# Research Papers — 2025-10 · bundle page 212 of 217

> Full-text markdown of papers #10551–10600 of 10843 published in 2025-10, newest first. AI-readable mirrors of the full 10843-paper 2025-10 archive (50 papers per file).

Canonical page: https://aipod.dev/papers/archive/2025-10/page/212
[← Previous bundle](https://aipod.dev/papers/archive/2025-10/page/211.md)
[Next bundle →](https://aipod.dev/papers/archive/2025-10/page/213.md)

### Double projection for reconstructing dynamical systems: between stochastic and deterministic regimes

Authors: Viktor Sip, Martin Breyton, Spase Petkoski, Viktor Jirsa
Published: 2025-10-01
Categories: cs.LG, q-bio.QM
arXiv: https://arxiv.org/abs/2510.01089v2

Learning stochastic models of dynamical systems from observed data is of interest in many scientific fields. Here, we propose a new method for this task within the family of dynamical variational autoencoders. The proposed double projection method estimates both the system state trajectories and the noise time series from data. This approach naturally allows us to perform multi-step system evolution and to learn models with a comparatively low-dimensional state space. We evaluate the performance of the method on six benchmark problems, including both simulated and experimental data. We further illustrate the effects of the teacher forcing interval of the multi-step scheme on the nature of the internal dynamics and compare the resulting behavior to that of deterministic models of equivalent architecture.

---

### Safety Instincts: LLMs Learn to Trust Their Internal Compass for Self-Defense

Authors: Guobin Shen, Dongcheng Zhao, Haibo Tong, Jindong Li, Feifei Zhao, Yi Zeng
Published: 2025-10-01
Categories: cs.AI
arXiv: https://arxiv.org/abs/2510.01088v1

Ensuring Large Language Model (LLM) safety remains challenging due to the absence of universal standards and reliable content validators, making it difficult to obtain effective training signals. We discover that aligned models already possess robust internal safety beliefs: they consistently produce high-confidence refusals to harmful requests while exhibiting high entropy when generating potentially dangerous content. This entropy gap reveals an untapped signal--models intrinsically "know" when to refuse. We introduce Safety Instincts Reinforcement Learning (SIRL), which transforms this internal confidence into a self-generated reward signal, eliminating dependence on external validators or human annotations. SIRL teaches models to trust their safety instincts by reinforcing low-entropy refusal behaviors. Evaluated on Llama and Qwen models, SIRL maintains 89%+ Defense Success Rates (DSRs) against 20+ jailbreak methods, from static prompts to adaptive attacks. Using only 15,000 unlabeled prompts, SIRL surpasses resource-intensive supervised methods while preserving performance on mathematics, coding, and conversation benchmarks. Our work demonstrates that effective alignment can emerge from within, paving the way for more autonomous and robust AI safety mechanisms that scale without extensive human oversight.

---

### Multi-Actor Multi-Critic Deep Deterministic Reinforcement Learning with a Novel Q-Ensemble Method

Authors: Andy Wu, Chun-Cheng Lin, Rung-Tzuo Liaw, Yuehua Huang, Chihjung Kuo, Chia Tong Weng
Published: 2025-10-01
Categories: cs.LG
arXiv: https://arxiv.org/abs/2510.01083v1

Reinforcement learning has gathered much attention in recent years due to its rapid development and rich applications, especially on control systems and robotics. When tackling real-world applications with reinforcement learning method, the corresponded Markov decision process may have huge discrete or even continuous state/action space. Deep reinforcement learning has been studied for handling these issues through deep learning for years, and one promising branch is the actor-critic architecture. Many past studies leveraged multiple critics to enhance the accuracy of evaluation of a policy for addressing the overestimation and underestimation issues. However, few studies have considered the architecture with multiple actors together with multiple critics. This study proposes a novel multi-actor multi-critic (MAMC) deep deterministic reinforcement learning method. The proposed method has three main features, including selection of actors based on non-dominated sorting for exploration with respect to skill and creativity factors, evaluation for actors and critics using a quantile-based ensemble strategy, and exploiting actors with best skill factor. Theoretical analysis proves the learning stability and bounded estimation bias for the MAMC. The present study examines the performance on a well-known reinforcement learning benchmark MuJoCo. Experimental results show that the proposed framework outperforms state-of-the-art deep deterministic based reinforcement learning methods. Experimental analysis also indicates the proposed components are effective. Empirical analysis further investigates the validity of the proposed method, and shows its benefit on complicated problems. The source code can be found at https://github.com/AndyWu101/MAMC.

---

### CodeGenLink: A Tool to Find the Likely Origin and License of Automatically Generated Code

Authors: Daniele Bifolco, Guido Annicchiarico, Pierluigi Barbiero, Massimiliano Di Penta, Fiorella Zampetti
Published: 2025-10-01
Categories: cs.SE, cs.AI
arXiv: https://arxiv.org/abs/2510.01077v1

Large Language Models (LLMs) are widely used in software development tasks nowadays. Unlike reusing code taken from the Web, for LLMs' generated code, developers are concerned about its lack of trustworthiness and possible copyright or licensing violations, due to the lack of code provenance information. This paper proposes CodeGenLink, a GitHub CoPilot extension for Visual Studio Code aimed at (i) suggesting links containing code very similar to automatically generated code, and (ii) whenever possible, indicating the license of the likely origin of the code. CodeGenLink retrieves candidate links by combining LLMs with their web search features and then performs similarity analysis between the generated and retrieved code. Preliminary results show that CodeGenLink effectively filters unrelated links via similarity analysis and provides licensing information when available. Tool URL: https://github.com/danielebifolco/CodeGenLink Tool Video: https://youtu.be/M6nqjBf9_pw

---

### Research on the Integration of Embodied Intelligence and Reinforcement Learning in Textual Domains

Authors: Haonan Wang, Junfeng Sun, Mingjia Zhao, Wei Liu
Published: 2025-10-01
Categories: cs.CL
arXiv: https://arxiv.org/abs/2510.01076v1

This article addresses embodied intelligence and reinforcement learning integration in the field of text processing, aiming to enhance text handling with more intelligence on the basis of embodied intelligence's perception and action superiority and reinforcement learning's decision optimization capability. Through detailed theoretical explanation and experimental exploration, a novel integration model is introduced. This model has been demonstrated to be very effective in a wide range oftext processing tasks, validating its applicative potential

---

### Predicting Diabetic Retinopathy Using a Two-Level Ensemble Model

Authors: Mahyar Mahmoudi, Tieming Liu
Published: 2025-10-01
Categories: cs.LG
arXiv: https://arxiv.org/abs/2510.01074v1

Preprint Note: This is the author preprint version of a paper accepted for presentation at the IISE Annual Conference & Expo 2025. The final version will appear in the official proceedings. Diabetic retinopathy (DR) is a leading cause of blindness in working-age adults, and current diagnostic methods rely on resource-intensive eye exams and specialized equipment. Image-based AI tools have shown limitations in early-stage detection, motivating the need for alternative approaches. We propose a non-image-based, two-level ensemble model for DR prediction using routine laboratory test results. In the first stage, base models (Linear SVC, Random Forest, Gradient Boosting, and XGBoost) are hyperparameter tuned and internally stacked across different configurations to optimize metrics such as accuracy, recall, and precision. In the second stage, predictions are aggregated using Random Forest as a meta-learner. This hierarchical stacking strategy improves generalization, balances performance across multiple metrics, and remains computationally efficient compared to deep learning approaches. The model achieved Accuracy 0.9433, F1 Score 0.9425, Recall 0.9207, Precision 0.9653, ROC-AUC 0.9844, and AUPRC 0.9875, surpassing one-level stacking and FCN baselines. These results highlight the model potential for accurate and interpretable DR risk prediction in clinical settings.

---

### Eliciting Secret Knowledge from Language Models

Authors: Bartosz Cywiński, Emil Ryd, Rowan Wang, Senthooran Rajamanoharan, Neel Nanda, Arthur Conmy, Samuel Marks
Published: 2025-10-01
Categories: cs.LG
arXiv: https://arxiv.org/abs/2510.01070v2

We study secret elicitation: discovering knowledge that an AI possesses but does not explicitly verbalize. As a testbed, we train three families of large language models (LLMs) to possess specific knowledge that they apply downstream but deny knowing when asked directly. For example, in one setting, we train an LLM to generate replies that are consistent with knowing the user is female, while denying this knowledge when asked directly. We then design various black-box and white-box secret elicitation techniques and evaluate them based on whether they can help an LLM auditor successfully guess the secret knowledge. Many of our techniques improve on simple baselines. Our most effective techniques (performing best in all settings) are based on prefill attacks, a black-box technique where the LLM reveals secret knowledge when generating a completion from a predefined prefix. Our white-box techniques based on logit lens and sparse autoencoders (SAEs) also consistently increase the success rate of the LLM auditor, but are less effective. We release our models and code, establishing a public benchmark for evaluating secret elicitation methods.

---

### Typed Chain-of-Thought: A Curry-Howard Framework for Verifying LLM Reasoning

Authors: Elija Perrier
Published: 2025-10-01
Categories: cs.AI
arXiv: https://arxiv.org/abs/2510.01069v1

While Chain-of-Thought (CoT) prompting enhances the reasoning capabilities of large language models, the faithfulness of the generated rationales remains an open problem for model interpretability. We propose a novel theoretical lens for this problem grounded in the Curry-Howard correspondence, which posits a direct relationship between formal proofs and computer programs. Under this paradigm, a faithful reasoning trace is analogous to a well-typed program, where each intermediate step corresponds to a typed logical inference. We operationalise this analogy, presenting methods to extract and map the informal, natural language steps of CoT into a formal, typed proof structure. Successfully converting a CoT trace into a well-typed proof serves as a strong, verifiable certificate of its computational faithfulness, moving beyond heuristic interpretability towards formal verification. Our framework provides a methodology to transform plausible narrative explanations into formally verifiable programs, offering a path towards building more reliable and trustworthy AI systems.

---

### Compose Your Policies! Improving Diffusion-based or Flow-based Robot Policies via Test-time Distribution-level Composition

Authors: Jiahang Cao, Yize Huang, Hanzhong Guo, Rui Zhang, Mu Nan, Weijian Mai, Jiaxu Wang, Hao Cheng, Jingkai Sun, Gang Han, Wen Zhao, Qiang Zhang et al.
Published: 2025-10-01
Categories: cs.RO, cs.LG
arXiv: https://arxiv.org/abs/2510.01068v2

Diffusion-based models for robotic control, including vision-language-action (VLA) and vision-action (VA) policies, have demonstrated significant capabilities. Yet their advancement is constrained by the high cost of acquiring large-scale interaction datasets. This work introduces an alternative paradigm for enhancing policy performance without additional model training. Perhaps surprisingly, we demonstrate that the composed policies can exceed the performance of either parent policy. Our contribution is threefold. First, we establish a theoretical foundation showing that the convex composition of distributional scores from multiple diffusion models can yield a superior one-step functional objective compared to any individual score. A Grönwall-type bound is then used to show that this single-step improvement propagates through entire generation trajectories, leading to systemic performance gains. Second, motivated by these results, we propose General Policy Composition (GPC), a training-free method that enhances performance by combining the distributional scores of multiple pre-trained policies via a convex combination and test-time search. GPC is versatile, allowing for the plug-and-play composition of heterogeneous policies, including VA and VLA models, as well as those based on diffusion or flow-matching, irrespective of their input visual modalities. Third, we provide extensive empirical validation. Experiments on Robomimic, PushT, and RoboTwin benchmarks, alongside real-world robotic evaluations, confirm that GPC consistently improves performance and adaptability across a diverse set of tasks. Further analysis of alternative composition operators and weighting strategies offers insights into the mechanisms underlying the success of GPC. These results establish GPC as a simple yet effective method for improving control performance by leveraging existing policies.

---

### ReSWD: ReSTIR'd, not shaken. Combining Reservoir Sampling and Sliced Wasserstein Distance for Variance Reduction

Authors: Mark Boss, Andreas Engelhardt, Simon Donné, Varun Jampani
Published: 2025-10-01
Categories: cs.GR, cs.CV, cs.LG
arXiv: https://arxiv.org/abs/2510.01061v1

Distribution matching is central to many vision and graphics tasks, where the widely used Wasserstein distance is too costly to compute for high dimensional distributions. The Sliced Wasserstein Distance (SWD) offers a scalable alternative, yet its Monte Carlo estimator suffers from high variance, resulting in noisy gradients and slow convergence. We introduce Reservoir SWD (ReSWD), which integrates Weighted Reservoir Sampling into SWD to adaptively retain informative projection directions in optimization steps, resulting in stable gradients while remaining unbiased. Experiments on synthetic benchmarks and real-world tasks such as color correction and diffusion guidance show that ReSWD consistently outperforms standard SWD and other variance reduction baselines. Project page: https://reservoirswd.github.io/

---

### Hybrid Dialogue State Tracking for Persian Chatbots: A Language Model-Based Approach

Authors: Samin Mahdipour Aghabagher, Saeedeh Momtazi
Published: 2025-10-01
Categories: cs.CL, cs.AI
arXiv: https://arxiv.org/abs/2510.01052v1

Dialogue State Tracking (DST) is an essential element of conversational AI with the objective of deeply understanding the conversation context and leading it toward answering user requests. Due to high demands for open-domain and multi-turn chatbots, the traditional rule-based DST is not efficient enough, since it cannot provide the required adaptability and coherence for human-like experiences in complex conversations. This study proposes a hybrid DST model that utilizes rule-based methods along with language models, including BERT for slot filling and intent detection, XGBoost for intent validation, GPT for DST, and online agents for real-time answer generation. This model is uniquely designed to be evaluated on a comprehensive Persian multi-turn dialogue dataset and demonstrated significantly improved accuracy and coherence over existing methods in Persian-based chatbots. The results demonstrate how effectively a hybrid approach may improve DST capabilities, paving the way for conversational AI systems that are more customized, adaptable, and human-like.

---

### GEM: A Gym for Agentic LLMs

Authors: Zichen Liu, Anya Sims, Keyu Duan, Changyu Chen, Simon Yu, Xiangxin Zhou, Haotian Xu, Shaopan Xiong, Bo Liu, Chenmien Tan, Chuen Yang Beh, Weixun Wang et al.
Published: 2025-10-01
Categories: cs.LG, cs.AI, cs.CL
arXiv: https://arxiv.org/abs/2510.01051v2

The training paradigm for large language models (LLMs) is moving from static datasets to experience-based learning, where agents acquire skills via interacting with complex environments. To facilitate this transition we introduce GEM (General Experience Maker), an open-source environment simulator designed for the age of LLMs. Analogous to OpenAI-Gym for traditional reinforcement learning (RL), GEM provides a standardized framework for the environment-agent interface, including asynchronous vectorized execution for high throughput, and flexible wrappers for easy extensibility. GEM also features a diverse suite of environments, robust integrated tools, and single-file example scripts demonstrating using GEM with five popular RL training frameworks. Along with this, we also provide a set of baselines across 24 environments using REINFORCE with Return Batch Normalization (ReBN), which -- unlike GRPO -- is compatible with the full RL setting of dense per-turn rewards and offers better credit assignment. We further conduct apple-to-apple benchmarking of PPO, GRPO and REINFORCE in both single- and multi-turn settings using GEM to shed light on the algorithmic designs. Lastly, GEM also functions as a convenient evaluation toolkit besides a training environment. We hope this framework can help accelerate future agentic LLM research.

---

### KeySG: Hierarchical Keyframe-Based 3D Scene Graphs

Authors: Abdelrhman Werby, Dennis Rotondi, Fabio Scaparro, Kai O. Arras
Published: 2025-10-01
Categories: cs.CV, cs.RO
arXiv: https://arxiv.org/abs/2510.01049v2

In recent years, 3D scene graphs have emerged as a powerful world representation, offering both geometric accuracy and semantic richness. Combining 3D scene graphs with large language models enables robots to reason, plan, and navigate in complex human-centered environments. However, current approaches for constructing 3D scene graphs are semantically limited to a predefined set of relationships, and their serialization in large environments can easily exceed an LLM's context window. We introduce KeySG, a framework that represents 3D scenes as a hierarchical graph consisting of floors, rooms, objects, and functional elements, where nodes are augmented with multi-modal information extracted from keyframes selected to optimize geometric and visual coverage. The keyframes allow us to efficiently leverage VLMs to extract scene information, alleviating the need to explicitly model relationship edges between objects, enabling more general, task-agnostic reasoning and planning. Our approach can process complex and ambiguous queries while mitigating the scalability issues associated with large scene graphs by utilizing a hierarchical multi-modal retrieval-augmented generation (RAG) pipeline to extract relevant context from the graph. Evaluated across three distinct benchmarks, 3D object semantic segmentation, functional element segmentation, and complex query retrieval, KeySG outperforms prior approaches on most metrics, demonstrating its superior semantic richness and efficiency.

---

### Interpreting Language Models Through Concept Descriptions: A Survey

Authors: Nils Feldhus, Laura Kopf
Published: 2025-10-01
Categories: cs.CL, cs.AI, cs.LG
arXiv: https://arxiv.org/abs/2510.01048v1

Understanding the decision-making processes of neural networks is a central goal of mechanistic interpretability. In the context of Large Language Models (LLMs), this involves uncovering the underlying mechanisms and identifying the roles of individual model components such as neurons and attention heads, as well as model abstractions such as the learned sparse features extracted by Sparse Autoencoders (SAEs). A rapidly growing line of work tackles this challenge by using powerful generator models to produce open-vocabulary, natural language concept descriptions for these components. In this paper, we provide the first survey of the emerging field of concept descriptions for model components and abstractions. We chart the key methods for generating these descriptions, the evolving landscape of automated and human metrics for evaluating them, and the datasets that underpin this research. Our synthesis reveals a growing demand for more rigorous, causal evaluation. By outlining the state of the art and identifying key challenges, this survey provides a roadmap for future research toward making models more transparent.

---

### In-Situ Tweedie Discrete Diffusion Models

Authors: Xiao Li, Jiaqi Zhang, Shuxiang Zhang, Tianshui Chen, Liang Lin, Guangrun Wang
Published: 2025-10-01
Categories: cs.CV, cs.AI, cs.CL, cs.LG
arXiv: https://arxiv.org/abs/2510.01047v2

While diffusion models excel at generating continuous data such as images, adapting them to discrete tasks has relied on indirect approaches that either operate in continuous embedding spaces or use token masking mechanisms, both of which deviate from modeling the true discrete data distribution that can be theoretically guaranteed by Tweedie's formula. We propose in-situ Tweedie Discrete Diffusion (TDD), a framework that performs diffusion guaranteed by Tweedie's formula directly within the discrete one-hot space, hence "in-situ." Unlike prior methods that diffuse continuous embeddings or mask tokens, TDD directly corrupts one-hot vectors with Gaussian noise and performs iterative denoising through a timestep-conditioned cross-entropy objective rather than mean-squared-error reconstruction. At each denoising step, the model predicts class probabilities, applies argmax to obtain discrete predictions, converts them to one-hot vectors, and feeds them into the next iteration with progressively reduced noise. This process naturally unifies discriminative classification and generative modeling under a single framework. Experiments demonstrate that TDD achieves strong performance on both image classification and text generation tasks, with extensive ablation studies confirming the effectiveness of each design component. Our work establishes a principled approach to discrete diffusion that preserves the core characteristics of diffusion models while operating natively in discrete space.

---

### Gated X-TFC: Soft Domain Decomposition for Forward and Inverse Problems in Sharp-Gradient PDEs

Authors: Vikas Dwivedi, Enrico Schiassi, Monica Sigovan, Bruno Sixou
Published: 2025-10-01
Categories: cs.LG
arXiv: https://arxiv.org/abs/2510.01039v1

Physics-informed neural networks (PINNs) and related methods struggle to resolve sharp gradients in singularly perturbed boundary value problems without resorting to some form of domain decomposition, which often introduce complex interface penalties. While the Extreme Theory of Functional Connections (X-TFC) avoids multi-objective optimization by employing exact boundary condition enforcement, it remains computationally inefficient for boundary layers and incompatible with decomposition. We propose Gated X-TFC, a novel framework for both forward and inverse problems, that overcomes these limitations through a soft, learned domain decomposition. Our method replaces hard interfaces with a differentiable logistic gate that dynamically adapts radial basis function (RBF) kernel widths across the domain, eliminating the need for interface penalties. This approach yields not only superior accuracy but also dramatic improvements in computational efficiency: on a benchmark one dimensional (1D) convection-diffusion, Gated X-TFC achieves an order-of-magnitude lower error than standard X-TFC while using 80 percent fewer collocation points and reducing training time by 66 percent. In addition, we introduce an operator-conditioned meta-learning layer that learns a probabilistic mapping from PDE parameters to optimal gate configurations, enabling fast, uncertainty-aware warm-starting for new problem instances. We further demonstrate scalability to multiple subdomains and higher dimensions by solving a twin boundary-layer equation and a 2D Poisson problem with a sharp Gaussian source. Overall, Gated X-TFC delivers a simple alternative alternative to PINNs that is both accurate and computationally efficient for challenging boundar-layer regimes. Future work will focus on nonlinear problems.

---

### Activation-Deactivation: A General Framework for Robust Post-hoc Explainable AI

Authors: Akchunya Chanchal, David A. Kelly, Hana Chockler
Published: 2025-10-01
Categories: cs.AI, cs.CV, cs.LG
arXiv: https://arxiv.org/abs/2510.01038v2

Perturbation-based explainability methods face criticism due to their reliance on out-of-distribution mutants. This raises doubts about the quality of the explanations. In this paper, we introduce a novel forward pass paradigm, Activation-Deactivation (AD), which obviates the need for perturbation of the input. AD replaces perturbation of input features with switching off parts of the model corresponding to to the intended perturbations. We implement ConvAD, an AD approximation algorithm for CNNs. ConvAD is a drop-in mechanism that can be easily added to any trained CNN and, without any additional training, generates more robust and more transferable explanations. We provide evaluation results across multiple architectures, datasets, methods and perturbation strategies, demonstrating the superior quality of ConvAD compared to the SOTA.

---

### CurES: From Gradient Analysis to Efficient Curriculum Learning for Reasoning LLMs

Authors: Yongcheng Zeng, Zexu Sun, Bokai Ji, Erxue Min, Hengyi Cai, Shuaiqiang Wang, Dawei Yin, Haifeng Zhang, Xu Chen, Jun Wang
Published: 2025-10-01
Categories: cs.LG, cs.AI
arXiv: https://arxiv.org/abs/2510.01037v2

Curriculum learning plays a crucial role in enhancing the training efficiency of large language models (LLMs) on reasoning tasks. However, existing methods often fail to adequately account for variations in prompt difficulty or rely on simplistic filtering mechanisms to select prompt datasets within a narrow criterion range, resulting in significant computational waste. In this work, we approach the problem from the perspective of reinforcement learning gradient optimization, offering a systematic and theoretical investigation into how to improve the training efficiency of LLMs. We identify two key factors influencing training efficiency: the selection of training prompts and the allocation of rollout quantities across different prompts. Our theoretical analysis reveals that the sampling distribution of prompts dictates the convergence rate of gradient descent, while the allocation of the rollout quantity influences the consistency and stability of overall gradient updates. Based on these insights, we propose CurES, an efficient training method that accelerates convergence and employs Bayesian posterior estimation to minimize computational overhead. Experiments demonstrate that our CurES outperforms Group Relative Policy Optimization (GRPO) by +3.30 points and +4.82 points with 1.5B and 7B models, respectively, and exceeds the best prior sample efficient methods by +2.12 points on average across eight math reasoning benchmarks. Additionally, CurES exhibits faster convergence compared to baselines, including GRPO.

---

### Meaningless Tokens, Meaningful Gains: How Activation Shifts Enhance LLM Reasoning

Authors: Zeru Shi, Yingjia Wan, Zhenting Wang, Qifan Wang, Fan Yang, Elisa Kreiss, Ruixiang Tang
Published: 2025-10-01
Categories: cs.LG
arXiv: https://arxiv.org/abs/2510.01032v1

Motivated by the puzzling observation that inserting long sequences of meaningless tokens before the query prompt can consistently enhance LLM reasoning performance, this work analyzes the underlying mechanism driving this phenomenon and based on these insights proposes a more principled method that allows for similar performance gains. First, we find that the improvements arise from a redistribution of activations in the LLM's MLP layers, where near zero activations become less frequent while large magnitude activations increase. This redistribution enhances the model's representational capacity by suppressing weak signals and promoting stronger, more informative ones. Building on this insight, we propose the Activation Redistribution Module (ARM), a lightweight inference-time technique that modifies activations directly without altering the input sequence. ARM adaptively identifies near-zero activations after the non-linear function and shifts them outward, implicitly reproducing the beneficial effects of meaningless tokens in a controlled manner. Extensive experiments across diverse benchmarks and model architectures clearly show that ARM consistently improves LLM performance on reasoning tasks while requiring only a few lines of simple code to implement. Our findings deliver both a clear mechanistic explanation for the unexpected benefits of meaningless tokens and a simple yet effective technique that harnesses activation redistribution to further improve LLM performance.

---

### Secure and reversible face anonymization with diffusion models

Authors: Pol Labarbarie, Vincent Itier, William Puech
Published: 2025-10-01
Categories: cs.CV, cs.LG
arXiv: https://arxiv.org/abs/2510.01031v2

Face anonymization aims to protect sensitive identity information by altering faces while preserving visual realism and utility for downstream computer vision tasks. Current methods struggle to simultaneously ensure high image quality, strong security guarantees, and controlled reversibility for authorized identity recovery at a later time. To improve the image quality of generated anonymized faces, recent methods have adopted diffusion models. However, these new diffusion-based anonymization methods do not provide a mechanism to restrict de-anonymization to trusted parties, limiting their real-world applicability. In this paper, we present the first diffusion-based framework for secure, reversible face anonymization via secret-key conditioning. Our method injects the secret key directly into the diffusion process, enabling anonymization and authorized face reconstruction while preventing unauthorized de-anonymization. The use of deterministic forward and reverse diffusion steps guarantees exact identity recovery when the correct secret key is available. Experiments on CelebA-HQ and LFW demonstrate that our approach achieves better anonymization and de-anonymization capabilities than prior work. We also show that our method remains robust to incorrect or adversarial key de-anonymization. Our code will be made publicly available.

---

### Uncovering the Computational Ingredients of Human-Like Representations in LLMs

Authors: Zach Studdiford, Timothy T. Rogers, Kushin Mukherjee, Siddharth Suresh
Published: 2025-10-01
Categories: cs.AI
arXiv: https://arxiv.org/abs/2510.01030v2

The human ability to translate diverse perceptual and linguistic inputs into structured behavior has been thought to rest on learning robust representations of concepts. The rapid advancement of transformer-based large language models (LLMs) has surfaced a diversity of computational ingredients relevant for model building - architectures, fine-tuning methods, and training datasets among others - yet it remains unclear which are most crucial for developing human-like conceptual representations. Further, most current benchmarks are ill-suited to measuring representational alignment, making LLMs' scores on them unreliable for assessing whether they are progressing as cognitive models. We address these limitations by evaluating over 75 models on a triplet similarity task, a method well established in cognitive science for measuring conceptual representations, using concepts from the THINGS database. We find that instruction fine-tuning and larger attention head dimensionality are among the strongest predictors of human alignment, while activation function choice, multimodal pretraining, and parameter size have limited influence on alignment. Correlations between alignment scores and existing benchmark scores reveal that while some benchmarks (e.g., BigBenchHard) better capture representational alignment than others (e.g., MUSR), none fully accounts for the variance in human-model alignment, demonstrating their insufficiency. Taken together, our findings highlight key computational ingredients for advancing LLMs as models of human conceptual representation and address a key gap in LLM evaluation.

---

### Syntax-Guided Diffusion Language Models with User-Integrated Personalization

Authors: Ruqian Zhang, Yijiao Zhang, Juan Shen, Zhongyi Zhu, Annie Qu
Published: 2025-10-01
Categories: cs.CL, stat.ME
arXiv: https://arxiv.org/abs/2510.01028v2

Large language models have made revolutionary progress in generating human-like text, yet their outputs often tend to be generic, exhibiting insufficient structural diversity, which limits personalized expression. Recent advances in diffusion models have opened new opportunities for improving language generation beyond the limitations of autoregressive paradigms. In this work, we propose a syntax-guided diffusion language model that integrates structural supervision and personalized conditioning to enhance text quality, diversity, and controllability. We introduce a cascaded framework that generates syntactic guidance before conditional text generation, and further generalize it to a novel noncascaded architecture for better alignment between structure and content. By incorporating syntactic information in the generating process, the proposed model better captures the lexical and structural characteristics of stylistic sentence construction. To enable fine-grained personalization, we develop a shared representation mechanism that facilitates information integration across users, supporting both faithful stylistic generation and generalizable zero-shot inference. Extensive experiments on multiple tasks demonstrate the superiority of our approach in fluency, diversity, and stylistic fidelity. Further qualitative analyses highlight its interpretability and flexibility in learning personalized patterns.

---

### Hypothesis-Driven Feature Manifold Analysis in LLMs via Supervised Multi-Dimensional Scaling

Authors: Federico Tiblias, Irina Bigoulaeva, Jingcheng Niu, Simone Balloccu, Iryna Gurevych
Published: 2025-10-01
Categories: cs.AI, cs.CL
arXiv: https://arxiv.org/abs/2510.01025v2

The linear representation hypothesis states that language models (LMs) encode concepts as directions in their latent space, forming organized, multidimensional manifolds. Prior work has largely focused on identifying specific geometries for individual features, limiting its ability to generalize. We introduce Supervised Multi-Dimensional Scaling (SMDS), a model-agnostic method for evaluating and comparing competing feature manifold hypotheses. We apply SMDS to temporal reasoning as a case study and find that different features instantiate distinct geometric structures, including circles, lines, and clusters. SMDS reveals several consistent characteristics of these structures: they reflect the semantic properties of the concepts they represent, remain stable across model families and sizes, actively support reasoning, and dynamically reshape in response to contextual changes. Together, our findings shed light on the functional role of feature manifolds, supporting a model of entity-based reasoning in which LMs encode and transform structured representations.

---

### VDW-GNNs: Vector diffusion wavelets for geometric graph neural networks

Authors: David R. Johnson, Alexander Sietsema, Rishabh Anand, Deanna Needell, Smita Krishnaswamy, Michael Perlmutter
Published: 2025-10-01
Categories: cs.LG, eess.SP, stat.ML
arXiv: https://arxiv.org/abs/2510.01022v3

We introduce vector diffusion wavelets (VDWs), a novel family of wavelets inspired by the vector diffusion maps algorithm that was introduced to analyze data lying in the tangent bundle of a Riemannian manifold. We show that these wavelets may be effectively incorporated into a family of geometric graph neural networks, which we refer to as VDW-GNNs. We demonstrate that such networks are effective on synthetic point cloud data, as well as on real-world data derived from wind field and neural activity measurements. Theoretically, we prove that these new wavelets have desirable frame theoretic properties, similar to traditional diffusion wavelets. Additionally, we prove that these wavelets have useful symmetries with respect to rotations and translations.

---

### The Good, the Bad, and the Sampled: a No-Regret Approach to Safe Online Classification

Authors: Tavor Z. Baharav, Spyros Dragazis, Aldo Pacchiano
Published: 2025-10-01
Categories: cs.LG, cs.AI, math.ST, stat.ML
arXiv: https://arxiv.org/abs/2510.01020v2

We study sequential testing for a binary disease outcome when risk follows an unknown logistic model. At each round, the decision maker may either pay for a test revealing the true label or predict the outcome based on patient features and past data. The goal is to minimize costly tests while ensuring the misclassification rate stays below $α$ with probability at least $1-δ$. We propose a method that jointly estimates the logistic parameter $θ^{\star}$ and the feature distribution, using a conservative threshold on the logistic score to decide when to test. We prove our procedure achieves the target error with high probability and requires only $\widetilde O(\sqrt{T})$ more tests than an oracle with full knowledge. This is the first no-regret guarantee for error-constrained logistic testing, with direct applications to medical screening. Simulations corroborate our theoretical results, showing safe classification of patients and efficient estimation of $θ^{\star}$ with few excess tests.

---

### Quantifying the noise sensitivity of the Wasserstein metric for images

Authors: Erik Lager, Gilles Mordant, Amit Moscovich
Published: 2025-10-01
Categories: math.ST, cs.LG, eess.IV
arXiv: https://arxiv.org/abs/2510.01015v3

Wasserstein metrics are increasingly adopted as similarity scores for images. We consider the sensitivity of Wasserstein metrics with respect to pixel-wise additive noise when the images are treated as discrete measures on the pixel grid. We derive finite-sample expectation bounds for a Gaussian noise model. Among other results, we prove that the error in the signed 2-Wasserstein discrepancy scales with the square root of the noise standard deviation. This is favorable compared to the Euclidean metric that scales linearly, and thus provides a theoretical basis for the benefits of optimal transport distances in noisy settings. We present experiments that support our theoretical findings and point to a peculiar phenomenon where increasing the level of noise can decrease the Wasserstein distance. A case study on cryo-electron microscopy images demonstrates that the Wasserstein metric can capture the geometry of the data manifold in high noise settings even when the Euclidean metric fails.

---

### Revisiting Adversarial Training under Hyperspectral Image

Authors: Weihua Zhang, Chengze Jiang, Minjing Dong, Jie Gui, Lu Dong, Zhipeng Gui, Yuan Yan Tang, James Tin-Yau Kwok
Published: 2025-10-01
Categories: cs.CV
arXiv: https://arxiv.org/abs/2510.01014v2

Recent studies have shown that deep learning-based hyperspectral image (HSI) classification models are highly vulnerable to adversarial attacks, posing significant security risks. Although most approaches attempt to enhance robustness by optimizing network architectures, these methods often rely on customized designs with limited scalability and struggle to defend against strong attacks. To address this issue, we introduce adversarial training (AT), one of the most effective defense strategies, into the hyperspectral domain. However, unlike conventional RGB image classification, directly applying AT to HSI classification introduces unique challenges due to the high-dimensional spectral signatures and strong inter-band correlations of hyperspectral data, where discriminative information relies on subtle spectral semantics and spectral-spatial consistency that are highly sensitive to adversarial perturbations. Through extensive empirical analyses, we observe that adversarial perturbations and the non-smooth nature of adversarial examples can distort or even eliminate important spectral semantic information. To mitigate this issue, we propose two hyperspectral-specific AT methods, termed AT-HARL and AT-RA. Specifically, AT-HARL exploits spectral characteristic differences and class distribution ratios to design a novel loss function that alleviates semantic distortion caused by adversarial perturbations. Meanwhile, AT-RA introduces spectral data augmentation to enhance spectral diversity while preserving spatial smoothness. Experiments on four benchmark HSI datasets demonstrate that the proposed methods achieve competitive performance compared with state-of-the-art approaches under adversarial attacks.

---

### Random Feature Spiking Neural Networks

Authors: Maximilian Gollwitzer, Felix Dietrich
Published: 2025-10-01
Categories: cs.LG, cs.NE
arXiv: https://arxiv.org/abs/2510.01012v2

Spiking Neural Networks (SNNs) as Machine Learning (ML) models have recently received a lot of attention as a potentially more energy-efficient alternative to conventional Artificial Neural Networks. The non-differentiability and sparsity of the spiking mechanism can make these models very difficult to train with algorithms based on propagating gradients through the spiking non-linearity. We address this problem by adapting the paradigm of Random Feature Methods (RFMs) from Artificial Neural Networks (ANNs) to Spike Response Model (SRM) SNNs. This approach allows training of SNNs without approximation of the spike function gradient. Concretely, we propose a novel data-driven, fast, high-performance, and interpretable algorithm for end-to-end training of SNNs inspired by the SWIM algorithm for RFM-ANNs, which we coin S-SWIM. We provide a thorough theoretical discussion and supplementary numerical experiments showing that S-SWIM can reach high accuracies on time series forecasting as a standalone strategy and serve as an effective initialisation strategy before gradient-based training. Additional ablation studies show that our proposed method performs better than random sampling of network weights.

---

### ImageDoctor: Diagnosing Text-to-Image Generation via Grounded Image Reasoning

Authors: Yuxiang Guo, Jiang Liu, Ze Wang, Hao Chen, Ximeng Sun, Yang Zhao, Jialian Wu, Xiaodong Yu, Zicheng Liu, Emad Barsoum
Published: 2025-10-01
Categories: cs.CV
arXiv: https://arxiv.org/abs/2510.01010v1

The rapid advancement of text-to-image (T2I) models has increased the need for reliable human preference modeling, a demand further amplified by recent progress in reinforcement learning for preference alignment. However, existing approaches typically quantify the quality of a generated image using a single scalar, limiting their ability to provide comprehensive and interpretable feedback on image quality. To address this, we introduce ImageDoctor, a unified multi-aspect T2I model evaluation framework that assesses image quality across four complementary dimensions: plausibility, semantic alignment, aesthetics, and overall quality. ImageDoctor also provides pixel-level flaw indicators in the form of heatmaps, which highlight misaligned or implausible regions, and can be used as a dense reward for T2I model preference alignment. Inspired by the diagnostic process, we improve the detail sensitivity and reasoning capability of ImageDoctor by introducing a "look-think-predict" paradigm, where the model first localizes potential flaws, then generates reasoning, and finally concludes the evaluation with quantitative scores. Built on top of a vision-language model and trained through a combination of supervised fine-tuning and reinforcement learning, ImageDoctor demonstrates strong alignment with human preference across multiple datasets, establishing its effectiveness as an evaluation metric. Furthermore, when used as a reward model for preference tuning, ImageDoctor significantly improves generation quality -- achieving an improvement of 10% over scalar-based reward models.

---

### POVQA: Preference-Optimized Video Question Answering with Rationales for Data Efficiency

Authors: Ashim Dahal, Ankit Ghimire, Saydul Akbar Murad, Nick Rahimi
Published: 2025-10-01
Categories: cs.CV, cs.MM
arXiv: https://arxiv.org/abs/2510.01009v3

Long-video multimodal question answering requires structured reasoning over visual evidence and dialogue, but Large Vision-Language Models (LVLMs) are constrained by context-window and compute limits. We propose POVQA, which compresses each second into a temporally pooled image (1 fps pooled images) to maintain dense temporal coverage under a fixed token budget. We then train Qwen2.5-VL-7B with supervised fine-tuning (SFT) on rationale+answer targets, and optionally apply Direct Preference Optimization (DPO) for preference alignment. We introduce ReasonVQA as a pilot diagnostic dataset with 12 movies and 239 human-annotated QA+rationale triplets for controlled analysis of long-context multimodal reasoning under compression. On ReasonVQA, SFT improves the best pooled-only baseline from 0.212 to 0.550 F1, showing that pooled evidence plus rationale supervision provides the main performance gains in this setting. In zero-shot transfer, POVQA also reaches 64.7\% on TVQA after SFT+DPO. These results are preliminary: ReasonVQA is small, pooling can lose fine-grained temporal order, and DPO effects are not uniformly positive across settings. Code, dataset, and additional qualitative evaluations are available at \href{https://povqa.github.io}{https://povqa.github.io}.

---

### Integrating AI and Ensemble Forecasting: Explainable Materials Planning with Scorecards and Trend Insights for a Large-Scale Manufacturer

Authors: Saravanan Venkatachalam
Published: 2025-10-01
Categories: cs.AI, cs.LG
arXiv: https://arxiv.org/abs/2510.01006v1

This paper presents a practical architecture for after-sales demand forecasting and monitoring that unifies a revenue- and cluster-aware ensemble of statistical, machine-learning, and deep-learning models with a role-driven analytics layer for scorecards and trend diagnostics. The framework ingests exogenous signals (installed base, pricing, macro indicators, life cycle, seasonality) and treats COVID-19 as a distinct regime, producing country-part forecasts with calibrated intervals. A Pareto-aware segmentation forecasts high-revenue items individually and pools the long tail via clusters, while horizon-aware ensembling aligns weights with business-relevant losses (e.g., WMAPE). Beyond forecasts, a performance scorecard delivers decision-focused insights: accuracy within tolerance thresholds by revenue share and count, bias decomposition (over- vs under-forecast), geographic and product-family hotspots, and ranked root causes tied to high-impact part-country pairs. A trend module tracks trajectories of MAPE/WMAPE and bias across recent months, flags entities that are improving or deteriorating, detects change points aligned with known regimes, and attributes movements to lifecycle and seasonal factors. LLMs are embedded in the analytics layer to generate role-aware narratives and enforce reporting contracts. They standardize business definitions, automate quality checks and reconciliations, and translate quantitative results into concise, explainable summaries for planners and executives. The system exposes a reproducible workflow -- request specification, model execution, database-backed artifacts, and AI-generated narratives -- so planners can move from "How accurate are we now?" to "Where is accuracy heading and which levers should we pull?", closing the loop between forecasting, monitoring, and inventory decisions across more than 90 countries and about 6,000 parts.

---

### TextCAM: Explaining Class Activation Map with Text

Authors: Qiming Zhao, Xingjian Li, Xiaoyu Cao, Xiaolong Wu, Min Xu
Published: 2025-10-01
Categories: cs.CV, cs.AI, cs.LG
arXiv: https://arxiv.org/abs/2510.01004v1

Deep neural networks (DNNs) have achieved remarkable success across domains but remain difficult to interpret, limiting their trustworthiness in high-stakes applications. This paper focuses on deep vision models, for which a dominant line of explainability methods are Class Activation Mapping (CAM) and its variants working by highlighting spatial regions that drive predictions. We figure out that CAM provides little semantic insight into what attributes underlie these activations. To address this limitation, we propose TextCAM, a novel explanation framework that enriches CAM with natural languages. TextCAM combines the precise spatial localization of CAM with the semantic alignment of vision-language models (VLMs). Specifically, we derive channel-level semantic representations using CLIP embeddings and linear discriminant analysis, and aggregate them with CAM weights to produce textual descriptions of salient visual evidence. This yields explanations that jointly specify where the model attends and what visual attributes likely support its decision. We further extend TextCAM to generate feature channels into semantically coherent groups, enabling more fine-grained visual-textual explanations. Experiments on ImageNet, CLEVR, and CUB demonstrate that TextCAM produces faithful and interpretable rationales that improve human understanding, detect spurious correlations, and preserve model fidelity.

---

### Improving Code Localization with Repository Memory

Authors: Boshi Wang, Weijian Xu, Yunsheng Li, Mei Gao, Yujia Xie, Huan Sun, Dongdong Chen
Published: 2025-10-01
Categories: cs.SE, cs.CL
arXiv: https://arxiv.org/abs/2510.01003v2

Code localization is a fundamental challenge in repository-level software engineering tasks such as bug fixing. While existing methods equip language agents with comprehensive tools/interfaces to fetch information from the repository, they overlook the critical aspect of memory, where each instance is typically handled from scratch assuming no prior repository knowledge. In contrast, human developers naturally build long-term repository memory, such as the functionality of key modules and associations between various bug types and their likely fix locations. In this work, we augment language agents with such memory by leveraging a repository's commit history -- a rich yet underutilized resource that chronicles the codebase's evolution. We introduce tools that allow the agent to retrieve from a non-parametric memory encompassing recent historical commits and linked issues, as well as functionality summaries of actively evolving parts of the codebase identified via commit patterns. We demonstrate that augmenting such a memory can significantly improve LocAgent, a state-of-the-art localization framework, on both SWE-bench-verified and the more recent SWE-bench-live benchmarks. Our research contributes towards developing agents that can accumulate and leverage past experience for long-horizon tasks, more closely emulating the expertise of human developers.

---

### SoftCFG: Uncertainty-guided Stable Guidance for Visual Autoregressive Model

Authors: Dongli Xu, Aleksei Tiulpin, Matthew B. Blaschko
Published: 2025-10-01
Categories: cs.CV
arXiv: https://arxiv.org/abs/2510.00996v2

Autoregressive (AR) models have emerged as powerful tools for image generation by modeling images as sequences of discrete tokens. While Classifier-Free Guidance (CFG) has been adopted to improve conditional generation, its application in AR models faces two key issues: guidance diminishing, where the conditional-unconditional gap quickly vanishes as decoding progresses, and over-guidance, where strong conditions distort visual coherence. To address these challenges, we propose SoftCFG, an uncertainty-guided inference method that distributes adaptive perturbations across all tokens in the sequence. The key idea behind SoftCFG is to let each generated token contribute certainty-weighted guidance, ensuring that the signal persists across steps while resolving conflicts between text guidance and visual context. To further stabilize long-sequence generation, we introduce Step Normalization, which bounds cumulative perturbations of SoftCFG. Our method is training-free, model-agnostic, and seamlessly integrates with existing AR pipelines. Experiments show that SoftCFG significantly improves image quality over standard CFG and achieves state-of-the-art FID on ImageNet 256*256 among autoregressive models.

---

### Visual Self-Refinement for Autoregressive Models

Authors: Jiamian Wang, Ziqi Zhou, Chaithanya Kumar Mummadi, Sohail Dianat, Majid Rabbani, Raghuveer Rao, Chen Qiu, Zhiqiang Tao
Published: 2025-10-01
Categories: cs.CV
arXiv: https://arxiv.org/abs/2510.00993v1

Autoregressive models excel in sequential modeling and have proven to be effective for vision-language data. However, the spatial nature of visual signals conflicts with the sequential dependencies of next-token prediction, leading to suboptimal results. This work proposes a plug-and-play refinement module to enhance the complex spatial correspondence modeling within the generated visual sequence. This module operates as a post-pretraining step to jointly refine all generated tokens of autoregressive model, enhancing vision-language modeling under a shared sequential prediction framework. By leveraging global context and relationship across the tokens, our method mitigates the error accumulation issue within the sequential generation. Experiments demonstrate that the proposed method improves the generation quality, enhancing the model's ability to produce semantically consistent results.

---

### Riemannian Consistency Model

Authors: Chaoran Cheng, Yusong Wang, Yuxin Chen, Xiangxin Zhou, Nanning Zheng, Ge Liu
Published: 2025-10-01
Categories: cs.LG
arXiv: https://arxiv.org/abs/2510.00983v2

Consistency models are a class of generative models that enable few-step generation for diffusion and flow matching models. While consistency models have achieved promising results on Euclidean domains like images, their applications to Riemannian manifolds remain challenging due to the curved geometry. In this work, we propose the Riemannian Consistency Model (RCM), which, for the first time, enables few-step consistency modeling while respecting the intrinsic manifold constraint imposed by the Riemannian geometry. Leveraging the covariant derivative and exponential-map-based parameterization, we derive the closed-form solutions for both discrete- and continuous-time training objectives for RCM. We then demonstrate theoretical equivalence between the two variants of RCM: Riemannian consistency distillation (RCD) that relies on a teacher model to approximate the marginal vector field, and Riemannian consistency training (RCT) that utilizes the conditional vector field for training. We further propose a simplified training objective that eliminates the need for the complicated differential calculation. Finally, we provide a unique kinematics perspective for interpreting the RCM objective, offering new theoretical angles. Through extensive experiments, we manifest the superior generative quality of RCM in few-step generation on various non-Euclidean manifolds, including flat-tori, spheres, and the 3D rotation group SO(3).

---

### Spiralformer: Low Latency Encoder for Streaming Speech Recognition with Circular Layer Skipping and Early Exiting

Authors: Emiru Tsunoo, Hayato Futami, Yosuke Kashiwagi, Siddhant Arora, Shinji Watanabe
Published: 2025-10-01
Categories: eess.AS, cs.CL, cs.SD
arXiv: https://arxiv.org/abs/2510.00982v1

For streaming speech recognition, a Transformer-based encoder has been widely used with block processing. Although many studies addressed improving emission latency of transducers, little work has been explored for improving encoding latency of the block processing. We seek to reduce latency by frequently emitting a chunk with a small shift rather than scarce large-chunk emissions, resulting in higher computational costs. To efficiently compute with the small chunk shift, we propose a new encoder, Spiralformer, tailored for block processing by combining layer dropping and early exiting. We skip layer computation in a cyclic manner and shift the computed layer in each block spirally, which completes computation for all the layers over the block processing. Experimentally, we observed that our method achieved 21.6% reduction in the averaged token emission delay in Librispeech, and 7.0% in CSJ, compared with the baseline with similar computational cost and word error rates.

---

### A Scene is Worth a Thousand Features: Feed-Forward Camera Localization from a Collection of Image Features

Authors: Axel Barroso-Laguna, Tommaso Cavallari, Victor Adrian Prisacariu, Eric Brachmann
Published: 2025-10-01
Categories: cs.CV
arXiv: https://arxiv.org/abs/2510.00978v2

Visually localizing an image, i.e., estimating its camera pose, requires building a scene representation that serves as a visual map. The representation we choose has direct consequences towards the practicability of our system. Even when starting from mapping images with known camera poses, state-of-the-art approaches still require hours of mapping time in the worst case, and several minutes in the best. This work raises the question whether we can achieve competitive accuracy much faster. We introduce FastForward, a method that creates a map representation and relocalizes a query image on-the-fly in a single feed-forward pass. At the core, we represent multiple mapping images as a collection of features anchored in 3D space. FastForward utilizes these mapping features to predict image-to-scene correspondences for the query image, enabling the estimation of its camera pose. We couple FastForward with image retrieval and achieve state-of-the-art accuracy when compared to other approaches with minimal map preparation time. Furthermore, FastForward demonstrates robust generalization to unseen domains, including challenging large-scale outdoor environments.

---

### It Takes Two: Your GRPO Is Secretly DPO

Authors: Yihong Wu, Liheng Ma, Lei Ding, Muzhi Li, Xinyu Wang, Kejia Chen, Zhan Su, Zhanguang Zhang, Chenyang Huang, Yingxue Zhang, Mark Coates, Jian-Yun Nie
Published: 2025-10-01
Categories: cs.LG, cs.CL
arXiv: https://arxiv.org/abs/2510.00977v3

GRPO has emerged as a prominent reinforcement learning algorithm for post-training LLMs. Unlike critic-based methods, GRPO computes advantages by estimating the \emph{value baselines} from group-level statistics, eliminating the need for a critic network. Consequently, the prevailing view emphasizes the necessity of large group sizes, which are assumed to yield more accurate statistical estimates. In this paper, we propose a different view that the efficacy of GRPO stems from its implicit contrastive objective in the optimization, which helps reduce variance via the control variate method. This makes GRPO structurally related to preference learning methods such as DPO. This perspective motivates 2-GRPO, a minimal group-size variant that constructs contrastive signals with only two rollouts. We provide a rigorous theoretical analysis of 2-GRPO and empirically validate its effectiveness: 2-GRPO retains $97.6\%$ of the performance of 16-GRPO, while requiring only $12.5\%$ of the rollouts and $21\%$ of the training time.

---

### Adaptive Federated Few-Shot Rare-Disease Diagnosis with Energy-Aware Secure Aggregation

Authors: Aueaphum Aueawatthanaphisut
Published: 2025-10-01
Categories: cs.AI, cs.CR, cs.DC, cs.LG, q-bio.QM
arXiv: https://arxiv.org/abs/2510.00976v1

Rare-disease diagnosis remains one of the most pressing challenges in digital health, hindered by extreme data scarcity, privacy concerns, and the limited resources of edge devices. This paper proposes the Adaptive Federated Few-Shot Rare-Disease Diagnosis (AFFR) framework, which integrates three pillars: (i) few-shot federated optimization with meta-learning to generalize from limited patient samples, (ii) energy-aware client scheduling to mitigate device dropouts and ensure balanced participation, and (iii) secure aggregation with calibrated differential privacy to safeguard sensitive model updates. Unlike prior work that addresses these aspects in isolation, AFFR unifies them into a modular pipeline deployable on real-world clinical networks. Experimental evaluation on simulated rare-disease detection datasets demonstrates up to 10% improvement in accuracy compared with baseline FL, while reducing client dropouts by over 50% without degrading convergence. Furthermore, privacy-utility trade-offs remain within clinically acceptable bounds. These findings highlight AFFR as a practical pathway for equitable and trustworthy federated diagnosis of rare conditions.

---

### JEPA-T: Joint-Embedding Predictive Architecture with Text Fusion for Image Generation

Authors: Siheng Wan, Zhengtao Yao, Zhengdao Li, Junhao Dong, Yanshu Li, Yikai Li, Linshan Li, Haoyan Xu, Yijiang Li, Zhikang Dong, Huacan Wang, Jifeng Shen
Published: 2025-10-01
Categories: cs.CV
arXiv: https://arxiv.org/abs/2510.00974v1

Modern Text-to-Image (T2I) generation increasingly relies on token-centric architectures that are trained with self-supervision, yet effectively fusing text with visual tokens remains a challenge. We propose \textbf{JEPA-T}, a unified multimodal framework that encodes images and captions into discrete visual and textual tokens, processed by a joint-embedding predictive Transformer. To enhance fusion, we incorporate cross-attention after the feature predictor for conditional denoising while maintaining a task-agnostic backbone. Additionally, raw texts embeddings are injected prior to the flow matching loss to improve alignment during training. During inference, the same network performs both class-conditional and free-text image generation by iteratively denoising visual tokens conditioned on text. Evaluations on ImageNet-1K demonstrate that JEPA-T achieves strong data efficiency, open-vocabulary generalization, and consistently outperforms non-fusion and late-fusion baselines. Our approach shows that late architectural fusion combined with objective-level alignment offers an effective balance between conditioning strength and backbone generality in token-based T2I.The code is now available: https://github.com/justin-herry/JEPA-T.git

---

### QUASAR: Quantum Assembly Code Generation Using Tool-Augmented LLMs via Agentic RL

Authors: Cong Yu, Valter Uotila, Shilong Deng, Qingyuan Wu, Tuo Shi, Songlin Jiang, Lei You, Bo Zhao
Published: 2025-10-01
Categories: cs.AI, quant-ph
arXiv: https://arxiv.org/abs/2510.00967v1

Designing and optimizing task-specific quantum circuits are crucial to leverage the advantage of quantum computing. Recent large language model (LLM)-based quantum circuit generation has emerged as a promising automatic solution. However, the fundamental challenges remain unaddressed: (i) parameterized quantum gates require precise numerical values for optimal performance, which also depend on multiple aspects, including the number of quantum gates, their parameters, and the layout/depth of the circuits. (ii) LLMs often generate low-quality or incorrect quantum circuits due to the lack of quantum domain-specific knowledge. We propose QUASAR, an agentic reinforcement learning (RL) framework for quantum circuits generation and optimization based on tool-augmented LLMs. To align the LLM with quantum-specific knowledge and improve the generated quantum circuits, QUASAR designs (i) a quantum circuit verification approach with external quantum simulators and (ii) a sophisticated hierarchical reward mechanism in RL training. Extensive evaluation shows improvements in both syntax and semantic performance of the generated quantum circuits. When augmenting a 4B LLM, QUASAR has achieved the validity of 99.31% in Pass@1 and 100% in Pass@10, outperforming industrial LLMs of GPT-4o, GPT-5 and DeepSeek-V3 and several supervised-fine-tuning (SFT)-only and RL-only baselines.

---

### Deep Learning-Based Approach for Improving Relational Aggregated Search

Authors: Sara Saad Soliman, Ahmed Younes, Islam Elkabani, Ashraf Elsayed
Published: 2025-10-01
Categories: cs.IR, cs.AI
arXiv: https://arxiv.org/abs/2510.00966v1

Due to an information explosion on the internet, there is a need for the development of aggregated search systems that can boost the retrieval and management of content in various formats. To further improve the clustering of Arabic text data in aggregated search environments, this research investigates the application of advanced natural language processing techniques, namely stacked autoencoders and AraBERT embeddings. By transcending the limitations of traditional search engines, which are imprecise, not contextually relevant, and not personalized, we offer more enriched, context-aware characterizations of search results, so we used a K-means clustering algorithm to discover distinctive features and relationships in these results, we then used our approach on different Arabic queries to evaluate its effectiveness. Our model illustrates that using stacked autoencoders in representation learning suits clustering tasks and can significantly improve clustering search results. It also demonstrates improved accuracy and relevance of search results.

---

### Analyzing Dialectical Biases in LLMs for Knowledge and Reasoning Benchmarks

Authors: Eileen Pan, Anna Seo Gyeong Choi, Maartje ter Hoeve, Skyler Seto, Allison Koenecke
Published: 2025-10-01
Categories: cs.CL
arXiv: https://arxiv.org/abs/2510.00962v1

Large language models (LLMs) are ubiquitous in modern day natural language processing. However, previous work has shown degraded LLM performance for under-represented English dialects. We analyze the effects of typifying "standard" American English language questions as non-"standard" dialectal variants on multiple choice question answering tasks and find up to a 20% reduction in accuracy. Additionally, we investigate the grammatical basis of under-performance in non-"standard" English questions. We find that individual grammatical rules have varied effects on performance, but some are more consequential than others: three specific grammar rules (existential "it", zero copula, and y'all) can explain the majority of performance degradation observed in multiple dialects. We call for future work to investigate bias mitigation methods focused on individual, high-impact grammatical structures.

---

### A Neuro-Fuzzy System for Interpretable Long-Term Stock Market Forecasting

Authors: Miha Ožbot, Igor Škrjanc, Vitomir Štruc
Published: 2025-10-01
Categories: cs.AI, cs.NE, eess.SY
arXiv: https://arxiv.org/abs/2510.00960v1

In the complex landscape of multivariate time series forecasting, achieving both accuracy and interpretability remains a significant challenge. This paper introduces the Fuzzy Transformer (Fuzzformer), a novel recurrent neural network architecture combined with multi-head self-attention and fuzzy inference systems to analyze multivariate stock market data and conduct long-term time series forecasting. The method leverages LSTM networks and temporal attention to condense multivariate data into interpretable features suitable for fuzzy inference systems. The resulting architecture offers comparable forecasting performance to conventional models such as ARIMA and LSTM while providing meaningful information flow within the network. The method was examined on the real world stock market index S\&P500. Initial results show potential for interpretable forecasting and identify current performance tradeoffs, suggesting practical application in understanding and forecasting stock market behavior.

---

### Test-Time Search in Neural Graph Coarsening Procedures for the Capacitated Vehicle Routing Problem

Authors: Yoonju Sim, Hyeonah Kim, Changhyun Kwon
Published: 2025-10-01
Categories: cs.AI, math.OC
arXiv: https://arxiv.org/abs/2510.00958v1

The identification of valid inequalities, such as the rounded capacity inequalities (RCIs), is a key component of cutting plane methods for the Capacitated Vehicle Routing Problem (CVRP). While a deep learning-based separation method can learn to find high-quality cuts, our analysis reveals that the model produces fewer cuts than expected because it is insufficiently sensitive to generate a diverse set of generated subsets. This paper proposes an alternative: enhancing the performance of a trained model at inference time through a new test-time search with stochasticity. First, we introduce stochastic edge selection into the graph coarsening procedure, replacing the previously proposed greedy approach. Second, we propose the Graph Coarsening History-based Partitioning (GraphCHiP) algorithm, which leverages coarsening history to identify not only RCIs but also, for the first time, the Framed capacity inequalities (FCIs). Experiments on randomly generated CVRP instances demonstrate the effectiveness of our approach in reducing the dual gap compared to the existing neural separation method. Additionally, our method discovers effective FCIs on a specific instance, despite the challenging nature of identifying such cuts.

---

### Bridging the Gap Between Simulated and Real Network Data Using Transfer Learning

Authors: Carlos Güemes-Palau, Miquel Ferriol-Galmés, Jordi Paillisse-Vilanova, Albert López-Brescó, Pere Barlet-Ros, Albert Cabellos-Aparicio
Published: 2025-10-01
Categories: cs.NI, cs.AI, cs.LG
arXiv: https://arxiv.org/abs/2510.00956v2

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%.

---

### Modeling Market States with Clustering and State Machines

Authors: Christian Oliva, Silviu Gabriel Tinjala
Published: 2025-10-01
Categories: cs.CE, cs.LG
arXiv: https://arxiv.org/abs/2510.00953v1

This work introduces a new framework for modeling financial markets through an interpretable probabilistic state machine. By clustering historical returns based on momentum and risk features across multiple time horizons, we identify distinct market states that capture underlying regimes, such as expansion phase, contraction, crisis, or recovery. From a transition matrix representing the dynamics between these states, we construct a probabilistic state machine that models the temporal evolution of the market. This state machine enables the generation of a custom distribution of returns based on a mixture of Gaussian components weighted by state frequencies. We show that the proposed benchmark significantly outperforms the traditional approach in capturing key statistical properties of asset returns, including skewness and kurtosis, and our experiments across random assets and time periods confirm its robustness.

---

### InfVSR: Toward Consistency-Driven Streaming Generative Video Super-Resolution

Authors: Ziqing Zhang, Kai Liu, Zheng Chen, Xi Li, Yucong Chen, Bingnan Duan, Linghe Kong, Yulun Zhang
Published: 2025-10-01
Categories: cs.CV
arXiv: https://arxiv.org/abs/2510.00948v3

Real-world videos often extend over thousands of frames. Existing generative video super-resolution (VSR) approaches, however, face two persistent challenges when processing long sequences: (1) inefficiency due to the heavy cost of multi-step denoising for full-length sequences; and (2) poor consistency is hindered by temporal decomposition that causes artifacts and discontinuities. To break these limits, we propose InfVSR, which reformulates VSR as an autoregressive-one-step-diffusion paradigm, and enables streaming inference with video diffusion priors. First, we adapt the pretrained DiT into a causal structure, maintaining both local and global coherence via rolling KV-cache and joint visual guidance. Second, we distill the diffusion process into a single step efficiently, with patch-wise pixel supervision and cross-chunk distribution matching. To fill the gap in long-form video evaluation, we build a new benchmark tailored for extended sequences and further introduce semantic-level metrics to comprehensively assess temporal consistency. Our method pushes the frontier of long-form VSR, achieves state-of-the-art quality with enhanced semantic consistency, and delivers up to 58x speed-up over existing methods such as MGLD-VSR. Our code and models are available at https://github.com/Kai-Liu001/InfVSR.

---

### Large Reasoning Models Learn Better Alignment from Flawed Thinking

Authors: ShengYun Peng, Pin-Yu Chen, Eric Smith, Song Jiang, Hongyuan Zhan, Haozhu Wang, Mahesh Pasupuleti, Duen Horng Chau, Jianfeng Chi
Published: 2025-10-01
Categories: cs.LG
arXiv: https://arxiv.org/abs/2510.00938v3

Large reasoning models (LRMs) "think" by generating structured chain-of-thought (CoT) before producing a final answer, yet they still lack the ability to reason critically about safety alignment and are easily biased when a flawed premise is injected into their thought process. We propose RECAP (Robust Safety Alignment via Counter-Aligned Prefilling), a principled reinforcement learning (RL) method for post-training that explicitly teaches models to override flawed reasoning trajectories and reroute to safe and helpful responses. RECAP trains on a mixture of synthetically generated counter-aligned CoT prefills and standard prompts, requires no additional training cost or modifications beyond vanilla reinforcement learning from human feedback (RLHF), and substantially improves safety and jailbreak robustness, reduces overrefusal, and preserves core reasoning capability -- all while maintaining inference token budget. Extensive analysis shows that RECAP-trained models engage in self-reflection more frequently and remain robust under adaptive attacks, preserving safety even after repeated attempts to override their reasoning.

---