review judgments compared to traditional passive MLLM inference. This research offers a novel, prompt-driven paradigm for intelligent and reliable power grid drawing review.
Authors: Hongxiao Li, Chenxi Wang, Fanda Fan, Zihan Wang, Wanling Gao, Lei Wang, Jianfeng Zhan
Abstract: Evaluation is the foundation of empirical science, yet the evaluation of evaluation itself -- so-called meta-evaluation -- remains strikingly underdeveloped. While methods such as observational studies, design of experiments (DoE), and randomized controlled trials (RCTs) have shaped modern scientific practice, there has been little systematic inquiry into their comparative validity and utility across domains. Here we introduce a formal framework for meta-evaluation by defining the evaluation space, its structured representation, and a benchmark we call AxiaBench. AxiaBench enables the first large-scale, quantitative comparison of ten widely used evaluation methods across eight representative application domains. Our analysis reveals a fundamental limitation: no existing method simultaneously achieves accuracy and efficiency across diverse scenarios, with DoE and observational designs in particular showing significant deviations from real-world ground truth. We further evaluate a unified method of entire-space stratified sampling from previous evaluatology research, and the results report that it consistently outperforms prior approaches across all tested domains. These results establish meta-evaluation as a scientific object in its own right and provide both a conceptual foundation and a pragmatic tool set for advancing trustworthy evaluation in computational and experimental research.
Authors: Wentao Jiang, Jingxin Wang, Zhang Hu, Zhengyuan Shi, Chengyu Ma, Qiang Xu, Weikang Qian, Zhufei Chu
Abstract: Traditional technology mapping suffers from systemic inaccuracies in delay estimation due to its reliance on abstract, technology-agnostic delay models that fail to capture the nuanced timing behavior behavior of real post-mapping circuits. To address this fundamental limitation, we introduce GPA(graph neural network (GNN)-based Path-Aware multi-view circuit learning), a novel GNN framework that learns precise, data-driven delay predictions by synergistically fusing three complementary views of circuit structure: And-Inverter Graphs (AIGs)-based functional encoding, post-mapping technology emphasizes critical timing paths. Trained exclusively on real cell delays extracted from critical paths of industrial-grade post-mapping netlists, GPA learns to classify cut delays with unprecedented accuracy, directly informing smarter mapping decisions. Evaluated on the 19 EPFL combinational benchmarks, GPA achieves 19.9%, 2.1% and 4.1% average delay reduction over the conventional heuristics methods (techmap, MCH) and the prior state-of-the-art ML-based approach SLAP, respectively-without compromising area efficiency.
Authors: Richard Hohensinner, Belgin Mutlu, Inti Gabriel Mendoza Estrada, Matej Vukovic, Simone Kopeinik, Roman Kern
Abstract: Large language models (LLMs) are deployed at scale, yet their training data life cycle remains opaque. This survey synthesizes research from the past ten years on three tightly coupled axes: (1) data provenance, (2) transparency, and (3) traceability, and three supporting pillars: (4) bias \& uncertainty, (5) data privacy, and (6) tools and techniques that operationalize them. A central contribution is a proposed taxonomy defining the field's domains and listing corresponding artifacts. Through analysis of 95 publications, this work identifies key methodologies concerning data generation, watermarking, bias measurement, data curation, data privacy, and the inherent trade-off between transparency and opacity.
Authors: Mengyu Sun, Ziyuan Yang, Andrew Beng Jin Teoh, Junxu Liu, Haibo Hu, Yi Zhang
Abstract: Concept erasure aims to suppress sensitive content in diffusion models, but recent studies show that erased concepts can still be reawakened, revealing vulnerabilities in erasure methods. Existing reawakening methods mainly rely on prompt-level optimization to manipulate sampling trajectories, neglecting other generative factors, which limits a comprehensive understanding of the underlying dynamics. In this paper, we model the generation process as an implicit function to enable a comprehensive theoretical analysis of multiple factors, including text conditions, model parameters, and latent states. We theoretically show that perturbing each factor can reawaken erased concepts. Building on this insight, we propose a novel concept reawakening method: Latent space Unblocking for concept REawakening (LURE), which reawakens erased concepts by reconstructing the latent space and guiding the sampling trajectory. Specifically, our semantic re-binding mechanism reconstructs the latent space by aligning denoising predictions with target distributions to reestablish severed text-visual associations. However, in multi-concept scenarios, naive reconstruction can cause gradient conflicts and feature entanglement. To address this, we introduce Gradient Field Orthogonalization, which enforces feature orthogonality to prevent mutual interference. Additionally, our Latent Semantic Identification-Guided Sampling (LSIS) ensures stability of the reawakening process via posterior density verification. Extensive experiments demonstrate that LURE enables simultaneous, high-fidelity reawakening of multiple erased concepts across diverse erasure tasks and methods.
Authors: Yiyang Lu, Jinwen He, Yue Zhao, Kai Chen, Ruigang Liang
Abstract: Large Language Models (LLMs) are widely integrated into interactive systems such as dialogue agents and task-oriented assistants. This growing ecosystem also raises supply-chain risks, where adversaries can distribute poisoned models that degrade downstream reliability and user trust. Existing backdoor attacks and defenses are largely prompt-centric, focusing on user-visible triggers while overlooking structural signals in multi-turn conversations. We propose Turn-based Structural Trigger (TST), a backdoor attack that activates from dialogue structure, using the turn index as the trigger and remaining independent of user inputs. Across four widely used open-source LLM models, TST achieves an average attack success rate (ASR) of 99.52% with minimal utility degradation, and remains effective under five representative defenses with an average ASR of 98.04%. The attack also generalizes well across instruction datasets, maintaining an average ASR of 99.19%. Our results suggest that dialogue structure constitutes an important and under-studied attack surface for multi-turn LLM systems, motivating structure-aware auditing and mitigation in practice.
Authors: Sunghyun Kim, Seokwoo Yun, Youngseo Yun, Youngrak Lee, Sangsoo Lim
Abstract: Motivation: Developing high-performing bioinformatics models typically requires repeated cycles of hypothesis formulation, architectural redesign, and empirical validation, making progress slow, labor-intensive, and difficult to reproduce. Although recent LLM-based assistants can automate isolated steps, they lack performance-grounded reasoning and stability-aware mechanisms required for reliable, iterative model improvement in bioinformatics workflows. Results: We introduce MARBLE, an execution-stable autonomous model refinement framework for bioinformatics models. MARBLE couples literature-aware reference selection with structured, debate-driven architectural reasoning among role-specialized agents, followed by autonomous execution, evaluation, and memory updates explicitly grounded in empirical performance. Across spatial transcriptomics domain segmentation, drug-target interaction prediction, and drug response prediction, MARBLE consistently achieves sustained performance improvements over strong baselines across multiple refinement cycles, while maintaining high execution robustness and low regression rates. Framework-level analyses demonstrate that structured debate, balanced evidence selection, and performance-grounded memory are critical for stable, repeatable model evolution, rather than single-run or brittle gains. Availability: Source code, data and Supplementary Information are available at https://github.com/PRISM-DGU/MARBLE.
Authors: Satvik Mishra, Roberto Trotta, Matteo Viel
Abstract: We demonstrate the capabilities of probabilistic diffusion models to reduce dramatically the computational cost of expensive hydrodynamical simulations to study the relationship between observable baryonic cosmological probes and dark matter at field level and well into the non-linear regime. We introduce a novel technique, Cosmo-FOLD (Cosmological Fields via Overlap Latent Diffusion) to rapidly generate accurate and arbitrarily large cosmological and astrophysical 3-dimensional fields, conditioned on a given input field. We are able to generate TNG300-2 dark matter density and gas temperature fields from a model trained only on ~1% of the volume (a process we refer to as `upscaling'), reproducing both large scale coherent dark matter filaments and power spectra to within 10% for wavenumbers k <= 5 h Mpc^-1. These results are obtained within a small fraction of the original simulation cost and produced on a single GPU. Beyond one and two points statistics, the bispectrum is also faithfully reproduced through the inclusion of positional encodings. Finally, we demonstrate Cosmo-FOLD's generalisation capabilities by upscaling a CAMELS volume of 25 (Mpc h^-1)^3 to a full TNG300-2 volume of 205 (Mpc h^-1)^3$ with no fine-tuning. Cosmo-FOLD opens the door to full field-level simulation-based inference on cosmological scale.
Authors: Peter Potaptchik, Adhi Saravanan, Abbas Mammadov, Alvaro Prat, Michael S. Albergo, Yee Whye Teh
Abstract: Controlling generative models is computationally expensive. This is because optimal alignment with a reward function--whether via inference-time steering or fine-tuning--requires estimating the value function. This task demands access to the conditional posterior $p_{1|t}(x_1|x_t)$, the distribution of clean data $x_1$ consistent with an intermediate state $x_t$, a requirement that typically compels methods to resort to costly trajectory simulations. To address this bottleneck, we introduce Meta Flow Maps (MFMs), a framework extending consistency models and flow maps into the stochastic regime. MFMs are trained to perform stochastic one-step posterior sampling, generating arbitrarily many i.i.d. draws of clean data $x_1$ from any intermediate state. Crucially, these samples provide a differentiable reparametrization that unlocks efficient value function estimation. We leverage this capability to solve bottlenecks in both paradigms: enabling inference-time steering without inner rollouts, and facilitating unbiased, off-policy fine-tuning to general rewards. Empirically, our single-particle steered-MFM sampler outperforms a Best-of-1000 baseline on ImageNet across multiple rewards at a fraction of the compute.
Authors: Hsin-Yi Lin, Huan-Hsin Tseng, Samuel Yen-Chi Chen, Shinjae Yoo
Abstract: Super-resolution (SR) seeks to reconstruct high-resolution (HR) data from low-resolution (LR) observations. Classical deep learning methods have advanced SR substantially, but require increasingly deeper networks, large datasets, and heavy computation to capture fine-grained correlations. In this work, we present the \emph{first study} to investigate quantum circuits for SR. We propose a framework based on Variational Quantum Circuits (VQCs) with \emph{Adaptive Non-Local Observable} (ANO) measurements. Unlike conventional VQCs with fixed Pauli readouts, ANO introduces trainable multi-qubit Hermitian observables, allowing the measurement process to adapt during training. This design leverages the high-dimensional Hilbert space of quantum systems and the representational structure provided by entanglement and superposition. Experiments demonstrate that ANO-VQCs achieve up to five-fold higher resolution with a relatively small model size, suggesting a promising new direction at the intersection of quantum machine learning and super-resolution.
Authors: Danial Sadrian Zadeh, Otman A. Basir, Behzad Moshiri
Abstract: Traffic scene understanding is essential for enabling autonomous vehicles to accurately perceive and interpret their environment, thereby ensuring safe navigation. This paper presents a novel framework that transforms a single frontal-view camera image into a concise natural language description, effectively capturing spatial layouts, semantic relationships, and driving-relevant cues. The proposed model leverages a hybrid attention mechanism to enhance spatial and semantic feature extraction and integrates these features to generate contextually rich and detailed scene descriptions. To address the limited availability of specialized datasets in this domain, a new dataset derived from the BDD100K dataset has been developed, with comprehensive guidelines provided for its construction. Furthermore, the study offers an in-depth discussion of relevant evaluation metrics, identifying the most appropriate measures for this task. Extensive quantitative evaluations using metrics such as CIDEr and SPICE, complemented by human judgment assessments, demonstrate that the proposed model achieves strong performance and effectively fulfills its intended objectives on the newly developed dataset.
Authors: Saeed Khaki, Ashudeep Singh, Nima Safaei, Kamal Ginotra
Abstract: Vision-language models (VLMs) lag behind text-only language models on mathematical reasoning when the same problems are presented as images rather than text. We empirically characterize this as a modality gap: the same question in text form yields markedly higher accuracy than its visually typeset counterpart, due to compounded failures in reading dense formulas, layout, and mixed symbolic-diagrammatic context. First, we introduce VisTIRA (Vision and Tool-Integrated Reasoning Agent), a tool-integrated reasoning framework that enables structured problem solving by iteratively decomposing a given math problem (as an image) into natural language rationales and executable Python steps to determine the final answer. Second, we build a framework to measure and improve visual math reasoning: a LaTeX-based pipeline that converts chain-of-thought math corpora (e.g., NuminaMath) into challenging image counterparts, and a large set of synthetic tool-use trajectories derived from a real-world, homework-style image dataset (called SnapAsk) for fine-tuning VLMs. Our experiments show that tool-integrated supervision improves image-based reasoning, and OCR grounding can further narrow the gap for smaller models, although its benefit diminishes at scale. These findings highlight that modality gap severity inversely correlates with model size, and that structured reasoning and OCR-based grounding are complementary strategies for advancing visual mathematical reasoning.
Authors: Valerio Belcamino, Nicholas Attolino, Alessio Capitanelli, Fulvio Mastrogiovanni
Abstract: Recent work shows that fine-tuned Large Language Models (LLMs) can achieve high valid plan rates on PDDL planning tasks. However, it remains unclear whether this reflects transferable planning competence or domain-specific memorization. In this work, we fine-tune a 1.7B-parameter LLM on 40,000 domain-problem-plan tuples from 10 IPC 2023 domains, and evaluate both in-domain and cross-domain generalization. While the model reaches 82.9% valid plan rate in in-domain conditions, it achieves 0% on two unseen domains. To analyze this failure, we introduce three diagnostic interventions, namely (i) instance-wise symbol anonymization, (ii) compact plan serialization, and (iii) verifier-reward fine-tuning using the VAL validator as a success-focused reinforcement signal. Symbol anonymization and compact serialization cause significant performance drops despite preserving plan semantics, thus revealing strong sensitivity to surface representations. Verifier-reward fine-tuning reaches performance saturation in half the supervised training epochs, but does not improve cross-domain generalization. For the explored configurations, in-domain performance plateaus around 80%, while cross-domain performance collapses, suggesting that our fine-tuned model relies heavily on domain-specific patterns rather than transferable planning competence in this setting. Our results highlight a persistent generalization gap in LLM-based planning and provide diagnostic tools for studying its causes.
Authors: Hunter Heidenreich, Ben Elliott, Olivia Dinica, Yosheb Getachew
Abstract: GutenOCR is a family of grounded OCR front-ends obtained by fine-tuning Qwen2.5-VL-3B and Qwen2.5-VL-7B. The resulting single-checkpoint vision-language models expose reading, detection, and grounding through a unified, prompt-based interface. Trained on business documents, scientific articles, and synthetic grounding data, the models support full-page and localized reading with line- and paragraph-level bounding boxes and conditional ``where is x?'' queries. We introduce a grounded OCR evaluation protocol and show that GutenOCR-7B more than doubles the composite grounded OCR score of its Qwen2.5-VL-7B backbone on 10.5K held-out business and scientific pages (0.40 to 0.82). On Fox and OmniDocBench v1.5, our approach substantially improves region- and line-level OCR as well as text-detection recall, but reveals trade-offs in page-level linearization, color-guided OCR, and formula-heavy layouts.
Authors: Mohammad Shamim Ahsan, Peng Liu
Abstract: In the network security domain, due to practical issues -- including imbalanced data and heterogeneous legitimate network traffic -- adversarial attacks in machine learning-based NIDSs have been viewed as attack packets misclassified as benign. Due to this prevailing belief, the possibility of (maliciously) perturbed benign packets being misclassified as attack has been largely ignored. In this paper, we demonstrate that this is not only theoretically possible, but also a particular threat to NIDS. In particular, we uncover a practical cyberattack, FPR manipulation attack (FPA), especially targeting industrial IoT networks, where domain-specific knowledge of the widely used MQTT protocol is exploited and a systematic simple packet-level perturbation is performed to alter the labels of benign traffic samples without employing traditional gradient-based or non-gradient-based methods. The experimental evaluations demonstrate that this novel attack results in a success rate of 80.19% to 100%. In addition, while estimating impacts in the Security Operations Center, we observe that even a small fraction of false positive alerts, irrespective of different budget constraints and alert traffic intensities, can increase the delay of genuine alerts investigations up to 2 hr in a single day under normal operating conditions. Furthermore, a series of relevant statistical and XAI analyses is conducted to understand the key factors behind this remarkable success. Finally, we explore the effectiveness of the FPA packets to enhance models' robustness through adversarial training and investigate the changes in decision boundaries accordingly.
Authors: Zhengang Zhong, Yury Korolev, Matthew Thorpe
Abstract: Laplace learning is a semi-supervised method, a solution for finding missing labels from a partially labeled dataset utilizing the geometry given by the unlabeled data points. The method minimizes a Dirichlet energy defined on a (discrete) graph constructed from the full dataset. In finite dimensions the asymptotics in the large (unlabeled) data limit are well understood with convergence from the graph setting to a continuum Sobolev semi-norm weighted by the Lebesgue density of the data-generating measure. The lack of the Lebesgue measure on infinite-dimensional spaces requires rethinking the analysis if the data aren't finite-dimensional. In this paper we make a first step in this direction by analyzing the setting when the data are generated by a Gaussian measure on a Hilbert space and proving pointwise convergence of the graph Dirichlet energy.
Authors: Leyi Zhao, Weijie Huang, Yitong Guo, Jiang Bian, Chenghong Wang, Xuhong Zhang
Abstract: Optimizing scientific computing algorithms for modern GPUs is a labor-intensive and iterative process involving repeated code modification, benchmarking, and tuning across complex hardware and software stacks. Recent work has explored large language model (LLM)-assisted evolutionary methods for automated code optimization, but these approaches primarily rely on outcome-based selection and random mutation, underutilizing the rich trajectory information generated during iterative optimization. We propose PhyloEvolve, an LLM-agent system that reframes GPU-oriented algorithm optimization as an In-Context Reinforcement Learning (ICRL) problem. This formulation enables trajectory-conditioned reuse of optimization experience without model retraining. PhyloEvolve integrates Algorithm Distillation and prompt-based Decision Transformers into an iterative workflow, treating sequences of algorithm modifications and performance feedback as first-class learning signals. To organize optimization history, we introduce a phylogenetic tree representation that captures inheritance, divergence, and recombination among algorithm variants, enabling backtracking, cross-lineage transfer, and reproducibility. The system combines elite trajectory pooling, multi-island parallel exploration, and containerized execution to balance exploration and exploitation across heterogeneous hardware. We evaluate PhyloEvolve on scientific computing workloads including PDE solvers, manifold learning, and spectral graph algorithms, demonstrating consistent improvements in runtime, memory efficiency, and correctness over baseline and evolutionary methods. Code is published at: https://github.com/annihi1ation/phylo_evolve
Authors: Chenglei Si, Zitong Yang, Yejin Choi, Emmanuel Cand\`es, Diyi Yang, Tatsunori Hashimoto
Abstract: Automated AI research holds great potential to accelerate scientific discovery. However, current LLMs often generate plausible-looking but ineffective ideas. Execution grounding may help, but it is unclear whether automated execution is feasible and whether LLMs can learn from the execution feedback. To investigate these, we first build an automated executor to implement ideas and launch large-scale parallel GPU experiments to verify their effectiveness. We then convert two realistic research problems - LLM pre-training and post-training - into execution environments and demonstrate that our automated executor can implement a large fraction of the ideas sampled from frontier LLMs. We analyze two methods to learn from the execution feedback: evolutionary search and reinforcement learning. Execution-guided evolutionary search is sample-efficient: it finds a method that significantly outperforms the GRPO baseline (69.4% vs 48.0%) on post-training, and finds a pre-training recipe that outperforms the nanoGPT baseline (19.7 minutes vs 35.9 minutes) on pre-training, all within just ten search epochs. Frontier LLMs often generate meaningful algorithmic ideas during search, but they tend to saturate early and only occasionally exhibit scaling trends. Reinforcement learning from execution reward, on the other hand, suffers from mode collapse. It successfully improves the average reward of the ideator model but not the upper-bound, due to models converging on simple ideas. We thoroughly analyze the executed ideas and training dynamics to facilitate future efforts towards execution-grounded automated AI research.
Authors: Luis Lazo, Hamed Jelodar, Roozbeh Razavi-Far
Abstract: In this study, we propose a homotopy-inspired prompt obfuscation framework to enhance understanding of security and safety vulnerabilities in Large Language Models (LLMs). By systematically applying carefully engineered prompts, we demonstrate how latent model behaviors can be influenced in unexpected ways. Our experiments encompassed 15,732 prompts, including 10,000 high-priority cases, across LLama, Deepseek, KIMI for code generation, and Claude to verify. The results reveal critical insights into current LLM safeguards, highlighting the need for more robust defense mechanisms, reliable detection strategies, and improved resilience. Importantly, this work provides a principled framework for analyzing and mitigating potential weaknesses, with the goal of advancing safe, responsible, and trustworthy AI technologies.
Authors: Bhaavanaa Thumu, Leena Mathur, Youssouf Kebe, Louis-Philippe Morency
Abstract: Social understanding abilities are crucial for multimodal large language models (MLLMs) to interpret human social interactions. We introduce Social Caption, a framework grounded in interaction theory to evaluate social understanding abilities of MLLMs along three dimensions: Social Inference (SI), the ability to make accurate inferences about interactions; Holistic Social Analysis (HSA), the ability to generate comprehensive descriptions of interactions; Directed Social Analysis (DSA), the ability to extract relevant social information from interactions. We analyze factors influencing model performance in social understanding, such as scale, architectural design, and spoken context. Experiments with MLLM judges contribute insights about scaling automated evaluation of multimodal social understanding.
Authors: Po-Kai Chiu, Hung-Hsuan Chen
Abstract: The requirement for expert annotations limits the effectiveness of deep learning for medical image analysis. Although 3D self-supervised methods like volume contrast learning (VoCo) are powerful and partially address the labeling scarcity issue, their high computational cost and memory consumption are barriers. We propose 2D-VoCo, an efficient adaptation of the VoCo framework for slice-level self-supervised pre-training that learns spatial-semantic features from unlabeled 2D CT slices via contrastive learning. The pre-trained CNN backbone is then integrated into a CNN-LSTM architecture to classify multi-organ injuries. In the RSNA 2023 Abdominal Trauma dataset, 2D-VoCo pre-training significantly improves mAP, precision, recall, and RSNA score over training from scratch. Our framework provides a practical method to reduce the dependency on labeled data and enhance model performance in clinical CT analysis. We release the code for reproducibility. https://github.com/tkz05/2D-VoCo-CT-Classifier
Authors: Weiwei Ma, Xiaobing Yu, Peijie Qiu, Jin Yang, Pan Xiao, Xiaoqi Zhao, Xiaofeng Liu, Tomo Miyazaki, Shinichiro Omachi, Yongsong Huang
Abstract: In clinical practice, medical segmentation datasets are often limited and heterogeneous, with variations in modalities, protocols, and anatomical targets across institutions. Existing deep learning models struggle to jointly learn from such diverse data, often sacrificing either generalization or domain-specific knowledge. To overcome these challenges, we propose a joint training method called Universal Harmonization (U-Harmony), which can be integrated into deep learning-based architectures with a domain-gated head, enabling a single segmentation model to learn from heterogeneous datasets simultaneously. By integrating U-Harmony, our approach sequentially normalizes and then denormalizes feature distributions to mitigate domain-specific variations while preserving original dataset-specific knowledge. More appealingly, our framework also supports universal modality adaptation, allowing the seamless learning of new imaging modalities and anatomical classes. Extensive experiments on cross-institutional brain lesion datasets demonstrate the effectiveness of our approach, establishing a new benchmark for robust and adaptable 3D medical image segmentation models in real-world clinical settings.
Authors: Ziwen Wang, Siqi Li, Marcus Eng Hock Ong, Nan Liu
Abstract: Privacy-preserving model co-training in medical research is often hindered by server-dependent architectures incompatible with protected hospital data systems and by the predominant focus on relative effect measures (hazard ratios) which lack clinical interpretability for absolute survival risk assessment. We propose FedRD, a communication-efficient framework for federated risk difference estimation in distributed survival data. Unlike typical federated learning frameworks (e.g., FedAvg) that require persistent server connections and extensive iterative communication, FedRD is server-independent with minimal communication: one round of summary statistics exchange for the stratified model and three rounds for the unstratified model. Crucially, FedRD provides valid confidence intervals and hypothesis testing--capabilities absent in FedAvg-based frameworks. We provide theoretical guarantees by establishing the asymptotic properties of FedRD and prove that FedRD (unstratified) is asymptotically equivalent to pooled individual-level analysis. Simulation studies and real-world clinical applications across different countries demonstrate that FedRD outperforms local and federated baselines in both estimation accuracy and prediction performance, providing an architecturally feasible solution for absolute risk assessment in privacy-restricted, multi-site clinical studies.
Authors: Wenda Zhang, Hongyu Jin, Siyi Wang, Zhiqiang Wei, Ting Dang
Abstract: Speech Emotion Recognition models typically use single categorical labels, overlooking the inherent ambiguity of human emotions. Ambiguous Emotion Recognition addresses this by representing emotions as probability distributions, but progress is limited by unreliable ground-truth distributions inferred from sparse human annotations. This paper explores whether Large Audio-Language Models (ALMs) can mitigate the annotation bottleneck by generating high-quality synthetic annotations. We introduce a framework leveraging ALMs to create Synthetic Perceptual Proxies, augmenting human annotations to improve ground-truth distribution reliability. We validate these proxies through statistical analysis of their alignment with human distributions and evaluate their impact by fine-tuning ALMs with the augmented emotion distributions. Furthermore, to address class imbalance and enable unbiased evaluation, we propose DiME-Aug, a Distribution-aware Multimodal Emotion Augmentation strategy. Experiments on IEMOCAP and MSP-Podcast show that synthetic annotations enhance emotion distribution, especially in low-ambiguity regions where annotation agreement is high. However, benefits diminish for highly ambiguous emotions with greater human disagreement. This work provides the first evidence that ALMs could address annotation scarcity in ambiguous emotion recognition, but highlights the need for more advanced prompting or generation strategies to handle highly ambiguous cases.
Authors: Yuze Chen, Yuan Zhou, Baichuan Mo, Jie Ying, Yufei Ruan, Zhou Ye
Abstract: We study an online linear programming (OLP) model in which inventory is not provided upfront but instead arrives gradually through an exogenous stochastic replenishment process. This replenishment-based formulation captures operational settings, such as e-commerce fulfillment, perishable supply chains, and renewable-powered systems, where resources are accumulated gradually and initial inventories are small or zero. The introduction of dispersed, uncertain replenishment fundamentally alters the structure of classical OLPs, creating persistent stockout risk and eliminating advance knowledge of the total budget. We develop new algorithms and regret analyses for three major distributional regimes studied in the OLP literature: bounded distributions, finite-support distributions, and continuous-support distributions with a non-degeneracy condition. For bounded distributions, we design an algorithm that achieves $\widetilde{\mathcal{O}}(\sqrt{T})$ regret. For finite-support distributions with a non-degenerate induced LP, we obtain $\mathcal{O}(\log T)$ regret, and we establish an $\Omega(\sqrt{T})$ lower bound for degenerate instances, demonstrating a sharp separation from the classical setting where $\mathcal{O}(1)$ regret is achievable. For continuous-support, non-degenerate distributions, we develop a two-stage accumulate-then-convert algorithm that achieves $\mathcal{O}(\log^2 T)$ regret, comparable to the $\mathcal{O}(\log T)$ regret in classical OLPs. Together, these results provide a near-complete characterization of the optimal regret achievable in OLP with replenishment. Finally, we empirically evaluate our algorithms and demonstrate their advantages over natural adaptations of classical OLP methods in the replenishment setting.
Authors: Jinyang Liao, Ziyang Lyu
Abstract: This paper presents a semi-supervised learning framework for Gaussian mixture modelling under a Missing at Random (MAR) mechanism. The method explicitly parameterizes the missingness mechanism by modelling the probability of missingness as a function of classification uncertainty. To quantify classification uncertainty, we introduce margin confidence and incorporate the Aranda Ordaz (AO) link function to flexibly capture the asymmetric relationships between uncertainty and missing probability. Based on this formulation, we develop an efficient Expectation Conditional Maximization (ECM) algorithm that jointly estimates all parameters appearing in both the Gaussian mixture model (GMM) and the missingness mechanism, and subsequently imputes the missing labels by a Bayesian classifier derived from the fitted mixture model. This method effectively alleviates the bias induced by ignoring the missingness mechanism while enhancing the robustness of semi-supervised learning. The resulting uncertainty-aware framework delivers reliable classification performance in realistic MAR scenarios with substantial proportions of missing labels.
Authors: Yuchen Fang, Xinshou Zheng, Javad Lavaei
Abstract: We propose a stochastic trust-region method for unconstrained nonconvex optimization that incorporates stochastic variance-reduced gradients (SVRG) to accelerate convergence. Unlike classical trust-region methods, the proposed algorithm relies solely on stochastic gradient information and does not require function value evaluations. The trust-region radius is adaptively adjusted based on a radius-control parameter and the stochastic gradient estimate. Under mild assumptions, we establish that the algorithm converges in expectation to a first-order stationary point. Moreover, the method achieves iteration and sample complexity bounds that match those of SVRG-based first-order methods, while allowing stochastic and potentially gradient-dependent second-order information. Extensive numerical experiments demonstrate that incorporating SVRG accelerates convergence, and that the use of trust-region methods and Hessian information further improves performance. We also highlight the impact of batch size and inner-loop length on efficiency, and show that the proposed method outperforms SGD and Adam on several machine learning tasks.
Authors: Mingyang Xie, Numair Khan, Tianfu Wang, Naina Dhingra, Seonghyeon Nam, Haitao Yang, Zhuo Hui, Christopher Metzler, Andrea Vedaldi, Hamed Pirsiavash, Lei Luo
Abstract: Given a monocular video, the goal of video re-rendering is to generate views of the scene from a novel camera trajectory. Existing methods face two distinct challenges. Geometrically unconditioned models lack spatial awareness, leading to drift and deformation under viewpoint changes. On the other hand, geometrically-conditioned models depend on estimated depth and explicit reconstruction, making them susceptible to depth inaccuracies and calibration errors. We propose to address these challenges by using the implicit geometric knowledge embedded in the latent space of a large 4D reconstruction model to condition the video generation process. These latents capture scene structure in a continuous space without explicit reconstruction. Therefore, they provide a flexible representation that allows the pretrained diffusion prior to regularize errors more effectively. By jointly conditioning on these latents and source camera poses, we demonstrate that our model achieves state-of-the-art results on the video re-rendering task. Project webpage is https://lavr-4d-scene-rerender.github.io/
Authors: Justin Cheung, Samuel Savine, Calvin Nguyen, Lin Lu, Alhassan S. Yasin
Abstract: Supervised deep learning models often achieve excellent performance within their training distribution but struggle to generalize beyond it. In cancer histopathology, for example, a convolutional neural network (CNN) may classify cancer severity accurately for cancer types represented in its training data, yet fail on related but unseen types. Although adenocarcinomas from different organs share morphological features that might support limited cross-domain generalization, addressing domain shift directly is necessary for robust performance. Domain adaptation offers a way to transfer knowledge from labeled data in one cancer type to unlabeled data in another, helping mitigate the scarcity of annotated medical images. This work evaluates cross-domain classification performance among lung, colon, breast, and kidney adenocarcinomas. A ResNet50 trained on any single adenocarcinoma achieves over 98% accuracy on its own domain but shows minimal generalization to others. Ensembling multiple supervised models does not resolve this limitation. In contrast, converting the ResNet50 into a domain adversarial neural network (DANN) substantially improves performance on unlabeled target domains. A DANN trained on labeled breast and colon data and adapted to unlabeled lung data reaches 95.56% accuracy. We also examine the impact of stain normalization on domain adaptation. Its effects vary by target domain: for lung, accuracy drops from 95.56% to 66.60%, while for breast and colon targets, stain normalization boosts accuracy from 49.22% to 81.29% and from 78.48% to 83.36%, respectively. Finally, using Integrated Gradients reveals that DANNs consistently attribute importance to biologically meaningful regions such as densely packed nuclei, indicating that the model learns clinically relevant features and can apply them to unlabeled cancer types.
Authors: Shuai Wang, Yaoming Yang, Bingdong Li, Hao Hao, Aimin Zhou
Abstract: Learning Path Recommendation (LPR) aims to generate personalized sequences of learning items that maximize long-term learning effect while respecting pedagogical principles and operational constraints. Although large language models (LLMs) offer rich semantic understanding for free-form recommendation, applying them to long-horizon LPR is challenging due to (i) misalignment with pedagogical objectives such as the Zone of Proximal Development (ZPD) under sparse, delayed feedback, (ii) scarce and costly expert demonstrations, and (iii) multi-objective interactions among learning effect, difficulty scheduling, length controllability, and trajectory diversity. To address these issues, we propose IB-GRPO (Indicator-Based Group Relative Policy Optimization), an indicator-guided alignment approach for LLM-based LPR. To mitigate data scarcity, we construct hybrid expert demonstrations via Genetic Algorithm search and teacher RL agents and warm-start the LLM with supervised fine-tuning. Building on this warm-start, we design a within-session ZPD alignment score for difficulty scheduling. IB-GRPO then uses the $I_{\epsilon+}$ dominance indicator to compute group-relative advantages over multiple objectives, avoiding manual scalarization and improving Pareto trade-offs. Experiments on ASSIST09 and Junyi using the KES simulator with a Qwen2.5-7B backbone show consistent improvements over representative RL and LLM baselines.
Authors: Casimir Czworkowski, Stephen Hornish, Alhassan S. Yasin
Abstract: Proximal Policy Optimization (PPO) is a widely used reinforcement learning algorithm known for its stability and sample efficiency, but it often suffers from premature convergence due to limited exploration. In this paper, we propose POEM (Proximal Policy Optimization with Evolutionary Mutations), a novel modification to PPO that introduces an adaptive exploration mechanism inspired by evolutionary algorithms. POEM enhances policy diversity by monitoring the Kullback-Leibler (KL) divergence between the current policy and a moving average of previous policies. When policy changes become minimal, indicating stagnation, POEM triggers an adaptive mutation of policy parameters to promote exploration. We evaluate POEM on four OpenAI Gym environments: CarRacing, MountainCar, BipedalWalker, and LunarLander. Through extensive fine-tuning using Bayesian optimization techniques and statistical testing using Welch's t-test, we find that POEM significantly outperforms PPO on three of the four tasks (BipedalWalker: t=-2.0642, p=0.0495; CarRacing: t=-6.3987, p=0.0002; MountainCar: t=-6.2431, p<0.0001), while performance on LunarLander is not statistically significant (t=-1.8707, p=0.0778). Our results highlight the potential of integrating evolutionary principles into policy gradient methods to overcome exploration-exploitation tradeoffs.
Authors: Mingxuan Song, Yusen Huo, Bohan Zhou, Shenglin Yin, Zhen Xiao, Jieyi Long, Zhilin Zhang, Chuan Yu
Abstract: Optimizing the advertiser's cumulative value of winning impressions under budget constraints poses a complex challenge in online advertising, under the paradigm of AI-Generated Bidding (AIGB). Advertisers often have personalized objectives but limited historical interaction data, resulting in few-shot scenarios where traditional reinforcement learning (RL) methods struggle to perform effectively. Large Language Models (LLMs) offer a promising alternative for AIGB by leveraging their in-context learning capabilities to generalize from limited data. However, they lack the numerical precision required for fine-grained optimization. To address this limitation, we introduce GRPO-Adaptive, an efficient LLM post-training strategy that enhances both reasoning and numerical precision by dynamically updating the reference policy during training. Built upon this foundation, we further propose DARA, a novel dual-phase framework that decomposes the decision-making process into two stages: a few-shot reasoner that generates initial plans via in-context prompting, and a fine-grained optimizer that refines these plans using feedback-driven reasoning. This separation allows DARA to combine LLMs' in-context learning strengths with precise adaptability required by AIGB tasks. Extensive experiments on both real-world and synthetic data environments demonstrate that our approach consistently outperforms existing baselines in terms of cumulative advertiser value under budget constraints.
Authors: Chun-Yi Kuan, Kai-Wei Chang, Hung-yi Lee
Abstract: Although text-to-audio generation has made remarkable progress in realism and diversity, the development of evaluation metrics has not kept pace. Widely-adopted approaches, typically based on embedding similarity like CLAPScore, effectively measure general relevance but remain limited in fine-grained semantic alignment and compositional reasoning. To address this, we introduce AQAScore, a backbone-agnostic evaluation framework that leverages the reasoning capabilities of audio-aware large language models (ALLMs). AQAScore reformulates assessment as a probabilistic semantic verification task; rather than relying on open-ended text generation, it estimates alignment by computing the exact log-probability of a "Yes" answer to targeted semantic queries. We evaluate AQAScore across multiple benchmarks, including human-rated relevance, pairwise comparison, and compositional reasoning tasks. Experimental results show that AQAScore consistently achieves higher correlation with human judgments than similarity-based metrics and generative prompting baselines, showing its effectiveness in capturing subtle semantic inconsistencies and scaling with the capability of underlying ALLMs.
Authors: Shuning Ge, Fangyun Qin, Xiaohui Wan, Yang Liu, Qian Dai, Zheng Zheng
Abstract: Software systems that run for long periods often suffer from software aging, which is typically caused by Aging-Related Bugs (ARBs). To mitigate the risk of ARBs early in the development phase, ARB prediction has been introduced into software aging research. However, due to the difficulty of collecting ARBs, within-project ARB prediction faces the challenge of data scarcity, leading to the proposal of cross-project ARB prediction. This task faces two major challenges: 1) domain adaptation issue caused by distribution difference between source and target projects; and 2) severe class imbalance between ARB-prone and ARB-free samples. Although various methods have been proposed for cross-project ARB prediction, existing approaches treat the input metrics independently and often neglect the rich inter-metric dependencies, which can lead to overlapping information and misjudgment of metric importance, potentially affecting the model's performance. Moreover, they typically use cross-entropy as the loss function during training, which cannot distinguish the difficulty of sample classification. To overcome these limitations, we propose ARFT-Transformer, a transformer-based cross-project ARB prediction framework that introduces a metric-level multi-head attention mechanism to capture metric interactions and incorporates Focal Loss function to effectively handle class imbalance. Experiments conducted on three large-scale open-source projects demonstrate that ARFT-Transformer on average outperforms state-of-the-art cross-project ARB prediction methods in both single-source and multi-source cases, achieving up to a 29.54% and 19.92% improvement in Balance metric.
Authors: Ziyao Ling, Silvia Mirri, Paola Salomoni, Giovanni Delnevo
Abstract: The scarcity of training data presents a fundamental challenge in applying deep learning to archaeological artifact classification, particularly for the rare types of Chinese porcelain. This study investigates whether synthetic images generated through Stable Diffusion with Low-Rank Adaptation (LoRA) can effectively augment limited real datasets for multi-task CNN-based porcelain classification. Using MobileNetV3 with transfer learning, we conducted controlled experiments comparing models trained on pure real data against those trained on mixed real-synthetic datasets (95:5 and 90:10 ratios) across four classification tasks: dynasty, glaze, kiln and type identification. Results demonstrate task-specific benefits: type classification showed the most substantial improvement (5.5\% F1-macro increase with 90:10 ratio), while dynasty and kiln tasks exhibited modest gains (3-4\%), suggesting that synthetic augmentation effectiveness depends on the alignment between generated features and task-relevant visual signatures. Our work contributes practical guidelines for deploying generative AI in archaeological research, demonstrating both the potential and limitations of synthetic data when archaeological authenticity must be balanced with data diversity.
Authors: Dimitrios G. Patsatzis, Alessandro Della Pia, Lucia Russo, Constantinos Siettos
Abstract: We introduce RANDSMAPs (Random-feature/multi-scale neural decoders with Mass Preservation), numerical analysis-informed, explainable neural decoders designed to explicitly respect conservation laws when solving the challenging ill-posed pre-image problem in manifold learning. We start by proving the equivalence of vanilla random Fourier feature neural networks to Radial Basis Function interpolation and the double Diffusion Maps (based on Geometric Harmonics) decoders in the deterministic limit. We then establish the theoretical foundations for RANDSMAP and introduce its multiscale variant to capture structures across multiple scales. We formulate and derive the closed-form solution of the corresponding constrained optimization problem and prove the mass preservation property. Numerically, we assess the performance of RANDSMAP on three benchmark problems/datasets with mass preservation obtained by the Lighthill-Whitham-Richards traffic flow PDE with shock waves, 2D rotated MRI brain images, and the Hughes crowd dynamics PDEs. We demonstrate that RANDSMAPs yield high reconstruction accuracy at low computational cost and maintain mass conservation at single-machine precision. In its vanilla formulation, the scheme remains applicable to the classical pre-image problem, i.e., when mass-preservation constraints are not imposed.
Authors: Anaclara Alvez-Canepa, Cyril Furtlehner, Fran\c{c}ois Landes
Abstract: Machine Learning (ML) has deeply changed some fields recently, like Language and Vision and we may expect it to be relevant also to the analysis of of complex systems. Here we want to tackle the question of how and to which extent can one regress scale-free processes, i.e. processes displaying power law behavior, like earthquakes or avalanches? We are interested in predicting the large ones, i.e. rare events in the training set which therefore require extrapolation capabilities of the model. For this we consider two paradigmatic problems that are statistically self-similar. The first one is a 2-dimensional fractional Gaussian field obeying linear dynamics, self-similar by construction and amenable to exact analysis. The second one is the Abelian sandpile model, exhibiting self-organized criticality. The emerging paradigm of Geometric Deep Learning shows that including known symmetries into the model's architecture is key to success. Here one may hope to extrapolate only by leveraging scale invariance. This is however a peculiar symmetry, as it involves possibly non-trivial coarse-graining operations and anomalous scaling. We perform experiments on various existing architectures like U-net, Riesz network (scale invariant by construction), or our own proposals: a wavelet-decomposition based Graph Neural Network (with discrete scale symmetry), a Fourier embedding layer and a Fourier-Mellin Neural Operator. Based on these experiments and a complete characterization of the linear case, we identify the main issues relative to spectral biases and coarse-grained representations, and discuss how to alleviate them with the relevant inductive biases.
Authors: Sidi Mohamed Sid El Moctar, Achraf Ait Laydi, Yousef El Mourabit, H\'el\`ene Bouvrais
Abstract: Microtubules are cytoskeletal filaments that play essential roles in many cellular processes and are key therapeutic targets in several diseases. Accurate segmentation of microtubule networks is critical for studying their organization and dynamics but remains challenging due to filament curvature, dense crossings, and image noise. We present MTFlow, a novel time-conditioned flow-matching model for microtubule segmentation. Unlike conventional U-Net variants that predict masks in a single pass, MTFlow learns vector fields that iteratively transport noisy masks toward the ground truth, enabling interpretable, trajectory-based refinement. Our architecture combines a U-Net backbone with temporal embeddings, allowing the model to capture the dynamics of uncertainty resolution along filament boundaries. We trained and evaluated MTFlow on synthetic and real microtubule datasets and assessed its generalization capability on public biomedical datasets of curvilinear structures such as retinal blood vessels and nerves. MTFlow achieves competitive segmentation accuracy comparable to state-of-the-art models, offering a powerful and time-efficient tool for filamentous structure analysis with more precise annotations than manual or semi-automatic approaches.
Authors: Hirofumi Ota, Masaaki Imaizumi
Abstract: We study a noisy linear observation model with an unknown permutation called permuted/shuffled linear regression, where responses and covariates are mismatched and the permutation forms a discrete, factorial-size parameter. This unknown permutation is a key component of the data-generating process, yet its statistical investigation remains challenging due to its discrete nature. In this study, we develop a general statistical inference framework on the permutation and regression coefficients. First, we introduce a localization step that reduces the permutation space to a small candidate set building on recent advances in the repro samples method, whose miscoverage decays polynomially with the number of Monte Carlo samples. Then, based on this localized set, we provide statistical inference procedures: a conditional Monte Carlo test of permutation structures with valid finite-sample Type-I error control. We also develop coefficient inference that remains valid under alignment uncertainty of permutations. For computational purposes, we develop a linear assignment problem computable in polynomial time complexity and demonstrate that its solution asymptotically converges to that of the conventional least squares problem with large computational cost. Extensions to partially permuted designs and ridge regularization are also discussed. Extensive simulations and an application to Beijing air-quality data corroborate finite-sample validity, strong power to detect mismatches, and practical scalability.
Authors: Miguel J. S. Martinho, Hamed Valizadegan, Jon M. Jenkins, Douglas A. Caldwell, Joseph D. Twicken, Ben Tofflemire, Marziye Jafariyazani
Abstract: The Transiting Exoplanet Survey Satellite (TESS) Full-Frame Images (FFIs) provide photometric time series for millions of stars, enabling transit searches beyond the limited set of pre-selected 2-minute targets. However, FFIs present additional challenges for transit identification and vetting. In this work, we apply ExoMiner++ 2.0, an adaptation of the ExoMiner++ framework originally developed for TESS 2-minute data, to FFI light curves. The model is used to perform large-scale planet versus non-planet classification of Threshold Crossing Events across the sectors analyzed in this study. We construct a uniform vetting catalog of all evaluated signals and assess model performance under different observing conditions. We find that ExoMiner++ 2.0 generalizes effectively to the FFI domain, providing robust discrimination between planetary signals, astrophysical false positives, and instrumental artifacts despite the limitations inherent to longer cadence data. This work extends the applicability of ExoMiner++ to the full TESS dataset and supports future population studies and follow-up prioritization.
Authors: Michelle Ching, Ioana Popescu, Nico Smith, Tianyi Ma, William G. Underwood, Richard J. Samworth
Abstract: We study in-context learning for nonparametric regression with $\alpha$-H\"older smooth regression functions, for some $\alpha>0$. We prove that, with $n$ in-context examples and $d$-dimensional regression covariates, a pretrained transformer with $\Theta(\log n)$ parameters and $\Omega\bigl(n^{2\alpha/(2\alpha+d)}\log^3 n\bigr)$ pretraining sequences can achieve the minimax-optimal rate of convergence $O\bigl(n^{-2\alpha/(2\alpha+d)}\bigr)$ in mean squared error. Our result requires substantially fewer transformer parameters and pretraining sequences than previous results in the literature. This is achieved by showing that transformers are able to approximate local polynomial estimators efficiently by implementing a kernel-weighted polynomial basis and then running gradient descent.
Authors: Isaac Baglin, Xiatian Zhu, Simon Hadfield
Abstract: Traditional defenses against Deep Leakage (DL) attacks in Federated Learning (FL) primarily focus on obfuscation, introducing noise, transformations or encryption to degrade an attacker's ability to reconstruct private data. While effective to some extent, these methods often still leak high-level information such as class distributions or feature representations, and are frequently broken by increasingly powerful denoising attacks. We propose a fundamentally different perspective on FL defense: framing it as a spoofing problem.We introduce SpooFL (Figure 1), a spoofing-based defense that deceives attackers into believing they have recovered the true training data, while actually providing convincing but entirely synthetic samples from an unrelated task. Unlike prior synthetic-data defenses that share classes or distributions with the private data and thus still leak semantic information, SpooFL uses a state-of-the-art generative model trained on an external dataset with no class overlap. As a result, attackers are misled into recovering plausible yet completely irrelevant samples, preventing meaningful data leakage while preserving FL training integrity. We implement the first example of such a spoofing defense, and evaluate our method against state-of-the-art DL defenses and demonstrate that it successfully misdirects attackers without compromising model performance significantly.
Authors: Srinath Srinivasan, Tim Menzies, Marcelo D'Amorim
Abstract: Differential fuzzers detect bugs by executing identical inputs across distinct implementations of the same specification, such as JavaScript interpreters. Validating the outputs requires an oracle and for differential testing of JavaScript, these are constructed manually, making them expensive, time-consuming, and prone to false positives. Worse, when the specification evolves, this manual effort must be repeated. Inspired by the success of agentic systems in other SE domains, this paper introduces SmartOracle. SmartOracle decomposes the manual triage workflow into specialized Large Language Model (LLM) sub-agents. These agents synthesize independently gathered evidence from terminal runs and targeted specification queries to reach a final verdict. For historical benchmarks, SmartOracle achieves 0.84 recall with an 18% false positive rate. Compared to a sequential Gemini 2.5 Pro baseline, it improves triage accuracy while reducing analysis time by 4$\times$ and API costs by 10$\times$. In active fuzzing campaigns, SmartOracle successfully identified and reported previously unknown specification-level issues across major engines, including bugs in V8, JavaScriptCore, and GraalJS. The success of SmartOracle's agentic architecture on Javascript suggests it might be useful other software systems- a research direction we will explore in future work.
Authors: Christopher Scofield
Abstract: Multi-agent systems (MAS) composed of large language models often exhibit improved problem-solving performance despite operating on identical information. In this work, we provide a formal explanation for this phenomenon grounded in operator theory and constrained optimization. We model each agent as enforcing a distinct family of validity constraints on a shared solution state, and show that a MAS implements a factorized composition of constraint-enforcement operators. Under mild conditions, these dynamics converge to invariant solution sets defined by the intersection of agent constraint sets. Such invariant structures are generally not dynamically accessible to a single agent applying all constraints simultaneously, even when expressive capacity and information are identical. We extend this result from exact constraint enforcement to soft constraints via proximal operators, and apply the formalism to contemporary text-based dialog systems.
Authors: Muntakimur Rahaman, Md Mahmudul Hoque, Md Mehedi Hassain
Abstract: Bangla music is enrich in its own music cultures. Now a days music genre classification is very significant because of the exponential increase in available music, both in digital and physical formats. It is necessary to index them accordingly to facilitate improved retrieval. Automatically classifying Bangla music by genre is essential for efficiently locating specific pieces within a vast and diverse music library. Prevailing methods for genre classification predominantly employ conventional machine learning or deep learning approaches. This work introduces a novel music dataset comprising ten distinct genres of Bangla music. For the task of audio classification, we utilize a recurrent neural network (RNN) architecture. Specifically, a Long Short-Term Memory (LSTM) network is implemented to train the model and perform the classification. Feature extraction represents a foundational stage in audio data processing. This study utilizes Mel-Frequency Cepstral Coefficients (MFCCs) to transform raw audio waveforms into a compact and representative set of features. The proposed framework facilitates music genre classification by leveraging these extracted features. Experimental results demonstrate a classification accuracy of 78%, indicating the system's strong potential to enhance and streamline the organization of Bangla music genres.
Authors: Giulio Bordieri, Giorgio Cartechini, Anna Bianchi, Anna Selva, Valeria Conte, Marta Missiaggia, Francesco G. Cordoni
Abstract: The relationship between the physical characteristics of the radiation field and biological damage is central to both radiotherapy and radioprotection, yet the link between spatial scales of energy deposition and biological effects remains not entirely understood. To address this, we developed an interpretable deep learning model that predicts cell survival after proton and carbon ion irradiation, leveraging sequential attention to highlight relevant features and provide insight into the contribution of different energy deposition scales. Trained and tested on the PIDE dataset, our model incorporates, beside LET, nanodosimetric and microdosimetric quantities simulated with MC-Startrack and Open-TOPAS, enabling multi-scale characterization. While achieving high predictive accuracy, our approach also emphasizes transparency in decision-making. We demonstrate high accuracy in predicting RBE for in vitro experiments. Multiple scales are utilized concurrently, with no single spatial scale being predominant. Quantities defined at smaller spatial domains generally have a greater influence, whereas the LET plays a lesser role.
Authors: Gokul Karthik Kumar, Ludovick Lepauloux, Hakim Hacid
Abstract: Whisper has become the de-facto encoder for extracting general-purpose audio features in large audio-language models, where a 30-second clip is typically represented by 1500 frame features projected into an LLM. In contrast, audio-text embedding models like CLAP-based models have largely relied on alternative audio encoders (e.g., HTS-AT, PaSST), and have not leveraged Whisper effectively. We present WavLink, a compact audio-text embedding model that augments Whisper encoder with a learnable global token, trained jointly with a text encoder. Through a systematic study of design choices, including pretrained text encoders, loss functions, training modes, and data mixtures, we identify configurations that yield state-of-the-art retrieval performance. Our two-stage training recipe across three model sizes, combined with Matryoshka-style supervision, improves scalability, enabling 8x smaller embeddings with minimal performance drop. WavLink also demonstrates competitive performance on AIR-Bench with MCQs and zero-shot classification.
Authors: Andr\'e Eberhard, Gerhard Neumann, Pascal Friederich
Abstract: Despite tremendous improvements in tasks such as image classification, object detection, and segmentation, the recognition of visual relationships, commonly modeled as the extraction of a graph from an image, remains a challenging task. We believe that this mainly stems from the fact that there is no canonical way to approach the visual graph recognition task. Most existing solutions are specific to a problem and cannot be transferred between different contexts out-of-the box, even though the conceptual problem remains the same. With broad applicability and simplicity in mind, in this paper we develop a method, \textbf{Gra}ph Recognition via \textbf{S}ubgraph \textbf{P}rediction (\textbf{GraSP}), for recognizing graphs in images. We show across several synthetic benchmarks and one real-world application that our method works with a set of diverse types of graphs and their drawings, and can be transferred between tasks without task-specific modifications, paving the way to a more unified framework for visual graph recognition.
Authors: Zanlin Ni, Shenzhi Wang, Yang Yue, Tianyu Yu, Weilin Zhao, Yeguo Hua, Tianyi Chen, Jun Song, Cheng Yu, Bo Zheng, Gao Huang
Abstract: Diffusion Large Language Models (dLLMs) break the rigid left-to-right constraint of traditional LLMs, enabling token generation in arbitrary orders. Intuitively, this flexibility implies a solution space that strictly supersets the fixed autoregressive trajectory, theoretically unlocking superior reasoning potential for general tasks like mathematics and coding. Consequently, numerous works have leveraged reinforcement learning (RL) to elicit the reasoning capability of dLLMs. In this paper, we reveal a counter-intuitive reality: arbitrary order generation, in its current form, narrows rather than expands the reasoning boundary of dLLMs. We find that dLLMs tend to exploit this order flexibility to bypass high-uncertainty tokens that are crucial for exploration, leading to a premature collapse of the solution space. This observation challenges the premise of existing RL approaches for dLLMs, where considerable complexities, such as handling combinatorial trajectories and intractable likelihoods, are often devoted to preserving this flexibility. We demonstrate that effective reasoning is better elicited by intentionally forgoing arbitrary order and applying standard Group Relative Policy Optimization (GRPO) instead. Our approach, JustGRPO, is minimalist yet surprisingly effective (e.g., 89.1% accuracy on GSM8K) while fully retaining the parallel decoding ability of dLLMs. Project page: https://nzl-thu.github.io/the-flexibility-trap
Authors: Fabi Nahian Madhurja, Rusab Sarmun, Muhammad E. H. Chowdhury, Adam Mushtak, Israa Al-Hashimi, Sohaib Bassam Zoghoul
Abstract: Cervical spine fractures are critical medical conditions requiring precise and efficient detection for effective clinical management. This study explores the viability of 2D projection-based vertebra segmentation for vertebra-level fracture detection in 3D CT volumes, presenting an end-to-end pipeline for automated analysis of cervical vertebrae (C1-C7). By approximating a 3D volume through optimized 2D axial, sagittal, and coronal projections, regions of interest are identified using the YOLOv8 model from all views and combined to approximate the 3D cervical spine area, achieving a 3D mIoU of 94.45 percent. This projection-based localization strategy reduces computational complexity compared to traditional 3D segmentation methods while maintaining high performance. It is followed by a DenseNet121-Unet-based multi-label segmentation leveraging variance- and energy-based projections, achieving a Dice score of 87.86 percent. Strategic approximation of 3D vertebral masks from these 2D segmentation masks enables the extraction of individual vertebra volumes. The volumes are analyzed for fractures using an ensemble of 2.5D Spatio-Sequential models incorporating both raw slices and projections per vertebra for complementary evaluation. This ensemble achieves vertebra-level and patient-level F1 scores of 68.15 and 82.26, and ROC-AUC scores of 91.62 and 83.04, respectively. We further validate our approach through an explainability study that provides saliency map visualizations highlighting anatomical regions relevant for diagnosis, and an interobserver variability analysis comparing our model's performance with expert radiologists, demonstrating competitive results.
Authors: Kexin Wang, Salil Bhate, Jo\~ao M. Pereira, Joe Kileel, Matylda Figlerowicz, Anna Seigal
Abstract: Principal component analysis (PCA) is a tool to capture factors that explain variation in data. Across domains, data are now collected across multiple contexts (for example, individuals with different diseases, cells of different types, or words across texts). While the factors explaining variation in data are undoubtedly shared across subsets of contexts, no tools currently exist to systematically recover such factors. We develop multi-context principal component analysis (MCPCA), a theoretical and algorithmic framework that decomposes data into factors shared across subsets of contexts. Applied to gene expression, MCPCA reveals axes of variation shared across subsets of cancer types and an axis whose variability in tumor cells, but not mean, is associated with lung cancer progression. Applied to contextualized word embeddings from language models, MCPCA maps stages of a debate on human nature, revealing a discussion between science and fiction over decades. These axes are not found by combining data across contexts or by restricting to individual contexts. MCPCA is a principled generalization of PCA to address the challenge of understanding factors underlying data across contexts.
Authors: Felix Schur, Niklas Pfister, Peng Ding, Sach Mukherjee, Jonas Peters
Abstract: We study the problem of estimating causal effects under hidden confounding in the following unpaired data setting: we observe some covariates $X$ and an outcome $Y$ under different experimental conditions (environments) but do not observe them jointly; we either observe $X$ or $Y$. Under appropriate regularity conditions, the problem can be cast as an instrumental variable (IV) regression with the environment acting as a (possibly high-dimensional) instrument. When there are many environments but only a few observations per environment, standard two-sample IV estimators fail to be consistent. We propose a GMM-type estimator based on cross-fold sample splitting of the instrument-covariate sample and prove that it is consistent as the number of environments grows but the sample size per environment remains constant. We further extend the method to sparse causal effects via $\ell_1$-regularized estimation and post-selection refitting.
Authors: Yu Wu, Minsik Jeon, Jen-Hao Rick Chang, Oncel Tuzel, Shubham Tulsiani
Abstract: We study positional encodings for multi-view transformers that process tokens from a set of posed input images, and seek a mechanism that encodes patches uniquely, allows SE(3)-invariant attention with multi-frequency similarity, and can be adaptive to the geometry of the underlying scene. We find that prior (absolute or relative) encoding schemes for multi-view attention do not meet the above desiderata, and present RayRoPE to address this gap. RayRoPE represents patch positions based on associated rays but leverages a predicted point along the ray instead of the direction for a geometry-aware encoding. To achieve SE(3) invariance, RayRoPE computes query-frame projective coordinates for computing multi-frequency similarity. Lastly, as the 'predicted' 3D point along a ray may not be precise, RayRoPE presents a mechanism to analytically compute the expected position encoding under uncertainty. We validate RayRoPE on the tasks of novel-view synthesis and stereo depth estimation and show that it consistently improves over alternate position encoding schemes (e.g. 15% relative improvement on LPIPS in CO3D). We also show that RayRoPE can seamlessly incorporate RGB-D input, resulting in even larger gains over alternatives that cannot positionally encode this information.
Authors: Shantanu Jaiswal, Mihir Prabhudesai, Nikash Bhardwaj, Zheyang Qin, Amir Zadeh, Chuan Li, Katerina Fragkiadaki, Deepak Pathak
Abstract: Text-to-image (T2I) models have achieved remarkable progress, yet they continue to struggle with complex prompts that require simultaneously handling multiple objects, relations, and attributes. Existing inference-time strategies, such as parallel sampling with verifiers or simply increasing denoising steps, can improve prompt alignment but remain inadequate for richly compositional settings where many constraints must be satisfied. Inspired by the success of chain-of-thought reasoning in large language models, we propose an iterative test-time strategy in which a T2I model progressively refines its generations across multiple steps, guided by feedback from a vision-language model as the critic in the loop. Our approach is simple, requires no external tools or priors, and can be flexibly applied to a wide range of image generators and vision-language models. Empirically, we demonstrate consistent gains on image generation across benchmarks: a 16.9% improvement in all-correct rate on ConceptMix (k=7), a 13.8% improvement on T2I-CompBench (3D-Spatial category) and a 12.5% improvement on Visual Jenga scene decomposition compared to compute-matched parallel sampling. Beyond quantitative gains, iterative refinement produces more faithful generations by decomposing complex prompts into sequential corrections, with human evaluators preferring our method 58.7% of the time over 41.3% for the parallel baseline. Together, these findings highlight iterative self-correction as a broadly applicable principle for compositional image generation. Results and visualizations are available at https://iterative-img-gen.github.io/
Authors: Przemyslaw A. Grabowicz, Nicholas Perello, Kenta Takatsu
Abstract: Supervised learning systems are trained using historical data and, if the data was tainted by discrimination, they may unintentionally learn to discriminate against protected groups. We propose that fair learning methods, despite training on potentially discriminatory datasets, shall perform well on fair test datasets. Such dataset shifts crystallize application scenarios for specific fair learning methods. For instance, the removal of direct discrimination can be represented as a particular dataset shift problem. For this scenario, we propose a learning method that provably minimizes model error on fair datasets, while blindly training on datasets poisoned with direct additive discrimination. The method is compatible with existing legal systems and provides a solution to the widely discussed issue of protected groups' intersectionality by striking a balance between the protected groups. Technically, the method applies probabilistic interventions, has causal and counterfactual formulations, and is computationally lightweight - it can be used with any supervised learning model to prevent direct and indirect discrimination via proxies while maximizing model accuracy for business necessity.
Authors: Hanyang Jiang, Yuehaw Khoo, Haizhao Yang
Abstract: Inverse wave scattering aims at determining the properties of an object using data on how the object scatters incoming waves. In order to collect information, sensors are put in different locations to send and receive waves from each other. The choice of sensor positions and incident wave frequencies determines the reconstruction quality of scatterer properties. This paper introduces reinforcement learning to develop precision imaging that decides sensor positions and wave frequencies adaptive to different scatterers in an intelligent way, thus obtaining a significant improvement in reconstruction quality with limited imaging resources. Extensive numerical results will be provided to demonstrate the superiority of the proposed method over existing methods.
Authors: Zhongyi Jiang, Chunmei Wang, Haizhao Yang
Abstract: Nonlinear dynamics is a pervasive phenomenon observed in scientific and engineering disciplines. However, the task of deriving analytical expressions to describe nonlinear dynamics from limited data remains challenging. In this paper, we shall present a novel deep symbolic learning method called the "finite expression method" (FEX) to discover governing equations within a function space containing a finite set of analytic expressions, based on observed dynamic data. The key concept is to employ FEX to generate analytical expressions of the governing equations by learning the derivatives of partial differential equation (PDE) solutions through convolutions. Our numerical results demonstrate that our FEX surpasses other existing methods (such as PDE-Net, SINDy, GP, and SPL) in terms of numerical performance across a range of problems, including time-dependent PDE problems and nonlinear dynamical systems with time-varying coefficients. Moreover, the results highlight FEX's flexibility and expressive power in accurately approximating symbolic governing equations.
Authors: Junchi Yan, Fangyu Ding, Jiawei Sun, Zhaoping Hu, Yunyi Zhou, Lei Zhu
Abstract: Graph invariant learning (GIL) seeks invariant relations between graphs and labels under distribution shifts. Recent works try to extract an invariant subgraph to improve out-of-distribution (OOD) generalization, yet existing approaches either lack explicit control over compactness or rely on hard top-$k$ selection that shrinks the solution space and is only partially differentiable. In this paper, we provide an in-depth analysis of the drawbacks of some existing works and propose a few general principles for invariant subgraph extraction: 1) separability, as encouraged by our sparsity-driven mechanism, to filter out the irrelevant common features; 2) softness, for a broader solution space; and 3) differentiability, for a soundly end-to-end optimization pipeline. Specifically, building on optimal transport, we propose Graph Sinkhorn Attention (GSINA), a fully differentiable, cardinality-constrained attention mechanism that assigns sparse-yet-soft edge weights via Sinkhorn iterations and induces node attention. GSINA provides explicit controls for separability and softness, and uses a Gumbel reparameterization to stabilize training. It convergence behavior is also theoretically studied. Extensive empirical experimental results on both synthetic and real-world
Authors: Hengguan Huang, Xing Shen, Songtao Wang, Lingfa Meng, Dianbo Liu, David Alejandro Duchene, Hao Wang, Samir Bhatt
Abstract: Human cognition excels at transcending sensory input and forming latent representations that structure our understanding of the world. While Large Language Model (LLM) agents demonstrate emergent reasoning and decision-making abilities, they lack a principled framework for capturing latent structures and modeling uncertainty. In this work, we explore for the first time how to bridge LLM agents with probabilistic graphical models (PGMs) to address agentic reasoning under uncertainty. To this end, we introduce Verbalized Probabilistic Graphical Modeling (vPGM), a Bayesian agentic framework that (i) guides LLM agents in following key principles of PGMs through natural language and (ii) refines the resulting posterior distributions via numerical Bayesian inference. Unlike many traditional probabilistic methods requiring substantial domain expertise, vPGM bypasses expert-driven model design, making it well-suited for scenarios with limited assumptions. We evaluated our model on several agentic reasoning tasks, both close-ended and open-ended. Our results indicate that the model effectively enhances confidence calibration and text generation quality.
Authors: Rick Wilming, Artur Dox, Hjalmar Schulz, Marta Oliveira, Benedict Clark, Stefan Haufe
Abstract: Large pre-trained language models have become a crucial backbone for many downstream tasks in natural language processing (NLP), and while they are trained on a plethora of data containing a variety of biases, such as gender biases, it has been shown that they can also inherit such biases in their weights, potentially affecting their prediction behavior. However, it is unclear to what extent these biases also affect feature attributions generated by applying "explainable artificial intelligence" (XAI) techniques, possibly in unfavorable ways. To systematically study this question, we create a gender-controlled text dataset, GECO, in which the alteration of grammatical gender forms induces class-specific words and provides ground truth feature attributions for gender classification tasks. This enables an objective evaluation of the correctness of XAI methods. We apply this dataset to the pre-trained BERT model, which we fine-tune to different degrees, to quantitatively measure how pre-training induces undesirable bias in feature attributions and to what extent fine-tuning can mitigate such explanation bias. To this extent, we provide GECOBench, a rigorous quantitative evaluation framework for benchmarking popular XAI methods. We show a clear dependency between explanation performance and the number of fine-tuned layers, where XAI methods are observed to benefit particularly from fine-tuning or complete retraining of embedding layers.
Authors: Grzegorz G{\l}uch, Berkant Turan, Sai Ganesh Nagarajan, Sebastian Pokutta
Abstract: We formalize and analyze the trade-off between backdoor-based watermarks and adversarial defenses, framing it as an interactive protocol between a verifier and a prover. While previous works have primarily focused on this trade-off, our analysis extends it by identifying transferable attacks as a third, counterintuitive, but necessary option. Our main result shows that for all learning tasks, at least one of the three exists: a watermark, an adversarial defense, or a transferable attack. By transferable attack, we refer to an efficient algorithm that generates queries indistinguishable from the data distribution and capable of fooling all efficient defenders. Using cryptographic techniques, specifically fully homomorphic encryption, we construct a transferable attack and prove its necessity in this trade-off. Finally, we show that tasks of bounded VC-dimension allow adversarial defenses against all attackers, while a subclass allows watermarks secure against fast adversaries.
Authors: Keru Chen, Honghao Wei, Zhigang Deng, Sen Lin
Abstract: The high costs and risks involved in extensive environment interactions hinder the practical application of current online safe reinforcement learning (RL) methods. While offline safe RL addresses this by learning policies from static datasets, the performance therein is usually limited due to reliance on data quality and challenges with out-of-distribution (OOD) actions. Inspired by recent successes in offline-to-online (O2O) RL, it is crucial to explore whether offline safe RL can be leveraged to facilitate faster and safer online policy learning, a direction that has yet to be fully investigated. To fill this gap, we first demonstrate that naively applying existing O2O algorithms from standard RL would not work well in the safe RL setting due to two unique challenges: \emph{erroneous Q-estimations}, resulted from offline-online objective mismatch and offline cost sparsity, and \emph{Lagrangian mismatch}, resulted from difficulties in aligning Lagrange multipliers between offline and online policies. To address these challenges, we introduce \textbf{Marvel}, a novel framework for O2O safe RL, comprising two key components that work in concert: \emph{Value Pre-Alignment} to align the Q-functions with the underlying truth before online learning, and \emph{Adaptive PID Control} to effectively adjust the Lagrange multipliers during online finetuning. Extensive experiments demonstrate that Marvel significantly outperforms existing baselines in both reward maximization and safety constraint satisfaction. By introducing the first policy-finetuning based framework for O2O safe RL, which is compatible with many offline and online safe RL methods, our work has the great potential to advance the field towards more efficient and practical safe RL solutions.
Authors: Kuan Lu, Menghao Huo, Yuxiao Li, Qiang Zhu, Zhenrui Chen
Abstract: Accurate forecasting of renewable energy generation is fundamental to enhancing the dynamic performance of modern power grids, especially under high renewable penetration. This paper presents Channel-Time Patch Time-Series Transformer (CT-PatchTST), a novel deep learning model designed to provide long-term, high-fidelity forecasts of wind and solar power. Unlike conventional time-series models, CT-PatchTST captures both temporal dependencies and inter-channel correlations-features that are critical for effective energy storage planning, control, and dispatch. Reliable forecasting enables proactive deployment of energy storage systems (ESSs), helping to mitigate uncertainties in renewable output, reduce system response time, and optimize storage operation based on location-specific flow and voltage conditions. Evaluated on real-world datasets from Denmark's offshore wind, onshore wind, and solar generation, CT-PatchTST outperforms existing methods in both accuracy and robustness. By enabling predictive, data-driven coordination of ESSs across integrated source-grid-load-storage systems, this work contributes to the design of more stable, responsive, and cost-efficient power networks.
Authors: Simone Macci\`o, Alessandro Carf\`i, Alessio Capitanelli, Peppino Tropea, Massimo Corbo, Fulvio Mastrogiovanni, Michela Picardi
Abstract: Background/Objectives: Falls represent a major health concern for stroke survivors, necessitating effective risk assessment tools. This study proposes the Instrumented Fall Risk Assessment (IFRA) scale, a novel screening tool derived from Instrumented Timed Up and Go (ITUG) test data, designed to capture mobility measures often missed by traditional scales. Methods: We employed a two-step machine learning approach to develop the IFRA scale: first, identifying predictive mobility features from ITUG data and, second, creating a stratification strategy to classify patients into low-, medium-, or high-fall-risk categories. This study included 142 participants, who were divided into training (including synthetic cases), validation, and testing sets (comprising 22 non-fallers and 10 fallers). IFRA's performance was compared against traditional clinical scales (e.g., standard TUG and Mini-BESTest) using Fisher's Exact test. Results: Machine learning analysis identified specific features as key predictors, namely vertical and medio-lateral acceleration, and angular velocity during walking and sit-to-walk transitions. IFRA demonstrated a statistically significant association with fall status (Fisher's Exact test p = 0.004) and was the only scale to assign more than half of the actual fallers to the high-risk category, outperforming the comparative clinical scales in this dataset. Conclusions: This proof-of-concept study demonstrates IFRA's potential as an automated, complementary approach for fall risk stratification in post-stroke patients. While IFRA shows promising discriminative capability, particularly for identifying high-risk individuals, these preliminary findings require validation in larger cohorts before clinical implementation.
Authors: Yerin Kim, Alexander Benvenuti, Bo Chen, Mustafa Karabag, Abhishek Kulkarni, Nathaniel D. Bastian, Ufuk Topcu, Matthew Hale
Abstract: Autonomous systems are increasingly expected to operate in the presence of adversaries, though adversaries may infer sensitive information simply by observing a system. Therefore, present a deceptive sequential decision-making framework that not only conceals sensitive information, but actively misleads adversaries about it. We model autonomous systems as Markov decision processes, with adversaries using inverse reinforcement learning to recover reward functions. To counter them, we present three regularization strategies for policy synthesis problems that actively deceive an adversary about a system's reward. ``Diversionary deception'' leads an adversary to draw any false conclusion about the system's reward function. ``Targeted deception'' leads an adversary to draw a specific false conclusion about the system's reward function. ``Equivocal deception'' leads an adversary to infer that the real reward and a false reward both explain the system's behavior. We show how each form of deception can be implemented in policy optimization problems and analytically bound the loss in total accumulated reward induced by deception. Next, we evaluate these developments in a multi-agent setting. We show that diversionary, targeted, and equivocal deception all steer the adversary to false beliefs while still attaining a total accumulated reward that is at least 98% of its optimal, non-deceptive value.
Authors: Wenyun Li, Wenjie Huang, Chen Sun
Abstract: In many real-world scenarios, reward signal for agents are exceedingly sparse, making it challenging to learn an effective reward function for reward shaping. To address this issue, the proposed approach in this paper performs reward shaping not only by utilizing non-zero-reward transitions but also by employing the \emph{Semi-Supervised Learning} (SSL) technique combined with a novel data augmentation to learn trajectory space representations from the majority of transitions, {i.e}., zero-reward transitions, thereby improving the efficacy of reward shaping. Experimental results in Atari and robotic manipulation demonstrate that our method outperforms supervised-based approaches in reward inference, leading to higher agent scores. Notably, in more sparse-reward environments, our method achieves up to twice the peak scores compared to supervised baselines. The proposed double entropy data augmentation enhances performance, showcasing a 15.8\% increase in best score over other augmentation methods
Authors: Qika Lin, Zhen Peng, Kaize Shi, Kai He, Yiming Xu, Jian Zhang, Erik Cambria, Mengling Feng
Abstract: Recent years have witnessed rapid advances in graph representation learning, with the continuous embedding approach emerging as the dominant paradigm. However, such methods encounter issues regarding parameter efficiency, interpretability, and robustness. Thus, Quantized Graph Representation (QGR) learning has recently gained increasing interest, which represents the graph structure with discrete codes instead of conventional continuous embeddings. Given its analogous representation form to natural language, QGR also possesses the capability to seamlessly integrate graph structures with large language models (LLMs). As this emerging paradigm is still in its infancy yet holds significant promise, we undertake this thorough survey to promote its rapid future prosperity. We first present the background of the general quantization methods and their merits. Moreover, we provide an in-depth demonstration of current QGR studies from the perspectives of quantized strategies, training objectives, distinctive designs, knowledge graph quantization, and applications. We further explore the strategies for code dependence learning and integration with LLMs. At last, we give discussions and conclude future directions, aiming to provide a comprehensive picture of QGR and inspire future research.
Authors: Xu Chu, Zhixin Zhang, Tianyu Jia, Yujie Jin
Abstract: Aligning large language models (LLMs) with human preferences typically demands vast amounts of meticulously curated data, which is both expensive and prone to labeling noise. We propose Stackelberg Game Preference Optimization (SGPO), a robust alignment framework that models alignment as a two-player Stackelberg game between a policy (leader) and a worst-case preference distribution (follower). The proposed SGPO guarantees $\mathcal{O}(\epsilon)$-bounded regret within an $\epsilon$-Wasserstein ball, offering formal robustness to (self-)annotation noise. We instantiate SGPO with Stackelberg Self-Annotated Preference Optimization (SSAPO), which uses minimal human-labeled "seed" preferences and iteratively self-annotates new prompts. In each iteration, SSAPO applies a distributionally robust reweighting of synthetic annotations, ensuring that noisy or biased self-labels do not derail training. Remarkably, using only 2K seed preferences -- about 1/30 of standard human labels -- SSAPO achieves strong win rates against GPT-4 across multiple benchmarks within three iterations. These results highlight that a principled Stackelberg formulation yields data-efficient alignment for LLMs, significantly reducing reliance on costly human annotations.
Authors: Jiayi Fu, Xuandong Zhao, Chengyuan Yao, Heng Wang, Qi Han, Yanghua Xiao
Abstract: Reinforcement Learning from Human Feedback (RLHF) is essential for aligning large language models (LLMs) with human values. However, RLHF is susceptible to \emph{reward hacking}, where the agent exploits flaws in the reward function rather than learning the intended behavior, thus degrading alignment. Although reward shaping helps stabilize RLHF and partially mitigate reward hacking, a systematic investigation into shaping techniques and their underlying principles remains lacking. To bridge this gap, we present a comprehensive study of the prevalent reward shaping methods. Our analysis suggests two key design principles: (1) the RL reward should be bounded, and (2) the RL reward benefits from rapid initial growth followed by gradual convergence. Guided by these insights, we propose Preference As Reward (PAR), a novel approach that leverages the latent preferences embedded within the reward model as the signal for reinforcement learning. Moreover, PAR exhibits two critical variance-reduction properties that contribute to stabilizing the RLHF training process and effectively extending the tolerance window for early stopping. We evaluated PAR on the base model Gemma2-2B using two datasets, Ultrafeedback-Binarized and HH-RLHF. Experimental results demonstrate PAR's superior performance over other reward shaping methods. On the AlpacaEval 2.0 benchmark, PAR achieves a win rate of at least 5 percentage points higher than competing approaches. Furthermore, PAR exhibits remarkable data efficiency, requiring only a single reference reward for optimal performance, and maintains robustness against reward hacking even after two full epochs of training. The code is available at https://github.com/PorUna-byte/PAR.
Authors: Monirul Islam Pavel, Siyi Hu, Mahardhika Pratama, Ryszard Kowalczyk
Abstract: Onboard learning is a transformative approach in edge AI, enabling real-time data processing, decision-making, and adaptive model training directly on resource-constrained devices without relying on centralized servers. This paradigm is crucial for applications demanding low latency, enhanced privacy, and energy efficiency. However, onboard learning faces challenges such as limited computational resources, high inference costs, and security vulnerabilities. This survey explores a comprehensive range of methodologies that address these challenges, focusing on techniques that optimize model efficiency, accelerate inference, and support collaborative learning across distributed devices. Approaches for reducing model complexity, improving inference speed, and ensuring privacy-preserving computation are examined alongside emerging strategies that enhance scalability and adaptability in dynamic environments. By bridging advancements in hardware-software co-design, model compression, and decentralized learning, this survey provides insights into the current state of onboard learning to enable robust, efficient, and secure AI deployment at the edge.
Authors: Dirk Tasche
Abstract: Recalibration of binary probabilistic classifiers to a target prior probability is an important task in areas like credit risk management. However, recalibration of a classifier learned on a training dataset to a target on a test dataset in general is not a well-defined problem because there might be more than one way to transform the original posterior probabilities such that the target is matched. In this paper, methods for recalibration are analysed from a distribution shift perspective. Distribution shift assumptions linked to the area under the curve (AUC) of a probabilistic classifier are found to be useful for the design of meaningful recalibration methods. Two new methods called parametric covariate shift with posterior drift (CSPD) and ROC-based quasi moment matching (QMM) are proposed and tested together with some other methods in an example setting. The outcomes of the test suggest that the QMM methods discussed in the paper can provide appropriately conservative results in evaluations with concave functions like for instance risk weights functions for credit risk.
Authors: Lingkai Kong, Haichuan Wang, Tonghan Wang, Guojun Xiong, Milind Tambe
Abstract: Incorporating pre-collected offline data can substantially improve the sample efficiency of reinforcement learning (RL), but its benefits can break down when the transition dynamics in the offline dataset differ from those encountered online. Existing approaches typically mitigate this issue by penalizing or filtering offline transitions in regions with large dynamics gap. However, their dynamics-gap estimators often rely on KL divergence or mutual information, which can be ill-defined when offline and online dynamics have mismatched support. To address this challenge, we propose CompFlow, a principled framework built on the theoretical connection between flow matching and optimal transport. Specifically, we model the online dynamics as a conditional flow built upon the output distribution of a pretrained offline flow, rather than learning it directly from a Gaussian prior. This composite structure provides two advantages: (1) improved generalization when learning online dynamics under limited interaction data, and (2) a well-defined and stable estimate of the dynamics gap via the Wasserstein distance between offline and online transitions. Building on this dynamics-gap estimator, we further develop an optimistic active data collection strategy that prioritizes exploration in high-gap regions, and show theoretically that it reduces the performance gap to the optimal policy. Empirically, CompFlow consistently outperforms strong baselines across a range of RL benchmarks with shifted-dynamics data.
Authors: Jiaheng Dong, Hong Jia, Soumyajit Chatterjee, Abhirup Ghosh, James Bailey, Ting Dang
Abstract: Speech Foundation Models encounter significant performance degradation when deployed in real-world scenarios involving acoustic domain shifts, such as background noise and speaker accents. Test-time adaptation (TTA) has recently emerged as a viable strategy to address such domain shifts at inference time without requiring access to source data or labels. However, existing TTA approaches, particularly those relying on backpropagation, are memory-intensive, limiting their applicability in speech tasks and resource-constrained settings. Although backpropagation-free methods offer improved efficiency, existing ones exhibit poor accuracy. This is because they are predominantly developed for vision tasks, which fundamentally differ from speech task formulations, noise characteristics, and model architecture, posing unique transferability challenges. In this paper, we introduce E-BATS, the first Efficient BAckpropagation-free TTA framework designed explicitly for speech foundation models. E-BATS achieves a balance between adaptation effectiveness and memory efficiency through three key components: (i) lightweight prompt adaptation for a forward-pass-based feature alignment, (ii) a multi-scale loss to capture both global (utterance-level) and local distribution shifts (token-level) and (iii) a test-time exponential moving average mechanism for stable adaptation across utterances. Experiments conducted on four noisy speech datasets spanning sixteen acoustic conditions demonstrate consistent improvements, with 4.1%-13.5% accuracy gains over backpropagation-free baselines and 2.0-6.4 times GPU memory savings compared to backpropagation-based methods. By enabling scalable and robust adaptation under acoustic variability, this work paves the way for developing more efficient adaptation approaches for practical speech processing systems in real-world environments.
Authors: Tianyi Xu, Jiaxin Liu, Nicholas Mattei, Zizhan Zheng
Abstract: We propose a multi-agent multi-armed bandit (MA-MAB) framework aimed at ensuring fair outcomes across agents while maximizing overall system performance. A key challenge in this setting is decision-making under limited information about arm rewards. To address this, we introduce a novel probing framework that strategically gathers information about selected arms before allocation. In the offline setting, where reward distributions are known, we leverage submodular properties to design a greedy probing algorithm with a provable performance bound. For the more complex online setting, we develop an algorithm that achieves sublinear regret while maintaining fairness. Extensive experiments on synthetic and real-world datasets show that our approach outperforms baseline methods, achieving better fairness and efficiency.
Authors: Andrey Goncharov, Daniil Vyazhev, Petr Sychev, Edvard Khalafyan, Alexey Zaytsev
Abstract: General-purpose Large Language Models (LLMs) are frequently fine-tuned through supervised fine-tuning (SFT) to enhance performance in specific domains. Better results can be achieved by distilling the chain-of-thought of a larger model at the cost of numerous expensive calls and a much greater amount of data. We propose a novel blueprint for efficient fine-tuning that uses reasoning only for complex data identified by entropy. Specifically, across three small open models ($\approx 3B$) we split the training data into complexity categories by a single token answer entropy (ROC AUC $0.73$), fine-tune large language models (LLMs) via SFT and distillation, and show that our pipeline significantly outperforms the standard SFT approach ($0.58$ vs $0.45$ average accuracy) and outperforms the distillation approach ($0.58$ vs $0.56$ average accuracy) while using $81\%$ less data.
Authors: Song Lai, Haohan Zhao, Rong Feng, Changyi Ma, Wenzhuo Liu, Hongbo Zhao, Xi Lin, Dong Yi, Qingfu Zhang, Hongbin Liu, Gaofeng Meng, Fei Zhu
Abstract: Continual post-training (CPT) is a popular and effective technique for adapting foundation models like multimodal large language models to specific and ever-evolving downstream tasks. While existing research has primarily concentrated on methods like data replay, model expansion, or parameter regularization, the fundamental role of the learning paradigm within CPT remains largely unexplored. This paper presents a comparative analysis of two core post-training paradigms: supervised fine-tuning (SFT) and reinforcement fine-tuning (RFT), investigating their respective impacts on knowledge retention during CPT. Our experiments are conducted on a benchmark comprising seven diverse multimodal tasks, utilizing Qwen2.5-VL-7B-Instruct as the base model for continual post-training. The investigation yields two significant findings: (1) When continuously learning on downstream tasks, SFT leads to catastrophic forgetting of previously learned tasks. In contrast, RFT inherently preserves prior knowledge and achieve performance comparable to multi-task training. (2) RFT successfully protects and even enhances the model's general knowledge on standard benchmarks (e.g., MMMU and MMLU-Pro). Conversely, SFT degrades general model capabilities severely. Further analysis reveals that this stability is not primarily due to explicit mechanisms like KL penalty or chain-of-thought reasoning. Instead, we identify an implicit regularization mechanism inherent to RFT as a key contributing factor. Our theoretical analysis suggests that RFT's gradient updates are naturally scaled by the reward variance, acting as a data-dependent regularizer that inherently protects previously acquired knowledge. Finally, we propose a rollout-based instance filtering algorithm to enhance the stability and efficiency of RFT. Our comprehensive study demonstrates the superiority of RFT as a robust paradigm for continual post-training.
Authors: Vismay Churiwala, Hardik Shukla, Manurag Khullar
Abstract: Physics-Informed Neural Networks (PINNs) have emerged as a powerful framework for solving partial differential equations (PDEs) by embedding physical constraints into the loss function. However, standard optimizers such as Adam often struggle to balance competing loss terms, particularly in stiff or ill-conditioned systems. In this work, we propose a PDE-aware optimizer that adapts parameter updates based on the variance of per-sample PDE residual gradients. This method addresses gradient misalignment without incurring the heavy computational costs of second-order optimizers such as SOAP. We benchmark the PDE-aware optimizer against Adam and SOAP on 1D Burgers', Allen-Cahn and Korteweg-de Vries(KdV) equations. Across both PDEs, the PDE-aware optimizer achieves smoother convergence and lower absolute errors, particularly in regions with sharp gradients. Our results demonstrate the effectiveness of PDE residual-aware adaptivity in enhancing stability in PINNs training. While promising, further scaling on larger architectures and hardware accelerators remains an important direction for future research.
Authors: Rui Zhu, Yuexing Peng, George C. Alexandropoulos, Peng Wang, Wenbo Wang, Wei Xiang
Abstract: Conventional computational electromagnetics (CEM) solvers can deliver high fidelity radar cross section (RCS) signatures by first solving the induced surface currents on 3-dimensional (3D) targets and then evaluating the scattered fields via radiation integrals. However, their computational cost becomes prohibitive for repeated queries and large-scale 3D scenarios. Recent purely data-driven networks improve efficiency, yet they often bypass this scattering mechanism, which may compromise physical consistency and generalization. To bridge this gap, in this paper, we propose U-PINet, a fully end-to-end, physics-informed hierarchical network for efficient RCS prediction via 3D electromagnetic scattering reconstruction. Once the scattering quantities are reconstructed, scattered fields and RCS can be evaluated for arbitrary observation directions via the radiation integral. U-PINet explicitly learns physics-consistent intermediate scattering representations by modeling local electromagnetic coupling and long-range radiation effects through a hierarchical operator design inspired by near-far field decomposition in fast solvers. A physics-guided graph neural network is incorporated to capture self- and mutual-coupling among mesh elements of complex targets, enabling physically interpretable intermediate representations. By embedding governing equations as residual constraints, U-PINet enables accurate object reconstruction of scattering quantities and consequently reliable RCS prediction across observation directions, while significantly reducing runtime. Extensive numerical experiments demonstrate that U-PINet achieves EM-solver-level RCS accuracy and 3D object reconstruction with orders-of-magnitude speedups, and generalizes well to unseen geometries under limited training data.
Authors: Amine Andam, Jamal Bentahar, Mustapha Hedabou
Abstract: Collaborative multi-agent reinforcement learning has rapidly evolved, offering state-of-the-art algorithms for real-world applications, including sensitive domains. However, a key challenge to its widespread adoption is the lack of a thorough investigation into its vulnerabilities to adversarial attacks. Existing work predominantly focuses on training-time attacks or unrealistic scenarios, such as access to policy weights or the ability to train surrogate policies. In this paper, we investigate new vulnerabilities under more challenging and constrained conditions, assuming an adversary can only collect and perturb the observations of deployed agents. We also consider scenarios where the adversary has no access at all (no observations, actions, or weights). Our main approach is to generate perturbations that intentionally misalign how victim agents see their environment. Our approach is empirically validated on three benchmarks and 22 environments, demonstrating its effectiveness across diverse algorithms and environments. Furthermore, we show that our algorithm is sample-efficient, requiring only 1,000 samples compared to the millions needed by previous methods.
Authors: Antoine Oriou, Philipp Krah, Julian Koellermeier
Abstract: This paper introduces the Intrinsic Dimension Estimating Autoencoder (IDEA), which identifies the underlying intrinsic dimension of a wide range of datasets whose samples lie on either linear or nonlinear manifolds. Beyond estimating the intrinsic dimension, IDEA is also able to reconstruct the original dataset after projecting it onto the corresponding latent space, which is structured using re-weighted double CancelOut layers. Our key contribution is the introduction of the projected reconstruction loss term, guiding the training of the model by continuously assessing the reconstruction quality under the removal of an additional latent dimension. We first assess the performance of IDEA on a series of theoretical benchmarks to validate its robustness. These experiments allow us to test its reconstruction ability and compare its performance with state-of-the-art intrinsic dimension estimators. The benchmarks show good accuracy and high versatility of our approach. Subsequently, we apply our model to data generated from the numerical solution of a vertically resolved one-dimensional free-surface flow, following a pointwise discretization of the vertical velocity profile in the horizontal direction, vertical direction, and time. IDEA succeeds in estimating the dataset's intrinsic dimension and then reconstructs the original solution by working directly within the projection space identified by the network.
Authors: Nour Jamoussi, Giuseppe Serra, Photios A. Stavrou, Marios Kountouris
Abstract: Bayesian Federated Learning (BFL) combines uncertainty modeling with decentralized training, enabling the development of personalized and reliable models under data heterogeneity and privacy constraints. Existing approaches typically rely on Markov Chain Monte Carlo (MCMC) sampling or variational inference, often incorporating personalization mechanisms to better adapt to local data distributions. In this work, we propose an information-geometric projection framework for personalization in parametric BFL. By projecting the global model onto a neighborhood of the user's local model, our method enables a tunable trade-off between global generalization and local specialization. Under mild assumptions, we show that this projection step is equivalent to computing a barycenter on the statistical manifold, allowing us to derive closed-form solutions and achieve cost-free personalization. We apply the proposed approach to a variational learning setup using the Improved Variational Online Newton (IVON) optimizer and extend its application to general aggregation schemes in BFL. Empirical evaluations under heterogeneous data distributions confirm that our method effectively balances global and local performance with minimal computational overhead.
Authors: Keanu Sisouk, Eloi Tanguy, Julie Delon, Julien Tierny
Abstract: This short paper presents a general approach for computing robust Wasserstein barycenters of persistence diagrams. The classical method consists in computing assignment arithmetic means after finding the optimal transport plans between the barycenter and the persistence diagrams. However, this procedure only works for the transportation cost related to the $q$-Wasserstein distance $W_q$ when $q=2$. We adapt an alternative fixed-point method to compute a barycenter diagram for generic transportation costs ($q > 1$), in particular those robust to outliers, $q \in (1,2)$. We show the utility of our work in two applications: \emph{(i)} the clustering of persistence diagrams on their metric space and \emph{(ii)} the dictionary encoding of persistence diagrams. In both scenarios, we demonstrate the added robustness to outliers provided by our generalized framework. Our Python implementation is available at this address: https://github.com/Keanu-Sisouk/RobustBarycenter .
Authors: Marko Tuononen, Heikki Penttinen, Ville Hautam\"aki
Abstract: We present the first use of influence functions for deep learning-based wireless receivers. Applied to DeepRx, a fully convolutional receiver, influence analysis reveals which training samples drive bit predictions, enabling targeted fine-tuning of poorly performing cases. We show that loss-relative influence with capacity-like binary cross-entropy loss and first-order updates on beneficial samples most consistently improves bit error rate toward genie-aided performance, outperforming random fine-tuning in single-target scenarios. Multi-target adaptation proved less effective, underscoring open challenges. Beyond experiments, we connect influence to self-influence corrections and propose a second-order, influence-aligned update strategy. Our results establish influence functions as both an interpretability tool and a basis for efficient receiver adaptation.
Authors: Tianyi Chen, Pengxiao Lin, Zhiwei Wang, Zhi-Qin John Xu
Abstract: State Space Models (SSMs) have emerged as promising alternatives to attention mechanisms, with the Mamba architecture demonstrating impressive performance and linear complexity for processing long sequences. However, the fundamental differences between Mamba and Transformer architectures remain incompletely understood. In this work, we use carefully designed synthetic tasks to reveal Mamba's inherent limitations. Through experiments, we identify that Mamba's nonlinear convolution introduces an asymmetry bias that significantly impairs its ability to recognize symmetrical patterns and relationships. Using composite function and inverse sequence matching tasks, we demonstrate that Mamba strongly favors compositional solutions over symmetrical ones and struggles with tasks requiring the matching of reversed sequences. We show these limitations stem not from the SSM module itself but from the nonlinear convolution preceding it, which fuses token information asymmetrically. These insights provide a new understanding of Mamba's constraints and suggest concrete architectural improvements for future sequence models.
Authors: Xiaocheng Fang, Jiarui Jin, Haoyu Wang, Che Liu, Jieyi Cai, Yujie Xiao, Guangkun Nie, Bo Liu, Shun Huang, Hongyan Li, Shenda Hong
Abstract: Electrocardiography (ECG) is the clinical gold standard for cardiovascular disease (CVD) assessment, yet continuous monitoring is constrained by the need for dedicated hardware and trained personnel. Photoplethysmography (PPG) is ubiquitous in wearable devices and readily scalable, but it lacks electrophysiological specificity, limiting diagnostic reliability. While generative methods aim to translate PPG into clinically useful ECG signals, existing approaches are limited by the misalignment of physiological semantics in generative models and the complexity of modeling in high-dimensional signals. To address these limitations, we propose PPGFlowECG, a two-stage framework that aligns PPG and ECG in a shared latent space using the CardioAlign Encoder and then synthesizes ECGs with latent rectified flow. We further provide a formal analysis of this coupling, showing that the CardioAlign Encoder is necessary to guarantee stable and semantically consistent ECG synthesis under our formulation. Extensive experiments on four datasets demonstrate improved synthesis fidelity and downstream diagnostic utility. These results indicate that PPGFlowECG supports scalable, wearable-first CVD screening when standard ECG acquisition is unavailable.
Authors: Xin Wang, Yu Wang, Yunchao Liu, Jens Meiler, Tyler Derr
Abstract: Ligand-based virtual screening (VS) is an essential step in drug discovery that evaluates large chemical libraries to identify compounds that potentially bind to a therapeutic target. However, VS faces three major challenges: class imbalance due to the low active rate, structural imbalance among active molecules where certain scaffolds dominate, and the need to identify structurally diverse active compounds for novel drug development. We introduce ScaffAug, a scaffold-aware VS framework that addresses these challenges through three modules. The augmentation module first generates synthetic data conditioned on scaffolds of actual hits using generative models, specifically a graph diffusion model. This helps mitigate the class imbalance and furthermore the structural imbalance, due to our proposed scaffold-aware sampling algorithm, designed to produce more samples for active molecules with underrepresented scaffolds. A model-agnostic self-training module is then used to safely integrate the generated synthetic data from our augmentation module with the original labeled data. Lastly, we introduce a reranking module that improves VS by enhancing scaffold diversity in the top recommended set of molecules, while still maintaining and even enhancing the overall general performance of identifying novel, active compounds. We conduct comprehensive computational experiments across five target classes, comparing ScaffAug against existing baseline methods by reporting the performance of multiple evaluation metrics and performing ablation studies on ScaffAug. Overall, this work introduces novel perspectives on effectively enhancing VS by leveraging generative augmentations, reranking, and general scaffold-awareness.
Authors: Manuel A. Hernandez Alonso, Michael Depass, Stephan Quessy, Ali Falaki, Soraya Rahimi, Numa Dancause, Ignasi Cos
Abstract: Electroencephalography (EEG) and local field potentials (LFP) are two widely used techniques to record electrical activity from the brain. These signals are used in both the clinical and research domains for multiple applications. However, most brain data recordings suffer from a myriad of artifacts and noise sources other than the brain itself. Thus, a major requirement for their use is proper and, given current volumes of data, a fully automatized conditioning. As a means to this end, here we introduce an unsupervised, multipurpose EEG/LFP preprocessing method, the NeuroClean pipeline. In addition to its completeness and reliability, NeuroClean is an unsupervised series of algorithms intended to mitigate reproducibility issues and biases caused by human intervention. The pipeline is designed as a five-step process, including the common bandpass and line noise filtering, and bad channel rejection. However, it incorporates an efficient independent component analysis with an automatic component rejection based on a clustering algorithm. This machine learning classifier is used to ensure that task-relevant information is preserved after each step of the cleaning process. We used several data sets to validate the pipeline. NeuroClean removed several common types of artifacts from the signal. Moreover, in the context of motor tasks of varying complexity, it yielded more than 97% accuracy (vs. a chance-level of 33.3%) in an optimized Multinomial Logistic Regression model after cleaning the data, compared to the raw data, which performed at 74% accuracy. These results show that NeuroClean is a promising pipeline and workflow that can be applied to future work and studies to achieve better generalization and performance on machine learning pipelines.
Authors: Dennis Thumm, Luis Ontaneda Mijares
Abstract: Market generators using deep generative models have shown promise for synthetic financial data generation, but existing approaches lack causal reasoning capabilities essential for counterfactual analysis and risk assessment. We propose a Time-series Neural Causal Model VAE (TNCM-VAE) that combines variational autoencoders with structural causal models to generate counterfactual financial time series while preserving both temporal dependencies and causal relationships. Our approach enforces causal constraints through directed acyclic graphs in the decoder architecture and employs the causal Wasserstein distance for training. We validate our method on synthetic autoregressive models inspired by the Ornstein-Uhlenbeck process, demonstrating superior performance in counterfactual probability estimation with L1 distances as low as 0.03-0.10 compared to ground truth. The model enables financial stress testing, scenario analysis, and enhanced backtesting by generating plausible counterfactual market trajectories that respect underlying causal mechanisms.
Authors: Zhuoyun Du, Runze Wang, Huiyu Bai, Zouying Cao, Xiaoyong Zhu, Yu Cheng, Bo Zheng, Wei Chen, Haochao Ying
Abstract: While natural language is the de facto communication medium for LLM-based agents, it presents a fundamental constraint. The process of downsampling rich, internal latent states into discrete tokens inherently limits the depth and nuance of information that can be transmitted, thereby hindering collaborative problem-solving. Inspired by telepathy, which bypasses symbolic language in communication, we propose Interlat (Inter-agent Latent Space Communication), a paradigm that leverages the continuous last hidden states of an LLM as a representation of its thought for direct communication (termed latent communication). An additional learned compression process further compresses latent communication via latent space reasoning. Experiments demonstrate that Interlat outperforms both fine-tuned chain-of-thought (CoT) prompting and single-agent baselines, even across heterogeneous models, promoting more exploratory behavior and enabling genuine utilization of latent information. Further compression not only substantially accelerates inference by up to 24 times but also maintains competitive performance through an efficient information-preserving mechanism. We position this work as a feasibility study of entirely latent space inter-agent communication, and our results highlight its potential, offering valuable insights for future research.
Authors: \c{C}a\u{g}r{\i} Eser, Zeynep Sonat Baltac{\i}, Emre Akba\c{s}, Sinan Kalkan
Abstract: Imbalance in classification tasks is commonly quantified by the cardinalities of examples across classes. This, however, disregards the presence of redundant examples and inherent differences in the learning difficulties of classes. Alternatively, one can use complex measures such as training loss and uncertainty, which, however, depend on training a machine learning model. Our paper proposes using data Intrinsic Dimensionality (ID) as an easy-to-compute, model-free measure of imbalance that can be seamlessly incorporated into various imbalance mitigation methods. Our results across five different datasets with a diverse range of imbalance ratios show that ID consistently outperforms cardinality-based re-weighting and re-sampling techniques used in the literature. Moreover, we show that combining ID with cardinality can further improve performance. Our code and models are available at https://github.com/cagries/IDIM.
Authors: Chengdong Ma, Th\'eo Tao Zhaowei, Pengyu Li, Minghao Liu, Haojun Chen, Zihao Mao, Yuan Cheng, Yuan Qi, Yaodong Yang
Abstract: Since Isaac Newton first studied the Kissing Number Problem in 1694, determining the maximal number of non-overlapping spheres around a central sphere has remained a fundamental challenge. This problem represents the local analogue of Hilbert's 18th problem on sphere packing, bridging geometry, number theory, and information theory. Although significant progress has been made through lattices and codes, the irregularities of high-dimensional geometry and exponentially growing combinatorial complexity beyond 8 dimensions, which exceeds the complexity of Go game, limit the scalability of existing methods. Here we model this problem as a two-player matrix completion game that can be fully parallelized at large scale, and train the game-theoretic reinforcement learning system, PackingStar, to efficiently explore high-dimensional spaces. The matrix entries represent pairwise cosines of sphere center vectors; one player fills entries while another corrects suboptimal ones, jointly maximizing the matrix size, corresponding to the kissing number. This cooperative dynamics substantially improves sample quality, making the extremely large spaces tractable. PackingStar reproduces previous configurations and surpasses all human-known records from dimensions 25 to 31, with the configuration in 25 dimensions geometrically corresponding to the Leech lattice and suggesting possible optimality. It achieves the first breakthrough beyond rational structures from 1971 in 13 dimensions, discovers over 6000 new structures in 14 and other dimensions, and establishes new records for generalized kissing configurations under various angular constraints. These results demonstrate AI's power to explore high-dimensional spaces beyond human intuition and open new pathways for the Kissing Number Problem and broader geometry problems.
Authors: Mohammad Pivezhandi, Mahdi Banisharif, Saeed Bakhshan, Abusayeed Saifullah, Ali Jannesari
Abstract: Autonomous AI agents on embedded platforms require real-time, risk-aware scheduling under resource and thermal constraints. Classical heuristics struggle with workload irregularity, tabular regressors discard structural information, and model-free reinforcement learning (RL) risks overheating. We introduce GraphPerf-RT, a graph neural network surrogate achieving deep learning accuracy at heuristic speeds (2-7ms). GraphPerf-RT is, to our knowledge, the first to unify task DAG topology, CFG-derived code semantics, and runtime context (per-core DVFS, thermal state, utilization) in a heterogeneous graph with typed edges encoding precedence, placement, and contention. Evidential regression with Normal-Inverse-Gamma priors provides calibrated uncertainty; we validate on makespan prediction for risk-aware scheduling. Experiments on three ARM platforms (Jetson TX2, Orin NX, RUBIK Pi) achieve R^2 = 0.81 on log-transformed makespan with Spearman rho = 0.95 and conservative uncertainty calibration (PICP = 99.9% at 95% confidence). Integration with four RL methods demonstrates that multi-agent model-based RL with GraphPerf-RT as the world model achieves 66% makespan reduction and 82% energy reduction versus model-free baselines, with zero thermal violations.
Authors: Haidong Hu, Xiaoyu Zheng, Jin Zhou, Yingxu Wang, Rui Wang, Pei Dong, Shiyuan Han, Lin Wang, C. L. Philip Chen, Tong Zhang, Yuehui Chen
Abstract: Deep clustering methods typically rely on a single, well-defined representation for clustering. In contrast, pretrained diffusion models provide abundant and diverse multi-scale representations across network layers and noise timesteps. However, a key challenge is how to efficiently identify the most clustering-friendly representation in the layer*timestep space. To address this issue, we propose Diffusion Embedded Clustering (DiEC), an unsupervised framework that performs clustering by leveraging optimal intermediate representations from pretrained diffusion models. DiEC systematically evaluates the clusterability of representations along the trajectory of network depth and noise timesteps. Meanwhile, an unsupervised search strategy is designed for recognizing the Clustering-optimal Layer (COL) and Clustering-optimal Timestep (COT) in the layer*timestep space of pretrained diffusion models, aiming to promote clustering performance and reduce computational overhead. DiEC is fine-tuned primarily with a structure-preserving DEC-style KL-divergence objective at the fixed COL + COT, together with a random-timestep diffusion denoising objective to maintain the generative capability of the pretrained model. Without relying on augmentation-based consistency constraints or contrastive learning, DiEC achieves excellent clustering performance across multiple benchmark datasets.
Authors: Deepit Sapru
Abstract: This paper presents a marketing analytics framework that operationalizes subscription pricing as a dynamic, guardrailed decision system, uniting multivariate demand forecasting, segment-level price elasticity, and churn propensity to optimize revenue, margin, and retention. The approach blends seasonal time-series models with tree-based learners, runs Monte Carlo scenario tests to map risk envelopes, and solves a constrained optimization that enforces business guardrails on customer experience, margin floors, and allowable churn. Validated across heterogeneous SaaS portfolios, the method consistently outperforms static tiers and uniform uplifts by reallocating price moves toward segments with higher willingness-to-pay while protecting price-sensitive cohorts. The system is designed for real-time recalibration via modular APIs and includes model explainability for governance and compliance. Managerially, the framework functions as a strategy playbook that clarifies when to shift from flat to dynamic pricing, how to align pricing with CLV and MRR targets, and how to embed ethical guardrails, enabling durable growth without eroding customer trust.
Authors: Dehao Yuan, Tyler Farnan, Stefan Tesliuc, Doron L Bergman, Yulun Wu, Xiaoyu Liu, Minghui Liu, James Montgomery, Nam H Nguyen, C. Bayan Bruss, Furong Huang
Abstract: Strict privacy regulations limit access to real transaction data, slowing open research in financial AI. Synthetic data can bridge this gap, but existing generators do not jointly achieve behavioral diversity and logical groundedness. Rule-driven simulators rely on hand-crafted workflows and shallow stochasticity, which miss the richness of human behavior. Learning-based generators such as GANs capture correlations yet often violate hard financial constraints and still require training on private data. We introduce PersonaLedger, a generation engine that uses a large language model conditioned on rich user personas to produce diverse transaction streams, coupled with an expert configurable programmatic engine that maintains correctness. The LLM and engine interact in a closed loop: after each event, the engine updates the user state, enforces financial rules, and returns a context aware "nextprompt" that guides the LLM toward feasible next actions. With this engine, we create a public dataset of 30 million transactions from 23,000 users and a benchmark suite with two tasks, illiquidity classification and identity theft segmentation. PersonaLedger offers a realistic, privacy preserving resource that supports rigorous evaluation of forecasting and anomaly detection models. PersonaLedger offers the community a rich, realistic, and privacy preserving resource -- complete with code, rules, and generation logs -- to accelerate innovation in financial AI and enable rigorous, reproducible evaluation.
Authors: Magnus B\"uhler, Lennart Purucker, Frank Hutter
Abstract: Fine-tuning tabular foundation models (TFMs) under data scarcity is challenging, as early stopping on even scarcer validation data often fails to capture true generalization performance. We propose CausalMixFT, a method that enhances fine-tuning robustness and downstream performance by generating structurally consistent synthetic samples using Structural Causal Models (SCMs) fitted on the target dataset. This approach augments limited real data with causally informed synthetic examples, preserving feature dependencies while expanding training diversity. Evaluated across 33 classification datasets from TabArena and over 2300 fine-tuning runs, our CausalMixFT method consistently improves median normalized ROC-AUC from 0.10 (standard fine-tuning) to 0.12, outperforming purely statistical generators such as CTGAN (-0.01), TabEBM (-0.04), and TableAugment (-0.09). Moreover, it narrows the median validation-test performance correlation gap from 0.67 to 0.30, enabling more reliable validation-based early stopping, a key step toward improving fine-tuning stability under data scarcity. These results demonstrate that incorporating causal structure into data augmentation provides an effective and principled route to fine-tuning tabular foundation models in low-data regimes.
Authors: Bodla Krishna Vamshi, Rohan Bhatnagar, Haizhao Yang
Abstract: Large language models (LLMs) frequently generate factually incorrect or unsupported content, commonly referred to as hallucinations. Prior work has explored decoding strategies, retrieval augmentation, and supervised fine-tuning for hallucination detection, while recent studies show that in-context learning (ICL) can substantially influence factual reliability. However, existing ICL demonstration selection methods often rely on surface-level similarity heuristics and exhibit limited robustness across tasks and models. We propose MB-ICL, a manifold-based demonstration sampling framework for selecting in-context demonstrations that leverages latent representations extracted from frozen LLMs. By jointly modeling local manifold structure and class-aware prototype geometry, MB-ICL selects demonstrations based on their proximity to learned prototypes rather than lexical or embedding similarity alone. Across factual verification (FEVER) and hallucination detection (HaluEval) benchmarks, MB-ICL outperforms standard ICL selection baselines in the majority of evaluated settings, with particularly strong gains on dialogue and summarization tasks. The method remains robust under temperature perturbations and model variation, indicating improved stability compared to heuristic retrieval strategies. While lexical retrieval can remain competitive in certain question-answering regimes, our results demonstrate that manifold-based prototype selection provides a reliable and training light approach for hallucination detection without modifying LLM parameters, offering a principled direction for improved ICL demonstration selection.
Authors: Mohammed Azeez Khan, Aaron D'Souza, Vijay Choyal
Abstract: Efficient materials discovery requires reducing costly first-principles calculations for training machine-learned interatomic potentials (MLIPs). We develop an active learning (AL) framework that iteratively selects informative structures from the Materials Project and Open Quantum Materials Database (OQMD) using compositional and property-based descriptors with a neural network ensemble model. Query-by-Committee enables real-time uncertainty quantification. We compare four strategies: random sampling (baseline), uncertainty-based sampling, diversity-based sampling (k-means clustering with farthest-point refinement), and a hybrid approach. Experiments across four material systems (C, Si, Fe, and TiO2) with 5 random seeds demonstrate that diversity sampling achieves competitive or superior performance, with 10.9% improvement on TiO2. Our approach achieves equivalent accuracy with 5-13% fewer labeled samples than random baselines. The complete pipeline executes on Google Colab in under 4 hours per system using less than 8 GB RAM, democratizing MLIP development for resource-limited researchers. Open-source code and configurations are available on GitHub. This multi-system evaluation provides practical guidelines for data-efficient MLIP training and highlights integration with symmetry-aware architectures as a promising future direction.
Authors: Zhipeng Zhang, Zhenjie Yao, Kai Li, Lei Yang
Abstract: Learning under unobservable feedback reliability poses a distinct challenge beyond optimization robustness: a system must decide whether to learn from an experience, not only how to learn stably. We study this setting as Epistemic Identifiability under Unobservable Reliability (EIUR), where each experience has a latent credibility, reliable and unreliable feedback can be locally indistinguishable, and data are generated in a closed loop by the learner's own evolving beliefs and actions. In EIUR, standard robust learning can converge stably yet form high-confidence, systematically wrong beliefs. We propose metacognitive regulation as a practical response: a second, introspective control loop that infers experience credibility from endogenous evidence in the learner's internal dynamics. We formalize this as a modular Monitor-Trust-Regulator (MTR) decomposition and instantiate it with self-diagnosis, which maintains a slowly varying experience-trust variable that softly modulates learning updates, without exogenous reliability labels or an explicit corruption model. Empirically, in the EIUR regimes studied here, self-diagnosis is associated with improved epistemic identifiability. In reinforcement learning, it enables calibrated skepticism and recovery under systematically corrupted rewards. In supervised learning, it exposes a critical dissociation: performance recovery does not imply epistemic recovery. Accuracy can rebound while internal belief dynamics remain locked-in by early misleading data, a failure detectable only through introspective diagnostics. Together, MTR and self-diagnosis provide an organizing abstraction and a concrete design template for intrinsic reliability assessment in autonomous learning under unobservable reliability.
Authors: Piyush Singh Pasi
Abstract: Multimodal models excel in English, supported by abundant image-text and audio-text data, but performance drops sharply for other languages due to limited multilingual multimodal resources. Existing solutions rely on machine translation, while advances in multilingual text modeling remain underutilized. We introduce M2M, a lightweight alignment method that learns only a few linear layers--using English text alone--to map multilingual text embeddings into multimodal space. Despite its simplicity, M2M matches baseline performance in English (94.9% Recall@10) and achieves strong zero-shot transfer (89.5% Recall@10 averaged across 11 languages, 10 unseen) on XTD Text-to-Image retrieval. Qualitative t-SNE visualizations show that multilingual embeddings align tightly with multimodal representations, while weight analysis reveals that the transformation reshapes embedding geometry rather than performing trivial rotations. Beyond image-text retrieval, M2M demonstrates robustness across datasets and tasks, extending to Audio-Text retrieval and Text-to-Image generation. We release code and checkpoints (https://github.com/piyushsinghpasi/M2M) along with multilingual evaluation datasets: MSCOCO Multilingual 30K (https://huggingface.co/datasets/piyushsinghpasi/mscoco-multilingual-30k), AudioCaps Multilingual (https://huggingface.co/datasets/piyushsinghpasi/audiocaps-multilingual), and Clotho Multilingual (https://huggingface.co/datasets/piyushsinghpasi/clotho-multilingual).
URLs: https://github.com/piyushsinghpasi/M2M), https://huggingface.co/datasets/piyushsinghpasi/mscoco-multilingual-30k),, https://huggingface.co/datasets/piyushsinghpasi/audiocaps-multilingual),, https://huggingface.co/datasets/piyushsinghpasi/clotho-multilingual).
Authors: Rapha\"el Razafindralambo, R\'emy Sun, Fr\'ed\'eric Precioso, Damien Garreau, Pierre-Alexandre Mattei
Abstract: Diffusion models now generate high-quality, diverse samples, with an increasing focus on more powerful models. Although ensembling is a well-known way to improve supervised models, its application to unconditional score-based diffusion models remains largely unexplored. In this work we investigate whether it provides tangible benefits for generative modelling. We find that while ensembling the scores generally improves the score-matching loss and model likelihood, it fails to consistently enhance perceptual quality metrics such as FID on image datasets. We confirm this observation across a breadth of aggregation rules using Deep Ensembles, Monte Carlo Dropout, on CIFAR-10 and FFHQ. We attempt to explain this discrepancy by investigating possible explanations, such as the link between score estimation and image quality. We also look into tabular data through random forests, and find that one aggregation strategy outperforms the others. Finally, we provide theoretical insights into the summing of score models, which shed light not only on ensembling but also on several model composition techniques (e.g. guidance).
Authors: Ho Fung Tsoi, Dylan Rankin
Abstract: Self-supervised learning is a powerful pre-training method for learning feature representations without labels, which often capture generic underlying semantics from the data and can later be fine-tuned for downstream tasks. In this work, we introduce jBOT, a pre-training method based on self-distillation for jet data from the CERN Large Hadron Collider, which combines local particle-level distillation with global jet-level distillation to learn jet representations that support downstream tasks such as anomaly detection and classification. We observe that pre-training on unlabeled jets leads to emergent semantic class clustering in the representation space. The clustering in the frozen embedding, when pre-trained on background jets only, enables anomaly detection via simple distance-based metrics, and the learned embedding can be fine-tuned for classification with improved performance compared to supervised models trained from scratch.
Authors: Wang Zixian
Abstract: Large language model alignment objectives are often presented as a collection of distinct algorithms, such as PPO, DPO, IPO, and their variants, each motivated by different derivations. In this work, we argue that this diversity obscures a simpler underlying structure. At a fundamental level, alignment objectives involve two independent design choices: (i) how training signals are sampled and weighted, and (ii) how deviations from a reference policy are geometrically penalized. Existing methods typically entangle these choices through a single divergence, most commonly the Kullback-Leibler divergence. We show that this entanglement is not merely a modeling convenience but a source of systematic instability. When the same divergence simultaneously determines sample weighting and optimization curvature, adjusting one aspect, such as exploration strength, inevitably alters the other, such as gradient geometry. This coupling is particularly problematic in preference-based reinforcement learning, where advantage signals are unbounded and high-confidence regimes are common. We propose a simple but structural remedy by formulating alignment as an orthogonal mirror descent problem, in which sampling geometry enters only as a linear driving force, while optimization geometry is determined independently by a mirror map. This perspective leads to a new alignment objective called Orthogonalized Policy Optimization (OPO), obtained by choosing a Euclidean mirror map in likelihood ratio space. The resulting objective admits a closed-form solution, linear and non-saturating gradient dynamics, and a well-conditioned trust region, while remaining fully compatible with standard large language model training pipelines.
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: Aryan Karmore
Abstract: Linear memory scaling stores $N$ independent expert weight matrices requiring $\mathcal{O}(N \cdot d^2)$ memory, which exceeds edge devices memory budget. Current compression methods like quantization, pruning and low-rank factorization reduce constant factors but leave the scaling bottleneck unresolved. We introduce ButterflyMoE, a method that treats experts not as independent weight matrices but as geometric reorientations of a unified shared quantized substrate. Diversity among experts arises from viewing different angles of shared capacity, not from redundant storage. By applying learned rotations to a shared ternary prototype, each expert yields $\mathcal{O}(d^2 + N \cdot d \log d)$ memory,sub-linear in the number of experts. The key insight: training these rotations with quantization reduces activation outliers and stabilizes extreme low bit training, where static methods collapse. Across language modeling benchmarks, ButterflyMoE achieves 150$\times$ memory reduction at 256 experts with negligible accuracy loss. ButterflyMoE allows multiple experts to fit on edge-constrained devices showing that geometric parameterization breaks linear scaling.
Authors: Jiasen Li, Yanwei Liu, Zhuoyi Shang, Xiaoyan Gu, Weiping Wang
Abstract: Graph-structured data is foundational to numerous web applications, and watermarking is crucial for protecting their intellectual property and ensuring data provenance. Existing watermarking methods primarily operate on graph structures or entangled graph representations, which compromise the transparency and robustness of watermarks due to the information coupling in representing graphs and uncontrollable discretization in transforming continuous numerical representations into graph structures. This motivates us to propose DRGW, the first graph watermarking framework that addresses these issues through disentangled representation learning. Specifically, we design an adversarially trained encoder that learns an invariant structural representation against diverse perturbations and derives a statistically independent watermark carrier, ensuring both robustness and transparency of watermarks. Meanwhile, we devise a graph-aware invertible neural network to provide a lossless channel for watermark embedding and extraction, guaranteeing high detectability and transparency of watermarks. Additionally, we develop a structure-aware editor that resolves the issue of latent modifications into discrete graph edits, ensuring robustness against structural perturbations. Experiments on diverse benchmark datasets demonstrate the superior effectiveness of DRGW.
Authors: Linrui Ma, Yufei Cui, Kai Han, Yunhe Wang
Abstract: One of the most compelling features of global discrete diffusion language models is their global bidirectional contextual capability. However, existing block-based diffusion studies tend to introduce autoregressive priors, which, while offering benefits, can cause models to lose this global coherence at the macro level. To regain global contextual understanding while preserving the advantages of the semi-autoregressive paradigm, we propose Diffusion in Diffusion, a 'draft-then-refine' framework designed to overcome the irreversibility and myopia problems inherent in block diffusion models. Our approach first employs block diffusion to generate rapid drafts using small blocks, then refines these drafts through global bidirectional diffusion with a larger bidirectional receptive field. We utilize snapshot confidence remasking to identify the most critical tokens that require modification, and apply mix-scale training to expand the block diffusion model's global capabilities. Empirical results demonstrate that our approach sets a new benchmark for discrete diffusion models on the OpenWebText dataset. Using only 26% of the fine-tuning budget of baseline models, we reduce generative perplexity from 25.7 to 21.9, significantly narrowing the performance gap with autoregressive models.
Authors: Cheol-Hui Lee, Hwa-Yeon Lee, Dong-Joo Kim
Abstract: The quality of data augmentation serves as a critical determinant for the performance of contrastive learning in EEG tasks. Although this paradigm is promising for utilizing unlabeled data, static or random augmentation strategies often fail to preserve intrinsic information due to the non-stationarity of EEG signals where statistical properties change over time. To address this, we propose RL-BioAug, a framework that leverages a label-efficient reinforcement learning (RL) agent to autonomously determine optimal augmentation policies. While utilizing only a minimal fraction (10%) of labeled data to guide the agent's policy, our method enables the encoder to learn robust representations in a strictly self-supervised manner. Experimental results demonstrate that RL-BioAug significantly outperforms the random selection strategy, achieving substantial improvements of 9.69% and 8.80% in Macro-F1 score on the Sleep-EDFX and CHB-MIT datasets, respectively. Notably, this agent mainly chose optimal strategies for each task--for example, Time Masking with a 62% probability for sleep stage classification and Crop & Resize with a 77% probability for seizure detection. Our framework suggests its potential to replace conventional heuristic-based augmentations and establish a new autonomous paradigm for data augmentation. The source code is available at https://github.com/dlcjfgmlnasa/RL-BioAug.
Authors: Ritam Raha, Rajarshi Roy, Nathana\"el Fijalkow, Daniel Neider
Abstract: Linear temporal logic (LTL) is a specification language for finite sequences (called traces) widely used in program verification, motion planning in robotics, process mining, and many other areas. We consider the problem of learning LTL formulas for classifying traces; despite a growing interest of the research community, existing solutions suffer from two limitations: they do not scale beyond small formulas, and they may exhaust computational resources without returning any result. We introduce a new algorithm addressing both issues: our algorithm is able to construct formulas an order of magnitude larger than previous methods, and it is anytime, meaning that it in most cases successfully outputs a formula, albeit possibly not of minimal size. We evaluate the performances of our algorithm using an open source implementation against publicly available benchmarks.
Authors: Soumick Chatterjee, Hadya Yassin, Florian Dubost, Andreas N\"urnberger, Oliver Speck
Abstract: Deep learning has demonstrated significant potential in medical imaging; however, the opacity of "black-box" models hinders clinical trust, while segmentation tasks typically necessitate labourious, hard-to-obtain pixel-wise annotations. To address these challenges simultaneously, this paper introduces a framework for three inherently explainable classifiers (GP-UNet, GP-ShuffleUNet, and GP-ReconResNet). By integrating a global pooling mechanism, these networks generate localisation heatmaps that directly influence classification decisions, offering inherent interpretability without relying on potentially unreliable post-hoc methods. These heatmaps are subsequently thresholded to achieve weakly-supervised segmentation, requiring only image-level classification labels for training. Validated on two datasets for multi-class brain tumour classification, the proposed models achieved a peak F1-score of 0.93. For the weakly-supervised segmentation task, a median Dice score of 0.728 (95% CI 0.715-0.739) was recorded. Notably, on a subset of tumour-only images, the best model achieved an accuracy of 98.7%, outperforming state-of-the-art glioma grading binary classifiers. Furthermore, comparative Precision-Recall analysis validated the framework's robustness against severe class imbalance, establishing a direct correlation between diagnostic confidence and segmentation fidelity. These results demonstrate that the proposed framework successfully combines high diagnostic accuracy with essential transparency, offering a promising direction for trustworthy clinical decision support. Code is available on GitHub: https://github.com/soumickmj/GPModels
Authors: Xiangyu Chang, Xi Chen, Zehua Lai, He Li, Zhihong Liu, Yichen Zhang
Abstract: With the fast development of big data, learning the optimal decision rule by recursively updating it and making online decisions has been easier than before. We study the online statistical inference of model parameters in a contextual bandit framework of sequential decision-making. We propose a general framework for an online and adaptive data collection environment that can update decision rules via weighted stochastic gradient descent. We allow different weighting schemes of the stochastic gradient and establish the asymptotic normality of the parameter estimator. Our proposed estimator significantly improves the asymptotic efficiency over the previous averaged SGD approach via inverse probability weights. We also conduct an optimality analysis on the weights in a linear regression setting. We provide a Bahadur representation of the proposed estimator and show that the remainder term in the Bahadur representation entails a slower convergence rate compared to classical SGD due to the adaptive data collection.
Authors: Marta Oliveira, Rick Wilming, Benedict Clark, C\'eline Budding, Fabian Eitel, Kerstin Ritter, Stefan Haufe
Abstract: Convolutional Neural Networks (CNNs) are frequently and successfully used in medical prediction tasks. They are often used in combination with transfer learning, leading to improved performance when training data for the task are scarce. The resulting models are highly complex and typically do not provide any insight into their predictive mechanisms, motivating the field of "explainable" artificial intelligence (XAI). However, previous studies have rarely quantitatively evaluated the "explanation performance" of XAI methods against ground-truth data, and transfer learning and its influence on objective measures of explanation performance has not been investigated. Here, we propose a benchmark dataset that allows for quantifying explanation performance in a realistic magnetic resonance imaging (MRI) classification task. We employ this benchmark to understand the influence of transfer learning on the quality of explanations. Experimental results show that popular XAI methods applied to the same underlying model differ vastly in performance, even when considering only correctly classified examples. We further observe that explanation performance strongly depends on the task used for pre-training and the number of CNN layers pre-trained. These results hold after correcting for a substantial correlation between explanation and classification performance.
Authors: Zezheng Song, Maria K. Cameron, Haizhao Yang
Abstract: Transition path theory (TPT) is a mathematical framework for quantifying rare transition events between a pair of selected metastable states $A$ and $B$. Central to TPT is the committor function, which describes the probability to hit the metastable state $B$ prior to $A$ from any given starting point of the phase space. Once the committor is computed, the transition channels and the transition rate can be readily found. The committor is the solution to the backward Kolmogorov equation with appropriate boundary conditions. However, solving it is a challenging task in high dimensions due to the need to mesh a whole region of the ambient space. In this work, we explore the finite expression method (FEX, Liang and Yang (2022)) as a tool for computing the committor. FEX approximates the committor by an algebraic expression involving a fixed finite number of nonlinear functions and binary arithmetic operations. The optimal nonlinear functions, the binary operations, and the numerical coefficients in the expression template are found via reinforcement learning. The FEX-based committor solver is tested on several high-dimensional benchmark problems. It gives comparable or better results than neural network-based solvers. Most importantly, FEX is capable of correctly identifying the algebraic structure of the solution which allows one to reduce the committor problem to a low-dimensional one and find the committor with any desired accuracy.
Authors: Gabriel Fern\'andez-Fern\'andez, Carlo Manzo, Maciej Lewenstein, Alexandre Dauphin, Gorka Mu\~noz-Gil
Abstract: Stochastic processes have found numerous applications in science, as they are broadly used to model a variety of natural phenomena. Due to their intrinsic randomness and uncertainty, they are, however, difficult to characterize. Here, we introduce an unsupervised machine learning approach to determine the minimal set of parameters required to effectively describe the dynamics of a stochastic process. Our method builds upon an extended $\beta$-variational autoencoder architecture. By means of simulated datasets corresponding to paradigmatic diffusion models, we showcase its effectiveness in extracting the minimal relevant parameters that accurately describe these dynamics. Furthermore, the method enables the generation of new trajectories that faithfully replicate the expected stochastic behavior. Overall, our approach enables the autonomous discovery of unknown parameters describing stochastic processes, hence enhancing our comprehension of complex phenomena across various fields.
Authors: Warren Buckley, Adrian Byrne, Nicholas Perello, Cyrus Cousins, Taha Yasseri, Yair Zick, Przemyslaw Grabowicz
Abstract: We propose that future AI transparency and accountability regulations are based on an open global standard for exchanging information about AI systems, which allows co-existence of potentially conflicting local regulations. Then, we discuss key components of a lightweight and effective AI transparency and/or accountability regulation. To prevent overregulation, the proposed approach encourages collaboration between regulators and industry to create a scalable and cost-efficient mutually beneficial solution. This includes using automated assessments and benchmarks with results transparently communicated through AI cards in an open AI register to facilitate meaningful public comparisons of competing AI systems. Such AI cards should report standardized measures tailored to the specific high-risk applications of AI systems and could be used for conformity assessments under AI transparency and accountability policies such as the European Union's AI Act.
Authors: Ernesto Araya, Mihai Cucuringu, Hemant Tyagi
Abstract: Given an undirected measurement graph $\mathcal{H} = ([n], \mathcal{E})$, the classical angular synchronization problem consists of recovering unknown angles $\theta_1^*,\dots,\theta_n^*$ from a collection of noisy pairwise measurements of the form $(\theta_i^* - \theta_j^*) \mod 2\pi$, for all $\{i,j\} \in \mathcal{E}$. This problem arises in a variety of applications, including computer vision, time synchronization of distributed networks, and ranking from pairwise comparisons. In this paper, we consider a dynamic version of this problem where the angles, and also the measurement graphs evolve over $T$ time points. Assuming a smoothness condition on the evolution of the latent angles, we derive three algorithms for joint estimation of the angles over all time points. Moreover, for one of the algorithms, we establish non-asymptotic recovery guarantees for the mean-squared error (MSE) under different statistical models. In particular, we show that the MSE converges to zero as $T$ increases under milder conditions than in the static setting. This includes the setting where the measurement graphs are highly sparse and disconnected, and also when the measurement noise is large and can potentially increase with $T$. We complement our theoretical results with experiments on synthetic data.
Authors: Yeasir Rayhan, Walid G. Aref
Abstract: Ever since the Dennard scaling broke down in the early 2000s and the frequency of the CPUs stalled, vendors have started to increase the core count in each CPU chip at the expense of introducing heterogeneity, thus ushering the era of NUMA and Chiplet processors. Since then, the heterogeneity in the design space of hardware has only increased to the point that DBMS performance may vary significantly up to an order of magnitude in modern servers. An important factor that affects performance includes the location of the logical cores where the DBMS queries execute, and the location where the data resides. This paper introduces P-MOSS, a learned spatial scheduling framework that schedules query execution to specific logical cores, and co-locates data on the corresponding NUMA node. For cross-hardware and workload adaptability, P-MOSS leverages core principles from Large Language Models, such as Next Token prediction, Generative Pre-training, and Fine-tuning. In the spirit of hardware-software synergy, P-MOSS guides its scheduling decision solely based on the low-level hardware statistics collected from the hardware Performance Monitoring Unit with the aid of a Decision Transformer. Experimental evaluation is performed in the context of the B$^+$-Tree index. Performance results demonstrate that P-MOSS offers an improvement of up to $6\times$ over traditional schedules in terms of query throughput.
Authors: Selim Furkan Tekin, Fatih Ilhan, Tiansheng Huang, Sihao Hu, Yichang Xu, Zachary Yahn, Ling Liu
Abstract: The alignment of pre-trained LLMs continues to draw significant attention from both industry and academia, aiming to ensure responses that are helpful, harmless, and honest. However, identifying a point in the model's representation subspace that simultaneously satisfies all these properties remains challenging. H3Fusion addresses this challenge by introducing a mixture-of-experts (MoE)-based fusion mechanism that models alignment as a controllable drift within the subspace, guided by a drift-regularization loss to balance competing alignment dimensions. Furthermore, we formulate the alignment by finding a dual objective of harnessing the distance of generated embeddings and alignment embeddings, and introduce a gating loss by canalizing the activations on the contributing experts. Extensive evaluations of three benchmark datasets show that H3Fusion is more helpful, less harmful, and more honest in three aspects: it outperforms each individually aligned model by 11.37%, and provides stronger robustness compared to the state-of-the-art LLM ensemble approaches by 13.77% and model-merging approaches by 6.18%. Code is available at https://github.com/git-disl/h3fusion.
Authors: Guanghui Wang, Krishna Acharya, Lokranjan Lakshmikanthan, Juba Ziani, Vidya Muthukumar
Abstract: We conduct a comprehensive analysis of the discrete-time exponential-weights dynamic with a constant step size on all general-sum and symmetric $2 \times 2$ normal-form games, i.e. games with $2$ pure strategies per player, and where the ensuing payoff tuple is of the form $(A,A^\top)$ (where $A$ is the $2 \times 2$ payoff matrix corresponding to the first player). Such symmetric games commonly arise in real-world interactions between 'symmetric" agents who have identically defined utility functions -- such as Bertrand competition and multi-agent performative prediction, and display a rich multiplicity of equilibria despite the seemingly simple setting. Somewhat surprisingly, we show through a first-principles analysis that the exponential weights dynamic, which is popular in online learning, converges in the last iterate for such games regardless of initialization with an appropriately chosen step size. For certain games and/or initializations, we further show that the convergence rate is in fact exponential and holds for any step size. We illustrate our theory with extensive simulations and applications to the aforementioned game-theoretic interactions. In the case of multi-agent performative prediction, we formulate a new "mortgage competition" game between lenders (i.e. banks) who interact with a population of customers, and show that it fits into our framework.
Authors: Richard Y. Zhang
Abstract: Low-rank matrix recovery is well-known to exhibit benign nonconvexity under the restricted isometry property (RIP): every second-order critical point is globally optimal, so local methods provably recover the ground truth. Motivated by the strong empirical performance of projected gradient methods for nonnegative low-rank recovery problems, we investigate whether this benign geometry persists when the factor matrices are constrained to be elementwise nonnegative. In the simple setting of a rank-1 nonnegative ground truth, we confirm that benign nonconvexity holds in the fully-observed case with RIP constant $\delta=0$. This benign nonconvexity, however, is unstable. It fails to extend to the partially-observed case with any arbitrarily small RIP constant $\delta>0$, and to higher-rank ground truths $r^{\star}>1$, regardless of how much the search rank $r\ge r^{\star}$ is overparameterized. Together, these results undermine the standard stability-based explanation for the empirical success of nonconvex methods and suggest that fundamentally different tools are needed to analyze nonnegative low-rank recovery.
Authors: Siyi Wang, Alexandre Leblanc, Paul D. McNicholas
Abstract: Cluster analysis, or clustering, plays a crucial role across numerous scientific and engineering domains. Despite the wealth of clustering methods proposed over the past decades, each method is typically designed for specific scenarios and presents certain limitations in practical applications. In this paper, we propose depth-based local center clustering (DLCC). This novel method makes use of data depth, which is known to produce a center-outward ordering of sample points in a multivariate space. However, data depth typically fails to capture the multimodal characteristics of {data}, something of the utmost importance in the context of clustering. To overcome this, DLCC makes use of a local version of data depth that is based on subsets of {data}. From this, local centers can be identified as well as clusters of varying shapes. Furthermore, we propose a new internal metric based on density-based clustering to evaluate clustering performance on {non-convex clusters}. Overall, DLCC is a flexible clustering approach that seems to overcome some limitations of traditional clustering methods, thereby enhancing data analysis capabilities across a wide range of application scenarios.
Authors: Francesco Orabona, Ryan D'Orazio
Abstract: The Polyak stepsize has been proven to be a fundamental stepsize in convex optimization, giving near optimal gradient descent rates across a wide range of assumptions. The universality of the Polyak stepsize has also inspired many stochastic variants, with theoretical guarantees and strong empirical performance. Despite the many theoretical results, our understanding of the convergence properties and shortcomings of the Polyak stepsize or its variants is both incomplete and fractured across different analyses. We propose a new, unified, and simple perspective for the Polyak stepsize and its variants as gradient descent on a surrogate loss. We show that each variant is equivalent to minimize a surrogate function with stepsizes that adapt to a guaranteed local curvature. Our general surrogate loss perspective is then used to provide a unified analysis of existing variants across different assumptions. Moreover, we show a number of negative results proving that the non-convergence results in some of the upper bounds is indeed real.
Authors: Md Jahangir Alam Khondkar, Ajan Ahmed, Stephanie Schuckers, Masudul Haider Imtiaz
Abstract: Speech enhancement, particularly denoising, is vital in improving the intelligibility and quality of speech signals for real-world applications, especially in noisy environments. While prior research has introduced various deep learning models for this purpose, many struggle to balance noise suppression, perceptual quality, and speaker-specific feature preservation, leaving a critical research gap in their comparative performance evaluation. This study benchmarks three state-of-the-art models Wave-U-Net, CMGAN, and U-Net, on diverse datasets such as SpEAR, VPQAD, and Clarkson datasets. These models were chosen due to their relevance in the literature and code accessibility. The evaluation reveals that U-Net achieves high noise suppression with SNR improvements of +71.96% on SpEAR, +64.83% on VPQAD, and +364.2% on the Clarkson dataset. CMGAN outperforms in perceptual quality, attaining the highest PESQ scores of 4.04 on SpEAR and 1.46 on VPQAD, making it well-suited for applications prioritizing natural and intelligible speech. Wave-U-Net balances these attributes with improvements in speaker-specific feature retention, evidenced by VeriSpeak score gains of +10.84% on SpEAR and +27.38% on VPQAD. This research indicates how advanced methods can optimize trade-offs between noise suppression, perceptual quality, and speaker recognition. The findings may contribute to advancing voice biometrics, forensic audio analysis, telecommunication, and speaker verification in challenging acoustic conditions.
Authors: Baturay Saglam, Paul Kassianik, Blaine Nelson, Sajana Weerawardhena, Yaron Singer, Amin Karbasi
Abstract: Understanding the latent space geometry of large language models (LLMs) is key to interpreting their behavior and improving alignment. Yet it remains unclear to what extent LLMs linearly organize representations related to semantic understanding. To explore this, we conduct a large-scale empirical study of hidden representations in 11 autoregressive models across six scientific topics. We find that high-level semantic information consistently resides in low-dimensional subspaces that form linearly separable representations across domains. This separability becomes more pronounced in deeper layers and under prompts that elicit structured reasoning or alignment behavior$\unicode{x2013}$even when surface content remains unchanged. These findings motivate geometry-aware tools that operate directly in latent space to detect and mitigate harmful and adversarial content. As a proof of concept, we train an MLP probe on final-layer hidden states as a lightweight latent-space guardrail. This approach substantially improves refusal rates on malicious queries and prompt injections that bypass both the model's built-in safety alignment and external token-level filters.
Authors: Runnan Fang, Yuan Liang, Xiaobin Wang, Jialong Wu, Shuofei Qiao, Pengjun Xie, Fei Huang, Huajun Chen, Ningyu Zhang
Abstract: Large Language Models (LLMs) based agents excel at diverse tasks, yet they suffer from brittle procedural memory that is manually engineered or entangled in static parameters. In this work, we investigate strategies to endow agents with a learnable, updatable, and lifelong procedural memory. We propose Memp that distills past agent trajectories into both fine-grained, step-by-step instructions and higher-level, script-like abstractions, and explore the impact of different strategies for Build, Retrieval, and Update of procedural memory. Coupled with a dynamic regimen that continuously updates, corrects, and deprecates its contents, this repository evolves in lockstep with new experience. Empirical evaluation on TravelPlanner and ALFWorld shows that as the memory repository is refined, agents achieve steadily higher success rates and greater efficiency on analogous tasks. Moreover, procedural memory built from a stronger model retains its value: migrating the procedural memory to a weaker model can also yield substantial performance gains. Code is available at https://github.com/zjunlp/MemP.
Authors: Fong-Chun Tsai, Kuan-Tang Huang, Bi-Cheng Yan, Tien-Hong Lo, Berlin Chen
Abstract: Automated Speaking Assessment (ASA) plays a crucial role in evaluating second-language (L2) learners proficiency. However, ASA models often suffer from class imbalance, leading to biased predictions. To address this, we introduce a novel objective for training ASA models, dubbed the Balancing Logit Variation (BLV) loss, which perturbs model predictions to improve feature representation for minority classes without modifying the dataset. Evaluations on the ICNALE benchmark dataset show that integrating the BLV loss into a celebrated text-based (BERT) model significantly enhances classification accuracy and fairness, making automated speech evaluation more robust for diverse learners.
Authors: Yue Li, Xin Yi, Dongsheng Shi, Yongyi Cui, Gerard de Melo, Linlin Wang
Abstract: Establishing reliable and verifiable fingerprinting mechanisms is fundamental to controlling the unauthorized redistribution of large language models (LLMs). However, existing approaches face two major challenges: (a) ensuring imperceptibility, including resistance to statistical identification and avoidance of accidental activation during fingerprint construction, and (b) preserving both model utility and fingerprint detectability under subsequent model modifications. To address these challenges, we propose an end-to-end fingerprinting framework with two components. First, we design a rule-based code-mixing fingerprint (CF) that maps natural-query-like prompts to multi-candidate targets, reducing accidental triggering via high-complexity code-mixing formulations. Second, we introduce Multi-Candidate Editing (MCEdit), which jointly optimizes multi-candidate targets and enforces margins between target and non-target outputs to improve post-modification detectability. Extensive experiments demonstrate that our framework provides a robust and practical solution for fingerprinting LLMs.
Authors: Gokul Karthik Kumar, Rishabh Saraf, Ludovick Lepauloux, Abdul Muneer, Billel Mokeddem, Hakim Hacid
Abstract: Large language models (LLMs) have transformed NLP, yet their integration with audio remains underexplored despite audio's centrality to human communication. We introduce Falcon3-Audio, a family of Audio-Language Models (ALMs) built on instruction-tuned LLMs and Whisper encoders. Using a remarkably small amount of public audio data, less than 30K hours (5K unique), Falcon3-Audio-7B matches the best reported performance among open-weight models on the MMAU benchmark, with a score of 64.14, matching R1-AQA, while distinguishing itself through superior data and parameter efficiency, single-stage training, and transparency. Notably, our smallest 1B model remains competitive with larger open models ranging from 2B to 13B parameters. Through extensive ablations, we find that common complexities such as curriculum learning, multiple audio encoders, and intricate cross-attention connectors are not required for strong performance, even compared to models trained on over 500K hours of data.
Authors: Youngwon Choi, Jaeyoon Jung, Hyeonyu Kim, Huu-Kim Nguyen, Hwayeon Kim
Abstract: Large Audio Language Models (LALMs) have emerged as powerful tools for speech-related tasks but remain underexplored for fine-tuning, especially with limited speech data. To bridge this gap, we systematically examine how different fine-tuning schemes including text-only, direct mixing, and curriculum learning affect spoken language understanding (SLU), focusing on scenarios where text-label pairs are abundant while paired speech-label data are limited. Results show that LALMs already achieve competitive performance with text-only fine-tuning, highlighting their strong generalization ability. Adding even small amounts of speech data (2-5%) yields substantial further gains, with curriculum learning particularly effective under scarce data. In cross-lingual SLU, combining source-language speech data with target-language text and minimal target-language speech data enables effective adaptation. Overall, this study provides practical insights into the LALM fine-tuning under realistic data constraints.
Authors: Mohammed Racim Moussa Boudjemaa, Alper Kalle, Xiaoyi Mai, Jos\'e Henrique de Morais Goulart, C\'edric F\'evotte
Abstract: Whitening is a classical technique in unsupervised learning that can facilitate estimation tasks by standardizing data. An important application is the estimation of latent variable models via the decomposition of tensors built from high-order moments. In particular, whitening orthogonalizes the means of a spherical Gaussian mixture model (GMM), thereby making the corresponding moment tensor orthogonally decomposable, hence easier to decompose. However, in the large-dimensional regime (LDR) where data are high-dimensional and scarce, the standard whitening matrix built from the sample covariance becomes ineffective because the latter is spectrally distorted. Consequently, whitened means of a spherical GMM are no longer orthogonal. Using random matrix theory, we derive exact limits for their dot products, which are generally nonzero in the LDR. As our main contribution, we then construct a corrected whitening matrix that restores asymptotic orthogonality, allowing for performance gains in spherical GMM estimation.
Authors: Krishna Vamshi Bodla, Haizhao Yang
Abstract: Since the introduction of Large Language Models (LLMs), they have been widely adopted for various tasks such as text summarization, question answering, speech-to-text translation, and more. In recent times, the use of LLMs for code generation has gained significant attention, with tools such as Cursor and Windsurf demonstrating the ability to analyze massive code repositories and recommend relevant changes. Big tech companies have also acknowledged the growing reliance on LLMs for code generation within their codebases. Although these advances significantly improve developer productivity, increasing reliance on automated code generation can proportionally increase the risk of suboptimal solutions and insecure code. Our work focuses on automatically sampling In-Context Learning (ICL) demonstrations which can improve model performance and enhance the interpretability of the generated code. Using AST-based analysis on outputs from the MBPP test set, we identify regions of code most influenced by the chosen demonstrations. In our experiments, we show that high-quality ICL demonstrations not only make outputs easier to interpret but also yield a positive performance improvement on the pass@10 metric. Conversely, poorly chosen ICL demonstrations affected the LLM performance on the pass@10 metric negatively compared to the base model. Overall, our approach highlights the importance of efficient sampling strategies for ICL, which can affect the performance of the model on any given task.
Authors: Carlos G\"uemes-Palau, Miquel Ferriol-Galm\'es, Jordi Paillisse-Vilanova, Albert L\'opez-Bresc\'o, Pere Barlet-Ros, Albert Cabellos-Aparicio
Abstract: Machine Learning (ML)-based network models provide fast and accurate predictions for complex network behaviors but require substantial training data. Collecting such data from real networks is often costly and limited, especially for critical scenarios like failures. As a result, researchers commonly rely on simulated data, which reduces accuracy when models are deployed in real environments. We propose a hybrid approach leveraging transfer learning to combine simulated and real-world data. Using RouteNet-Fermi, we show that fine-tuning a pre-trained model with a small real dataset significantly improves performance. Our experiments with OMNeT++ and a custom testbed reduce the Mean Absolute Percentage Error (MAPE) in packet delay prediction by up to 88%. With just 10 real scenarios, MAPE drops by 37%, and with 50 scenarios, by 48%.
Authors: Yujie Luo, Zhuoyun Yu, Xuehai Wang, Yuqi Zhu, Ningyu Zhang, Lanning Wei, Lun Du, Da Zheng, Huajun Chen
Abstract: Replicating AI research is a crucial yet challenging task for large language model (LLM) agents. Existing approaches often struggle to generate executable code, primarily due to insufficient background knowledge and the limitations of retrieval-augmented generation (RAG) methods, which fail to capture latent technical details hidden in referenced papers. Furthermore, previous approaches tend to overlook valuable implementation-level code signals and lack structured knowledge representations that support multi-granular retrieval and reuse. To overcome these challenges, we propose Executable Knowledge Graphs (xKG), a pluggable, paper-centric knowledge base that automatically integrates code snippets and technical insights extracted from scientific literature. When integrated into three agent frameworks with two different LLMs, xKG shows substantial performance gains (10.9% with o3-mini) on PaperBench, demonstrating its effectiveness as a general and extensible solution for automated AI research replication. Code is available at https://github.com/zjunlp/xKG.
Authors: Yihao Li, Saeed Salehi, Lyle Ungar, Konrad P. Kording
Abstract: Object binding, the brain's ability to bind the many features that collectively represent an object into a coherent whole, is central to human cognition. It groups low-level perceptual features into high-level object representations, stores those objects efficiently and compositionally in memory, and supports human reasoning about individual object instances. While prior work often imposes object-centric attention (e.g., Slot Attention) explicitly to probe these benefits, it remains unclear whether this ability naturally emerges in pre-trained Vision Transformers (ViTs). Intuitively, they could: recognizing which patches belong to the same object should be useful for downstream prediction and thus guide attention. Motivated by the quadratic nature of self-attention, we hypothesize that ViTs represent whether two patches belong to the same object, a property we term IsSameObject. We decode IsSameObject from patch embeddings across ViT layers using a quadratic similarity probe, which reaches over 90% accuracy. Crucially, this object-binding capability emerges reliably in DINO, CLIP, and ImageNet-supervised ViTs, but is markedly weaker in MAE, suggesting that binding is not a trivial architectural artifact, but an ability acquired through specific pretraining objectives. We further discover that IsSameObject is encoded in a low-dimensional subspace on top of object features, and that this signal actively guides attention. Ablating IsSameObject from model activations degrades downstream performance and works against the learning objective, implying that emergent object binding naturally serves the pretraining objective. Our findings challenge the view that ViTs lack object binding and highlight how symbolic knowledge of "which parts belong together" emerges naturally in a connectionist system.
Authors: Tim Strnad (Jo\v{z}ef Stefan Institute, Slovenia), Bla\v{z} Bertalani\v{c} (Jo\v{z}ef Stefan Institute, Slovenia), Carolina Fortuna (Jo\v{z}ef Stefan Institute, Slovenia)
Abstract: Machine learning is increasingly permeating radio-based localization services. To keep results credible and comparable, everyday workflows should make rigorous experiment specification and exact repeatability the default, without blocking advanced experimentation. However, in practice, researchers face a three-way gap that could be filled by a framework that offers (i) low coding effort for end-to-end studies, (ii) reproducibility by default, including versioned code, data, and configurations, controlled randomness, isolated runs, and recorded artifacts, and (iii) built-in extensibility so new models, metrics, and stages can be added with minimal integration effort. Existing tools rarely deliver all three for machine learning in general and localization workflows, supporting location-based services, in particular. In this paper, we introduce a low-code, configuration-first framework in which experiments are declared in human-readable configuration files, a workflow orchestrator executes standardized pipelines from data preparation to reporting, and all artifacts, such as datasets, models, metrics, and reports, are versioned. We instantiate the framework as LOCALIZE with preconfigured, versioned datasets that reduce initial setup effort and boilerplate, thereby accelerating model development and evaluation. The design, with explicit extension points, allows experts to add components without reworking the underlying infrastructure. Through a qualitative comparison and a head-to-head study against a plain Jupyter notebook baseline, we show that the framework reduces authoring effort while maintaining comparable runtime and memory behavior. Furthermore, using a example dataset, we demonstrate that scaling the training data from 1x to 10x keeps orchestration overheads bounded as data grows.
Authors: Phuc Nguyen, Josiah Couch, Rahul Bansal, Alexandra Morgan, Chris Tam, Miao Li, Rima Arnaout, Ramy Arnaout
Abstract: Shannon entropy is not the only entropy that is relevant to machine-learning datasets, nor possibly even the most important one. Traditional entropies such as Shannon entropy capture information represented by elements' frequencies but not the richer information encoded by their similarities and differences. Capturing the latter requires similarity-sensitive entropy (S-entropy). S-entropy can be measured using either the recently developed Leinster-Cobbold-Reeve framework (LCR) or the newer Vendi score (VS). This raises the practical question of which one to use: LCR or VS. Here we address this question conceptually, analytically, and experimentally, using 53 large and well-known imaging and tabular datasets. We find that LCR and VS values can differ by orders of magnitude and are complementary, except in limiting cases. We show that both LCR and VS results depend on how similarities are scaled, and introduce the notion of ``half-distance'' to parameterize this dependence. We prove that VS provides an upper bound on LCR for several values of the R\'enyi-Hill order parameter and present evidence that this bound holds for all values. We conclude that VS is preferable only when a dataset's elements can be usefully interpreted as linear combinations of a more fundamental set of ``ur-elements'' or when the system that the dataset describes has a quantum-mechanical character. In the broader case where one simply wishes to capture the rich information encoded by elements' similarities and differences as well as their frequencies, LCR is favored; nevertheless, for certain half-distances the two methods can complement each other.
Authors: Dennis Thumm
Abstract: Energy markets exhibit complex causal relationships between weather patterns, generation technologies, and price formation, with regime changes occurring continuously rather than at discrete break points. Current approaches model electricity prices without explicit causal interpretation or counterfactual reasoning capabilities. We introduce Augmented Time Series Causal Models (ATSCM) for energy markets, extending counterfactual reasoning frameworks to multivariate temporal data with learned causal structure. Our approach models energy systems through interpretable factors (weather, generation mix, demand patterns), rich grid dynamics, and observable market variables. We integrate neural causal discovery to learn time-varying causal graphs without requiring ground truth DAGs. Applied to real-world electricity price data, ATSCM enables novel counterfactual queries such as "What would prices be under different renewable generation scenarios?".
Authors: Philipp Dahlinger, Niklas Freymuth, Tai Hoang, Tobias W\"urth, Michael Volpp, Luise K\"arger, Gerhard Neumann
Abstract: Simulating object deformations is a critical challenge across many scientific domains, including robotics, manufacturing, and structural mechanics. Learned Graph Network Simulators (GNSs) offer a promising alternative to traditional mesh-based physics simulators. Their speed and inherent differentiability make them particularly well suited for applications that require fast and accurate simulations, such as robotic manipulation or manufacturing optimization. However, existing learned simulators typically rely on single-step observations, which limits their ability to exploit temporal context. Without this information, these models fail to infer, e.g., material properties. Further, they rely on auto-regressive rollouts, which quickly accumulate error for long trajectories. We instead frame mesh-based simulation as a trajectory-level meta-learning problem. Using Conditional Neural Processes, our method enables rapid adaptation to new simulation scenarios from limited initial data while capturing their latent simulation properties. We utilize movement primitives to directly predict fast, stable and accurate simulations from a single model call. The resulting approach, Movement-primitive Meta-MeshGraphNet (M3GN), provides higher simulation accuracy at a fraction of the runtime cost compared to state-of-the-art GNSs across several tasks.
Authors: Arthur Satouf, Yuxuan Zong, Habiboulaye Amadou-Boubacar, Pablo Piantanida, Benjamin Piwowarski
Abstract: Generative retrieval (GR) differs from the traditional index-then-retrieve pipeline by storing relevance in model parameters and generating retrieval cues directly from the query, but it can be brittle out of domain and expensive to scale. We introduce QueStER (QUEry SpecificaTion for gEnerative Keyword-Based Retrieval), which bridges GR and query reformulation by learning to generate explicit keyword-based search specifications. Given a user query, a lightweight LLM produces a keyword query that is executed by a standard retriever (BM25), combining the generalization benefits of generative query rewriting with the efficiency and scalability of lexical indexing. We train the rewriting policy with reinforcement learning techniques. Across in- and out-of-domain evaluations, QueStER consistently improves over BM25 and is competitive with neural IR baselines, while maintaining strong efficiency.
Authors: Mariafrancesca Patalano, Giovanna Capizzi, Kamran Paynabar
Abstract: Modern sensing technologies have enabled the collection of unstructured point cloud data (PCD) of varying sizes, which are used to monitor the geometric accuracy of 3D objects. PCD are widely applied in advanced manufacturing processes, including additive, subtractive, and hybrid manufacturing. To ensure the consistency of analysis and avoid false alarms, preprocessing steps such as registration and mesh reconstruction are commonly applied prior to monitoring. However, these steps are error-prone, time-consuming and may introduce artifacts, potentially affecting monitoring outcomes. In this paper, we present a novel registration-free approach for monitoring PCD of complex shapes, eliminating the need for both registration and mesh reconstruction. Our proposal consists of two alternative feature learning methods and a common monitoring scheme designed to handle hundreds of features. Feature learning methods leverage intrinsic geometric properties of the shape, captured via the Laplacian and geodesic distances. In the monitoring scheme, thresholding techniques are used to further select intrinsic features most indicative of potential out-of-control conditions. Numerical experiments and case studies highlight the effectiveness of the proposed approach in identifying different types of defects.
Authors: Julia Santaniello, Matthew Russell, Benson Jiang, Donatello Sassaroli, Robert Jacob, Jivko Sinapov
Abstract: Reinforcement Learning from Human Feedback (RLHF) is a methodology that aligns agent behavior with human preferences by integrating user feedback into the agent's training process. This paper introduces a framework that guides agent training through implicit neural signals, with a focus on the neural classification problem. Our work presents and releases a novel dataset of functional near-infrared spectroscopy (fNIRS) recordings collected from 25 human participants across three domains: Pick-and-Place Robot, Lunar Lander, and Flappy Bird. We train multiple classifiers to predict varying levels of agent performance (optimal, suboptimal, or worst-case) from windows of preprocessed fNIRS features, achieving an average F1 score of 67% for binary and 46% for multi-class classification across conditions and domains. We also train multiple regressors to predict the degree of deviation between an agent's chosen action and a set of near-optimal policy actions, providing a continuous measure of performance. Finally, we evaluate cross-subject generalization and show that fine-tuning pre-trained models with a small sample of subject-specific data increases average F1 scores by 17% and 41% for binary and multi-class models, respectively. Our results demonstrate that mapping implicit fNIRS signals to agent performance is feasible and can be improved, laying the foundation for future Reinforcement Learning from Neural Feedback (RLNF) systems.
Authors: Moran Yanuka, Paul Dixon, Eyal Finkelshtein, Daniel Rotman, Raja Giryes
Abstract: Speculative decoding accelerates autoregressive speech generation by letting a fast draft model propose tokens that a larger target model verifies. However, for speech LLMs that generate acoustic tokens, exact token matching is overly restrictive: many discrete tokens are acoustically or semantically interchangeable, reducing acceptance rates and limiting speedups. We introduce Principled Coarse-Graining (PCG), which verifies proposals at the level of Acoustic Similarity Groups (ASGs) derived from the target model's embedding space. By splitting each token's probability mass across the overlapping groups that contain it, we define an overlap-aware coarse-grained distribution and perform rejection sampling on the resulting group variable. This yields an exactness guarantee at the group level while allowing the accepted draft token to stand in for any member of the group in practice. On LibriTTS, PCG increases acceptance and throughput relative to standard speculative decoding and prior speech-specific relaxations while maintaining intelligibility and speaker similarity. These results suggest acoustically aware, group-level acceptance as a simple and general way to accelerate speech token generation while maintaining speech quality.
Authors: Tanishq Patil, Snigdha Sen, Kieran G. Foley, Fabrizio Fasano, Chantal M. W. Tax, Derek K. Jones, Mara Cercignani, Marco Palombo, Paddy J. Slator, Eleftheria Panagiotaki
Abstract: Diffusion MRI (dMRI) enables non-invasive assessment of prostate microstructure but conventional dMRI metrics such as the Apparent Diffusion Coefficient in multiparametric MRI and reflect a mixture of underlying tissues features rather than distinct histologic characteristics. Integrating dMRI with the compartment-based biophysical VERDICT (Vascular, Extracellular, and Restricted Diffusion for Cytometry in Tumours) framework offers richer microstructural insights, though clinical gradient systems (40-80 mT/m) often suffer from poor signal-to-noise ratio at stronger diffusion weightings due to prolonged echo times. Ultra-strong gradients (e.g., 300 mT/m) can mitigate these limitations by improving SNR and contrast-to-noise ratios. This study investigates whether physics-informed self-supervised VERDICT (ssVERDICT) fitting when combined with ultra-strong gradient data, enhances prostate microstructural characterization relative to current fitting approaches and clinical gradient systems. We developed enhanced ssVERDICT fitting approaches using dense multilayer perceptron and convolutional U-Net architectures, comparing them against non-linear least-squares (NLLS) VERDICT fitting, original ssVERDICT implementation, and Diffusion Kurtosis Imaging across clinical- to ultra-strong gradient systems. For the same ultra-strong gradient data, Dense ssVERDICT outperformed NLLS VERDICT, boosting median CNR by 47%, cutting inter-patient Coefficient of Variation by 52%, and reducing pooled $f_{ic}$ variation by 50%. Overall, Dense ssVERDICT delivered the highest CNR, the most stable parameter estimates, and the clearest tumour-normal contrast compared with conventional fitting methods and clinical gradient systems. These findings underscore that meaningful gains in non-invasive prostate cancer characterization arise from the combination of advanced gradient systems and deep learning-based modelling.
Authors: Francisco Ribeiro, Claudio Spiess, Prem Devanbu, Sarah Nadi
Abstract: Despite the effectiveness of large language models (LLMs) for code generation, they often output incorrect code. One reason is that model output probabilities are often not well-correlated with correctness, and reflect only the final output of the generation process. Inspired by findings that LLMs internally encode concepts like truthfulness, this paper explores if LLMs similarly represent code correctness. Specifically, we identify a correctness representation inside LLMs by contrasting the hidden states between pairs of correct and incorrect code for the same programming tasks. By experimenting on four LLMs, we show that exploiting this extracted correctness representation outperforms standard log-likelihood ranking, as well as verbalized model confidence. Furthermore, we explore how this internal correctness signal can be used to select higher-quality code samples, without requiring test execution. Ultimately, this work demonstrates how leveraging internal representations can enhance code generation systems and make LLMs more reliable, thus improving confidence in automatically generated code.
Authors: Yi Liu, Weixiang Han, Chengjun Cai, Xingliang Yuan, Cong Wang
Abstract: With the rise of large language models, service providers offer language models as a service, enabling users to fine-tune customized models via uploaded private datasets. However, this raises concerns about sensitive data leakage. Prior methods, relying on differential privacy within device-cloud collaboration frameworks, struggle to balance privacy and utility, exposing users to inference attacks or degrading fine-tuning performance. To address this, we propose PrivTune, an efficient and privacy-preserving fine-tuning framework via Split Learning (SL). The key idea of PrivTune is to inject crafted noise into token representations from the SL bottom model, making each token resemble the $n$-hop indirect neighbors. PrivTune formulates this as an optimization problem to compute the optimal noise vector, aligning with defense-utility goals. On this basis, it then adjusts the parameters (i.e., mean) of the $d_\chi$-Privacy noise distribution to align with the optimization direction and scales the noise according to token importance to minimize distortion. Experiments on five datasets (covering both classification and generation tasks) against three embedding inversion and three attribute inference attacks show that, using RoBERTa on the Stanford Sentiment Treebank dataset, PrivTune reduces the attack success rate to 10% with only a 3.33% drop in utility performance, outperforming state-of-the-art baselines.
Authors: Xin Huang, Jia Li, Jun Yu
Abstract: In the development of learning systems, there is an ongoing need to reconcile the strong predictive performance offered by opaque black-box models with the level of transparency required for critical applications. This work introduces a methodological framework that combines spectral representation learning with transparent statistical modeling to construct a mixture of generalized additive models (GAMs). The approach utilizes random Fourier feature embeddings to uncover locally adaptive structures within the data. High-dimensional random feature representations are compressed via principal component analysis to derive a latent space that informs a Gaussian mixture model, which performs soft clustering to partition the input space into distinct regimes. Within each cluster, a local GAM captures nonlinear univariate effects through interpretable spline-based smoothers. Numerical experiments across diverse regression benchmarks demonstrate that the proposed method consistently improves upon classical global interpretable models by effectively modeling data heterogeneity. Furthermore, the mixture-of-GAMs framework achieves performance comparable to explainable boosting machine, random forest, and multilayer perceptron on certain tasks. Overall, this construction provides a principled approach for integrating representation learning with transparent statistical modeling.
Authors: Kla Tantithamthavorn, Yaotian Zou, Andy Wong, Michael Gupta, Zhe Wang, Mike Buller, Ryan Jiang, Matthew Watson, Minwoo Jeong, Kun Chen, Ming Wu
Abstract: Large Language Models (LLMs)-powered code review automation has the potential to transform code review workflows. Despite the advances of LLM-powered code review comment generation approaches, several practical challenges remain for designing enterprise-grade code review automation tools. In particular, this paper aims at answering the practical question: how can we design a review-guided, context-aware, quality-checked code review comment generation without fine-tuning? In this paper, we present RovoDev Code Reviewer, an enterprise-grade LLM-based code review automation tool designed and deployed at scale within Atlassian's development ecosystem with seamless integration into Atlassian's Bitbucket. Through the offline, online, user feedback evaluations over a one-year period, we conclude that RovoDev Code Reviewer is effective in generating code review comments that could lead to code resolution for 38.70% (i.e., comments that triggered code changes in the subsequent commits); and offers the promise of accelerating feedback cycles (i.e., decreasing the PR cycle time by 30.8%), alleviating reviewer workload (i.e., reducing the number of human-written comments by 35.6%), and improving overall software quality (i.e., finding errors with actionable suggestions).
Authors: Sunki Hong, Jisoo Lee
Abstract: Accurate grid load forecasting is safety-critical: under-predictions risk supply shortfalls, while symmetric error metrics can mask this operational asymmetry. We introduce an operator-legible evaluation framework -- Under-Prediction Rate (UPR), tail Reserve$_{99.5}^{\%}$ requirements, and explicit inflation diagnostics (Bias$_{24h}$/OPR) -- to quantify one-sided reliability risk beyond MAPE. Using this framework, we evaluate state space models (Mamba variants) and strong baselines on a weather-aligned California Independent System Operator (CAISO) dataset spanning Nov 2023--Nov 2025 (84,498 hourly records across 5 regional transmission areas) under a rolling-origin walk-forward backtest. We develop and evaluate thermal-lag-aligned weather fusion strategies for these architectures. Our results demonstrate that standard accuracy metrics are insufficient proxies for operational safety: models with comparable MAPE can imply materially different tail reserve requirements (Reserve$_{99.5}^{\%}$). We show that explicit weather integration narrows error distributions, reducing the impact of temperature-driven demand spikes. Furthermore, while probabilistic calibration reduces large-error events, it can induce systematic schedule inflation. We introduce Bias/OPR-constrained objectives to enable auditable trade-offs between minimizing tail risk and preventing trivial over-forecasting.
Authors: Oshri Naparstek
Abstract: Standard autoregressive language models collapse uncertainty at every generation step by committing to discrete tokens through immediate sampling. This premature discretization underlies well-known failure modes, including degenerate repetition loops in greedy decoding and a heavy reliance on heuristic sampling strategies. We introduce \textbf{Token Maturation}, a continuous autoregressive framework in which tokens evolve as vector-valued trajectories prior to discretization. Rather than sampling from a categorical distribution at each step, the model resolves uncertainty through a deterministic dynamical process in embedding space, deferring discrete commitment until the representation has geometrically stabilized. We show that this formulation mitigates degeneration \emph{intrinsically}: Token Maturation generates coherent and diverse text under fully deterministic decoding (argmax), without repetition penalties, temperature scaling, or stochastic sampling. Moreover, we identify a novel convergence behavior in which token representations stabilize spatially while predictive entropy remains high, challenging the common assumption that commitment requires probability concentration. We propose continuous token dynamics with delayed commitment as an alternative formulation of autoregressive generation that exposes structural regularities obscured by immediate discretization.
Authors: Hanyu Li, Jiangshan Duo, Bofei Gao, Hailin Zhang, Sujian Li, Xiaotie Deng, Liang Zhao
Abstract: Chain-of-thought reasoning in large language models can trigger an "overthinking trap": longer rollouts raise cost and latency yet often yield unreliable accuracy gains. Existing methods use global, static controls that may suppress needed reasoning. We propose mastery-gated, sample-level, soft reinforcement learning compression that penalizes long rollouts only when the model already solves the problem and has produced a shorter rollout. Across benchmarks, it cuts response length by 20-40% with comparable or higher accuracy and generalizes across domains: a model trained on math spontaneously shortens unseen tasks (code, instruction following, general-knowledge QA) without hurting accuracy. We further show two-way transfer between non-agent CoT and tool-use agents: non-agent training reduces SWE-Bench Verified rounds by 13%, while compressing a thinking agent cuts SWE trajectories by 67% tokens and 52% rounds and shortens non-agent outputs by up to 44%. Compression is thus not cosmetic brevity, but an inherent computation policy -- what to keep, and what to forget.
Authors: Efstratios Manolakis, Christian Bongiorno, Rosario Nunzio Mantegna
Abstract: A new wave of work on covariance cleaning and nonlinear shrinkage has delivered asymptotically optimal analytical solutions for large covariance matrices. The same framework has been generalized to empirical cross-covariance matrices, whose singular value decomposition identifies canonical comovement modes between two asset sets, with singular values quantifying the strength of each mode and providing natural targets for shrinkage. Existing analytical cross-covariance cleaners are derived under strong stationarity and large-sample assumptions, and they typically rely on mesoscopic regularity conditions such as bounded spectra; macroscopic common modes (e.g., a global market factor) violate these conditions. When applied to real equity returns, where dependence structures drift over time and global modes are prominent, we find that these theoretically optimal formulas do not translate into robust out-of-sample performance. We address this gap by designing a random-matrix-inspired neural architecture that operates in the empirical singular-vector basis and learns a nonlinear mapping from empirical singular values to their corresponding cleaned values. By construction, the network can recover the analytical solution as a special case, yet it remains flexible enough to adapt to non-stationary dynamics and mode-driven distortions. Trained on a long history of equity returns, the proposed method achieves a more favorable bias-variance trade-off than purely analytical cleaners and delivers systematically lower out-of-sample cross-covariance prediction errors. Our results demonstrate that combining random-matrix theory with machine learning makes asymptotic theories practically effective in realistic time-varying markets.
Authors: Miaomiao Cai, Zhijie Zhang, Junfeng Fang, Zhiyong Cheng, Xiang Wang, Meng Wang
Abstract: Multi-behavior recommendation faces a critical challenge in practice: auxiliary behaviors (e.g., clicks, carts) are often noisy, weakly correlated, or semantically misaligned with the target behavior (e.g., purchase), which leads to biased preference learning and suboptimal performance. While existing methods attempt to fuse these heterogeneous signals, they inherently lack a principled mechanism to ensure robustness against such behavioral inconsistency. In this work, we propose Robust Multi-Behavior Recommendation towards Target Behaviors (RMBRec), a robust multi-behavior recommendation framework grounded in an information-theoretic robustness principle. We interpret robustness as a joint process of maximizing predictive information while minimizing its variance across heterogeneous behavioral environments. Under this perspective, the Representation Robustness Module (RRM) enhances local semantic consistency by maximizing the mutual information between users' auxiliary and target representations, whereas the Optimization Robustness Module (ORM) enforces global stability by minimizing the variance of predictive risks across behaviors, which is an efficient approximation to invariant risk minimization. This local-global collaboration bridges representation purification and optimization invariance in a theoretically coherent way. Extensive experiments on three real-world datasets demonstrate that RMBRec not only outperforms state-of-the-art methods in accuracy but also maintains remarkable stability under various noise perturbations. For reproducibility, our code is available at https://github.com/miaomiao-cai2/RMBRec/.
Authors: Syed Naveed Mahmood, Md. Rezaur Rahman Bhuiyan, Tasfia Zaman, Jareen Tasneem Khondaker, Md. Sameer Sakib, K. M. Shadman Wadith, Nazia Tasnim, Farig Sadeque
Abstract: Selective knowledge erasure from LLMs is critical for GDPR compliance and model safety, yet current unlearning methods conflate behavioral suppression with true knowledge removal, allowing latent capabilities to persist beneath surface-level refusals. In this work, we address this challenge by introducing Knowledge Immunization Framework (KIF), a representation-aware architecture that distinguishes genuine erasure from obfuscation by targeting internal activation signatures rather than surface outputs. Our approach combines dynamic suppression of subject-specific representations with parameter-efficient adaptation, enabling durable unlearning without full model retraining. KIF achieves near-oracle erasure (FQ approx 0.99 vs. 1.00) while preserving utility at oracle levels (MU = 0.62), effectively breaking the stability-erasure tradeoff that has constrained all prior work. We evaluate both standard foundation models (Llama and Mistral) and reasoning-prior models (Qwen and DeepSeek) across 3B to 14B parameters. Our observation shows that standard models exhibit scale-independent true erasure (<3% utility drift), while reasoning-prior models reveal fundamental architectural divergence. Our comprehensive dual-metric evaluation protocol, combining surface-level leakage with latent trace persistence, operationalizes the obfuscation - erasure distinction and enables the first systematic diagnosis of mechanism-level forgetting behavior across model families and scales.
Authors: Anthony Hur
Abstract: Accurate segmentation of brain tumors is essential for clinical diagnosis and treatment planning. Deep learning is currently the state-of-the-art for brain tumor segmentation, yet it requires either large datasets or extensive computational resources that are inaccessible in most areas. This makes the problem increasingly difficult: state-of-the-art models use thousands of training cases and vast computational power, where performance drops sharply when either is limited. The top performer in the Brats GLI 2023 competition relied on supercomputers trained on over 92,000 augmented MRI scans using an AMD EPYC 7402 CPU, six NVIDIA RTX 6000 GPUs (48GB VRAM each), and 1024GB of RAM over multiple weeks. To address this, the Karhunen--Lo\`eve Expansion (KLE) was implemented as a feature extraction step on downsampled, z-score normalized MRI volumes. Each 240$\times$240$\times$155 multi-modal scan is reduced to four $48^3$ channels and compressed into 32 KL coefficients. The resulting approximate reconstruction enables a residual-based anomaly map, which is upsampled and added as a fifth channel to a compact 3D U-Net. All experiments were run on a consumer workstation (AMD Ryzen 5 7600X CPU, RTX 4060Ti (8GB VRAM), and 64GB RAM while using far fewer training cases. This model achieves post-processed Dice scores of 0.929 (WT), 0.856 (TC), and 0.821 (ET), with HD95 distances of 2.93, 6.78, and 10.35 voxels. These results are significantly better than the winning BraTS 2023 methodology for HD95 distances and WT dice scores. This demonstrates that a KLE-based residual anomaly map can dramatically reduce computational cost and data requirements while retaining state-of-the-art performance.
Authors: Xin Xiong, Zijian Guo, Haobo Zhu, Chuan Hong, Jordan W Smoller, Tianxi Cai, Molei Liu
Abstract: Clinical AI systems frequently suffer performance decay post-deployment due to temporal data shifts, such as evolving populations, diagnostic coding updates (e.g., ICD-9 to ICD-10), and systemic shocks like the COVID-19 pandemic. Addressing this ``aging'' effect via frequent retraining is often impractical due to computational costs and privacy constraints. To overcome these hurdles, we introduce Adversarial Drift-Aware Predictive Transfer (ADAPT), a novel framework designed to confer durability against temporal drift with minimal retraining. ADAPT innovatively constructs an uncertainty set of plausible future models by combining historical source models and limited current data. By optimizing worst-case performance over this set, it balances current accuracy with robustness against degradation due to future drifts. Crucially, ADAPT requires only summary-level model estimators from historical periods, preserving data privacy and ensuring operational simplicity. Validated on longitudinal suicide risk prediction using electronic health records from Mass General Brigham (2005--2021) and Duke University Health Systems, ADAPT demonstrated superior stability across coding transitions and pandemic-induced shifts. By minimizing annual performance decay without labeling or retraining future data, ADAPT offers a scalable pathway for sustaining reliable AI in high-stakes healthcare environments.
Authors: Sharan Sahu, Cameron J. Hogan, Martin T. Wells
Abstract: While momentum-based acceleration has been studied extensively in deterministic optimization problems, its behavior in nonstationary environments -- where the data distribution and optimal parameters drift over time -- remains underexplored. We analyze the tracking performance of Stochastic Gradient Descent (SGD) and its momentum variants (Polyak heavy-ball and Nesterov) under uniform strong convexity and smoothness in varying stepsize regimes. We derive finite-time bounds in expectation and with high probability for the tracking error, establishing a sharp decomposition into three components: a transient initialization term, a noise-induced variance term, and a drift-induced tracking lag. Crucially, our analysis uncovers a fundamental trade-off: while momentum can suppress gradient noise, it incurs an explicit penalty on the tracking capability. We show that momentum can substantially amplify drift-induced tracking error, with amplification that becomes unbounded as the momentum parameter approaches one, formalizing the intuition that using 'stale' gradients hinders adaptation to rapid regime shifts. Complementing these upper bounds, we establish minimax lower bounds for dynamic regret under gradient-variation constraints. These lower bounds prove that the inertia-induced penalty is not an artifact of analysis but an information-theoretic barrier: in drift-dominated regimes, momentum creates an unavoidable 'inertia window' that fundamentally degrades performance. Collectively, these results provide a definitive theoretical grounding for the empirical instability of momentum in dynamic environments and delineate the precise regime boundaries where SGD provably outperforms its accelerated counterparts.
Authors: Jakob Kienegger, Timo Gerkmann
Abstract: Latest advances in deep spatial filtering for Ambisonics demonstrate strong performance in stationary multi-speaker scenarios by rotating the sound field toward a target speaker prior to multi-channel enhancement. For applicability in dynamic acoustic conditions with moving speakers, we propose to automate this rotary steering using an interleaved tracking algorithm conditioned on the target's initial direction. However, for nearby or crossing speakers, robust tracking becomes difficult and spatial cues less effective for enhancement. By incorporating the processed recording as additional guide into both algorithms, our novel joint autoregressive framework leverages temporal-spectral correlations of speech to resolve spatially challenging speaker constellations. Consequently, our proposed method significantly improves tracking and enhancement of closely spaced speakers, consistently outperforming comparable non-autoregressive methods on a synthetic dataset. Real-world recordings complement these findings in complex scenarios with multiple speaker crossings and varying speaker-to-array distances.
Authors: Daniel Holmberg, Ivan Zaitsev, Markku Alho, Ioanna Bouri, Fanni Franssila, Haewon Jeong, Minna Palmroth, Teemu Roos
Abstract: Hybrid-Vlasov simulations resolve ion-kinetic effects for modeling the solar wind-magnetosphere interaction, but even 5D (2D + 3V) simulations are computationally expensive. We show that graph-based machine learning emulators can learn the spatiotemporal evolution of electromagnetic fields and lower order moments of ion velocity distribution in the near-Earth space environment from four 5D Vlasiator runs performed with identical steady solar wind conditions. The initial ion number density is systematically varied, while the grid spacing is held constant, to scan the ratio of the characteristic ion skin depth to the numerical grid size. Using a graph neural network architecture operating on the 2D spatial simulation grid comprising 670k cells, we demonstrate that both a deterministic forecasting model (Graph-FM) and a probabilistic ensemble forecasting model (Graph-EFM) based on a latent variable formulation are capable of producing accurate predictions of future plasma states. A divergence penalty is incorporated during training to encourage divergence-freeness in the magnetic fields and improve physical consistency. For the probabilistic model, a continuous ranked probability score objective is added to improve the calibration of the ensemble forecasts. When trained, the emulators achieve more than two orders of magnitude speedup in generating the next time step relative to the original simulation on a single GPU compared to 100 CPUs for the Vlasiator runs, while closely matching physical magnetospheric response of the different runs. These results demonstrate that machine learning offers a way to make hybrid-Vlasov simulation tractable for real-time use while providing forecast uncertainty.
Authors: Miao Xie, Siguang Chen, Chunli Lv
Abstract: Large language models (LLMs) have become powerful and widely used systems for language understanding and generation, while multi-armed bandit (MAB) algorithms provide a principled framework for adaptive decision-making under uncertainty. This survey explores the potential at the intersection of these two fields. As we know, it is the first survey to systematically review the bidirectional interaction between large language models and multi-armed bandits at the component level. We highlight the bidirectional benefits: MAB algorithms address critical LLM challenges, spanning from pre-training to retrieval-augmented generation (RAG) and personalization. Conversely, LLMs enhance MAB systems by redefining core components such as arm definition and environment modeling, thereby improving decision-making in sequential tasks. We analyze existing LLM-enhanced bandit systems and bandit-enhanced LLM systems, providing insights into their design, methodologies, and performance. Key challenges and representative findings are identified to help guide future research. An accompanying GitHub repository that indexes relevant literature is available at https://github.com/bucky1119/Awesome-LLM-Bandit-Interaction.
URLs: https://github.com/bucky1119/Awesome-LLM-Bandit-Interaction.
Authors: Hongyu He, Shaowen Xiang, Ye Zhang, Yingtao Zhu, Jin Zhang, Hao Deng, Emily Alsentzer, Qingyu Chen, Kun-Hsing Yu, Andrew Marshall, Tingting Chen, Srinivas Anumasa, Daniel Ebner, Dean Ho, Kee Yuan Ngiam, Ching-Yu Cheng, Dianbo Liu
Abstract: Generative artificial intelligence (AI) is rapidly populating medical records with synthetic content, creating a feedback loop where future models are increasingly at risk of training on uncurated AI-generated data. However, the clinical consequences of this AI-generated data contamination remain unexplored. Here, we show that in the absence of mandatory human verification, this self-referential cycle drives a rapid erosion of pathological variability and diagnostic reliability. By analysing more than 800,000 synthetic data points across clinical text generation, vision-language reporting, and medical image synthesis, we find that models progressively converge toward generic phenotypes regardless of the model architecture. Specifically, rare but critical findings, including pneumothorax and effusions, vanish from the synthetic content generated by AI models, while demographic representations skew heavily toward middle-aged male phenotypes. Crucially, this degradation is masked by false diagnostic confidence; models continue to issue reassuring reports while failing to detect life-threatening pathology, with false reassurance rates tripling to 40%. Blinded physician evaluation confirms that this decoupling of confidence and accuracy renders AI-generated documentation clinically useless after just two generations. We systematically evaluate three mitigation strategies, finding that while synthetic volume scaling fails to prevent collapse, mixing real data with quality-aware filtering effectively preserves diversity. Ultimately, our results suggest that without policy-mandated human oversight, the deployment of generative AI threatens to degrade the very healthcare data ecosystems it relies upon.
Authors: Esteban G\'omez, Tom B\"ackstr\"om
Abstract: In speech machine learning, neural network models are typically designed by choosing an architecture with fixed layer sizes and structure. These models are then trained to maximize performance on metrics aligned with the task's objective. While the overall architecture is usually guided by prior knowledge of the task, the sizes of individual layers are often chosen heuristically. However, this approach does not guarantee an optimal trade-off between performance and computational complexity; consequently, post hoc methods such as weight quantization or model pruning are typically employed to reduce computational cost. This occurs because stochastic gradient descent (SGD) methods can only optimize differentiable functions, while factors influencing computational complexity, such as layer sizes and floating-point operations per second (FLOP/s), are non-differentiable and require modifying the model structure during training. We propose a reparameterization technique based on feature noise injection that enables joint optimization of performance and computational complexity during training using SGD-based methods. Unlike traditional pruning methods, our approach allows the model size to be dynamically optimized for a target performance-complexity trade-off, without relying on heuristic criteria to select which weights or structures to remove. We demonstrate the effectiveness of our method through three case studies, including a synthetic example and two practical real-world applications: voice activity detection and audio anti-spoofing. The code related to our work is publicly available to encourage further research.
Authors: Daniel Kyselica, Jon\'a\v{s} Herec, Oliver Kutis, Rado Pito\v{n}\'ak
Abstract: Natural disaster monitoring through continuous satellite observation requires processing multi-temporal data under strict operational constraints. This paper addresses flood detection, a critical application for hazard management, by developing an onboard change detection system that operates within the memory and computational limits of small satellites. We propose History Injection mechanism for Transformer models (HiT), that maintains historical context from previous observations while reducing data storage by over 99\% of original image size. Moreover, testing on the STTORM-CD flood dataset confirms that the HiT mechanism within the Prithvi-tiny foundation model maintains detection accuracy compared to the bitemporal baseline. The proposed HiT-Prithvi model achieved 43 FPS on Jetson Orin Nano, a representative onboard hardware used in nanosats. This work establishes a practical framework for satellite-based continuous monitoring of natural disasters, supporting real-time hazard assessment without dependency on ground-based processing infrastructure. Architecture as well as model checkpoints is available at https://github.com/zaitra/HiT-change-detection
Authors: George Mihaila
Abstract: Explainable AI (XAI) has become critical as transformer-based models are deployed in high-stakes applications including healthcare, legal systems, and financial services, where opacity hinders trust and accountability. Transformers self-attention mechanisms have proven valuable for model interpretability, with attention weights successfully used to understand model focus and behavior (Xu et al., 2015); (Wiegreffe and Pinter, 2019). However, existing attention-based explanation methods rely on manually defined aggregation strategies and fixed attribution rules (Abnar and Zuidema, 2020a); (Chefer et al., 2021), while model-agnostic approaches (LIME, SHAP) treat the model as a black box and incur significant computational costs through input perturbation. We introduce Explanation Network (ExpNet), a lightweight neural network that learns an explicit mapping from transformer attention patterns to token-level importance scores. Unlike prior methods, ExpNet discovers optimal attention feature combinations automatically rather than relying on predetermined rules. We evaluate ExpNet in a challenging cross-task setting and benchmark it against a broad spectrum of model-agnostic methods and attention-based techniques spanning four methodological families.