# Research Papers — 2025-10 · bundle page 207 of 217 > Full-text markdown of papers #10301–10350 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/207 [← Previous bundle](https://aipod.dev/papers/archive/2025-10/page/206.md) [Next bundle →](https://aipod.dev/papers/archive/2025-10/page/208.md) ### Discrete Facial Encoding: : A Framework for Data-driven Facial Display Discovery Authors: Minh Tran, Maksim Siniukov, Zhangyu Jin, Mohammad Soleymani Published: 2025-10-02 Categories: cs.CV arXiv: https://arxiv.org/abs/2510.01662v1 Facial expression analysis is central to understanding human behavior, yet existing coding systems such as the Facial Action Coding System (FACS) are constrained by limited coverage and costly manual annotation. In this work, we introduce Discrete Facial Encoding (DFE), an unsupervised, data-driven alternative of compact and interpretable dictionary of facial expressions from 3D mesh sequences learned through a Residual Vector Quantized Variational Autoencoder (RVQ-VAE). Our approach first extracts identity-invariant expression features from images using a 3D Morphable Model (3DMM), effectively disentangling factors such as head pose and facial geometry. We then encode these features using an RVQ-VAE, producing a sequence of discrete tokens from a shared codebook, where each token captures a specific, reusable facial deformation pattern that contributes to the overall expression. Through extensive experiments, we demonstrate that Discrete Facial Encoding captures more precise facial behaviors than FACS and other facial encoding alternatives. We evaluate the utility of our representation across three high-level psychological tasks: stress detection, personality prediction, and depression detection. Using a simple Bag-of-Words model built on top of the learned tokens, our system consistently outperforms both FACS-based pipelines and strong image and video representation learning models such as Masked Autoencoders. Further analysis reveals that our representation covers a wider variety of facial displays, highlighting its potential as a scalable and effective alternative to FACS for psychological and affective computing applications. --- ### VirDA: Reusing Backbone for Unsupervised Domain Adaptation with Visual Reprogramming Authors: Duy Nguyen, Dat Nguyen Published: 2025-10-02 Categories: cs.CV arXiv: https://arxiv.org/abs/2510.01660v4 Existing UDA pipelines fine-tune already well-trained backbone parameters for every new source-and-target pair, resulting in the number of training parameters and storage memory growing linearly with each new pair, and also preventing the reuse of these well-trained backbone parameters. Inspired by recent implications that existing backbones have textural biases, we propose making use of domain-specific textural bias for domain adaptation via visual reprogramming, namely VirDA. Instead of fine-tuning the full backbone, VirDA prepends a domain-specific visual reprogramming layer to the backbone. This layer produces visual prompts that act as an added textural bias to the input image, adapting its "style" to a target domain. To optimize these visual reprogramming layers, we use multiple objective functions that optimize the intra- and inter-domain distribution differences when domain-adapting visual prompts are applied. This process does not require modifying the backbone parameters, allowing the same backbone to be reused across different domains. We evaluate VirDA on Office-31 and obtain 92.8% mean accuracy with only 1.5M trainable parameters. VirDA surpasses PDA, the state-of-the-art parameter-efficient UDA baseline, by +1.6% accuracy while using just 46% of its parameters. Compared with full-backbone fine-tuning, VirDA outperforms CDTrans and FixBi by +0.2% and +1.4%, respectively, while requiring only 1.7% and 2.8% of their trainable parameters. Relative to the strongest current methods (PMTrans and TVT), VirDA uses ~1.7% of their parameters and trades off only 2.2% and 1.1% accuracy, respectively. --- ### MDSEval: A Meta-Evaluation Benchmark for Multimodal Dialogue Summarization Authors: Yinhong Liu, Jianfeng He, Hang Su, Ruixue Lian, Yi Nian, Jake Vincent, Srikanth Vishnubhotla, Robinson Piramuthu, Saab Mansour Published: 2025-10-02 Categories: cs.CL, cs.AI arXiv: https://arxiv.org/abs/2510.01659v1 Multimodal Dialogue Summarization (MDS) is a critical task with wide-ranging applications. To support the development of effective MDS models, robust automatic evaluation methods are essential for reducing both cost and human effort. However, such methods require a strong meta-evaluation benchmark grounded in human annotations. In this work, we introduce MDSEval, the first meta-evaluation benchmark for MDS, consisting image-sharing dialogues, corresponding summaries, and human judgments across eight well-defined quality aspects. To ensure data quality and richfulness, we propose a novel filtering framework leveraging Mutually Exclusive Key Information (MEKI) across modalities. Our work is the first to identify and formalize key evaluation dimensions specific to MDS. We benchmark state-of-the-art modal evaluation methods, revealing their limitations in distinguishing summaries from advanced MLLMs and their susceptibility to various bias. --- ### Learning Time-Series Representations by Hierarchical Uniformity-Tolerance Latent Balancing Authors: Amin Jalali, Milad Soltany, Michael Greenspan, Ali Etemad Published: 2025-10-02 Categories: cs.LG, cs.AI arXiv: https://arxiv.org/abs/2510.01658v1 We propose TimeHUT, a novel method for learning time-series representations by hierarchical uniformity-tolerance balancing of contrastive representations. Our method uses two distinct losses to learn strong representations with the aim of striking an effective balance between uniformity and tolerance in the embedding space. First, TimeHUT uses a hierarchical setup to learn both instance-wise and temporal information from input time-series. Next, we integrate a temperature scheduler within the vanilla contrastive loss to balance the uniformity and tolerance characteristics of the embeddings. Additionally, a hierarchical angular margin loss enforces instance-wise and temporal contrast losses, creating geometric margins between positive and negative pairs of temporal sequences. This approach improves the coherence of positive pairs and their separation from the negatives, enhancing the capture of temporal dependencies within a time-series sample. We evaluate our approach on a wide range of tasks, namely 128 UCR and 30 UAE datasets for univariate and multivariate classification, as well as Yahoo and KPI datasets for anomaly detection. The results demonstrate that TimeHUT outperforms prior methods by considerable margins on classification, while obtaining competitive results for anomaly detection. Finally, detailed sensitivity and ablation studies are performed to evaluate different components and hyperparameters of our method. --- ### Asymmetric Proximal Policy Optimization: mini-critics boost LLM reasoning Authors: Jiashun Liu, Johan Obando-Ceron, Han Lu, Yancheng He, Weixun Wang, Wenbo Su, Bo Zheng, Pablo Samuel Castro, Aaron Courville, Ling Pan Published: 2025-10-02 Categories: cs.LG, cs.AI arXiv: https://arxiv.org/abs/2510.01656v3 Most recent RL for LLMs (RL4LLM) methods avoid explicit critics, replacing them with average advantage baselines. This shift is largely pragmatic: conventional value functions are computationally expensive to train at LLM scale and often fail under sparse rewards and long reasoning horizons. We revisit this bottleneck from an architectural perspective and introduce Asymmetric Proximal Policy Optimization (AsyPPO), a simple and scalable framework that restores the critics role while remaining efficient in large-model settings. AsyPPO employs a set of lightweight mini-critics, each trained on disjoint prompt shards. This design encourages diversity while preserving calibration, reducing value-estimation bias. Beyond robust estimation, AsyPPO leverages inter-critic uncertainty to refine the policy update: (i) masking advantages in states where critics agree and gradients add little learning signal, and (ii) filtering high-divergence states from entropy regularization, suppressing spurious exploration. After training on open-source data with only 5,000 samples, AsyPPO consistently improves learning stability and performance across multiple benchmarks over strong baselines, such as GRPO, achieving performance gains of more than six percent on Qwen3-4b-Base and about three percent on Qwen3-8b-Base and Qwen3-14b-Base over classic PPO, without additional tricks. These results highlight the importance of architectural innovations for scalable, efficient algorithms. --- ### SoK: Measuring What Matters for Closed-Loop Security Agents Authors: Mudita Khurana, Raunak Jain Published: 2025-10-02 Categories: cs.CL, cs.AI arXiv: https://arxiv.org/abs/2510.01654v1 Cybersecurity is a relentless arms race, with AI driven offensive systems evolving faster than traditional defenses can adapt. Research and tooling remain fragmented across isolated defensive functions, creating blind spots that adversaries exploit. Autonomous agents capable of integrating, exploit confirmation, remediation, and validation into a single closed loop offer promise, but the field lacks three essentials: a framework defining the agentic capabilities of security systems across security life cycle, a principled method for evaluating closed loop agents, and a benchmark for measuring their performance in practice. We introduce CLASP: the Closed-Loop Autonomous Security Performance framework which aligns the security lifecycle (reconnaissance, exploitation, root cause analysis, patch synthesis, validation) with core agentic capabilities (planning, tool use, memory, reasoning, reflection & perception) providing a common vocabulary and rubric for assessing agentic capabilities in security tasks. By applying CLASP to 21 representative works, we map where systems demonstrate strengths, and where capability gaps persist. We then define the Closed-Loop Capability (CLC) Score, a composite metric quantifying both degree of loop closure and operational effectiveness, and outline the requirements for a closed loop benchmark. Together, CLASP and the CLC Score, provide the vocabulary, diagnostics, and measurements needed to advance both function level performance and measure closed loop security agents. --- ### Learning to Look at the Other Side: A Semantic Probing Study of Word Embeddings in LLMs with Enabled Bidirectional Attention Authors: Zhaoxin Feng, Jianfei Ma, Emmanuele Chersoni, Xiaojing Zhao, Xiaoyi Bao Published: 2025-10-02 Categories: cs.CL arXiv: https://arxiv.org/abs/2510.01652v1 Autoregressive Large Language Models (LLMs) demonstrate exceptional performance in language understanding and generation. However, their application in text embedding tasks has been relatively slow, along with the analysis of their semantic representation in probing tasks, due to the constraints of the unidirectional attention mechanism. This paper aims to explore whether such constraints can be overcome by enabling bidirectional attention in LLMs. We tested different variants of the Llama architecture through additional training steps, progressively enabling bidirectional attention and unsupervised/supervised contrastive learning. --- ### LadderMoE: Ladder-Side Mixture of Experts Adapters for Bronze Inscription Recognition Authors: Rixin Zhou, Peiqiang Qiu, Qian Zhang, Chuntao Li, Xi Yang Published: 2025-10-02 Categories: cs.CV arXiv: https://arxiv.org/abs/2510.01651v1 Bronze inscriptions (BI), engraved on ritual vessels, constitute a crucial stage of early Chinese writing and provide indispensable evidence for archaeological and historical studies. However, automatic BI recognition remains difficult due to severe visual degradation, multi-domain variability across photographs, rubbings, and tracings, and an extremely long-tailed character distribution. To address these challenges, we curate a large-scale BI dataset comprising 22454 full-page images and 198598 annotated characters spanning 6658 unique categories, enabling robust cross-domain evaluation. Building on this resource, we develop a two-stage detection-recognition pipeline that first localizes inscriptions and then transcribes individual characters. To handle heterogeneous domains and rare classes, we equip the pipeline with LadderMoE, which augments a pretrained CLIP encoder with ladder-style MoE adapters, enabling dynamic expert specialization and stronger robustness. Comprehensive experiments on single-character and full-page recognition tasks demonstrate that our method substantially outperforms state-of-the-art scene text recognition baselines, achieving superior accuracy across head, mid, and tail categories as well as all acquisition modalities. These results establish a strong foundation for bronze inscription recognition and downstream archaeological analysis. --- ### The Unseen Frontier: Pushing the Limits of LLM Sparsity with Surrogate-Free ADMM Authors: Kwanhee Lee, Hyeondo Jang, Dongyeop Lee, Dan Alistarh, Namhoon Lee Published: 2025-10-02 Categories: cs.LG, cs.AI arXiv: https://arxiv.org/abs/2510.01650v2 Neural network pruning is a promising technique to mitigate the excessive computational and memory requirements of large language models (LLMs). Despite its promise, however, progress in this area has diminished, as conventional methods are seemingly unable to surpass moderate sparsity levels (50-60%) without severely degrading model accuracy. This work breaks through the current impasse, presenting a principled and effective method called $\texttt{Elsa}$, which achieves extreme sparsity levels of up to 90% while retaining high model fidelity. This is done by identifying several limitations in current practice, all of which can be traced back to their reliance on a surrogate objective formulation. $\texttt{Elsa}$ tackles this issue directly and effectively via standard and well-established constrained optimization techniques based on ADMM. Our extensive experiments across a wide range of models and scales show that $\texttt{Elsa}$ achieves substantial improvements over existing methods; e.g., it achieves 7.8$\times$ less perplexity than the best existing method on LLaMA-2-7B at 90% sparsity. Moreover, we show that $\texttt{Elsa}$ remains stable even at extreme sparsity (e.g., 95\%), yielding up to $\times$3.98 inference speedup and $\times$7.80 memory compression over its dense counterpart. We also present $\texttt{Elsa}_{-L}$, a quantized variant that scales to extremely large models (27B), and establish its theoretical convergence guarantees.These results highlight meaningful progress in advancing the frontier of LLM sparsity, while promising that significant opportunities for further advancement may remain in directions that have so far attracted limited exploration. --- ### Source-Free Cross-Domain Continual Learning Authors: Muhammad Tanzil Furqon, Mahardhika Pratama, Igor Škrjanc, Lin Liu, Habibullah Habibullah, Kutluyil Dogancay Published: 2025-10-02 Categories: cs.LG, cs.AI arXiv: https://arxiv.org/abs/2510.01649v1 Although existing cross-domain continual learning approaches successfully address many streaming tasks having domain shifts, they call for a fully labeled source domain hindering their feasibility in the privacy constrained environments. This paper goes one step ahead with the problem of source-free cross-domain continual learning where the use of source-domain samples are completely prohibited. We propose the idea of rehearsal-free frequency-aware dynamic prompt collaborations (REFEREE) to cope with the absence of labeled source-domain samples in realm of cross-domain continual learning. REFEREE is built upon a synergy between a source-pre-trained model and a large-scale vision-language model, thus overcoming the problem of sub-optimal generalizations when relying only on a source pre-trained model. The domain shift problem between the source domain and the target domain is handled by a frequency-aware prompting technique encouraging low-frequency components while suppressing high-frequency components. This strategy generates frequency-aware augmented samples, robust against noisy pseudo labels. The noisy pseudo-label problem is further addressed with the uncertainty-aware weighting strategy where the mean and covariance matrix are weighted by prediction uncertainties, thus mitigating the adverse effects of the noisy pseudo label. Besides, the issue of catastrophic forgetting (CF) is overcome by kernel linear discriminant analysis (KLDA) where the backbone network is frozen while the classification is performed using the linear discriminant analysis approach guided by the random kernel method. Our rigorous numerical studies confirm the advantage of our approach where it beats prior arts having access to source domain samples with significant margins. --- ### Position: Privacy Is Not Just Memorization! Authors: Niloofar Mireshghallah, Tianshi Li Published: 2025-10-02 Categories: cs.CR, cs.AI, cs.CL, cs.LG arXiv: https://arxiv.org/abs/2510.01645v1 The discourse on privacy risks in Large Language Models (LLMs) has disproportionately focused on verbatim memorization of training data, while a constellation of more immediate and scalable privacy threats remain underexplored. This position paper argues that the privacy landscape of LLM systems extends far beyond training data extraction, encompassing risks from data collection practices, inference-time context leakage, autonomous agent capabilities, and the democratization of surveillance through deep inference attacks. We present a comprehensive taxonomy of privacy risks across the LLM lifecycle -- from data collection through deployment -- and demonstrate through case studies how current privacy frameworks fail to address these multifaceted threats. Through a longitudinal analysis of 1,322 AI/ML privacy papers published at leading conferences over the past decade (2016--2025), we reveal that while memorization receives outsized attention in technical research, the most pressing privacy harms lie elsewhere, where current technical approaches offer little traction and viable paths forward remain unclear. We call for a fundamental shift in how the research community approaches LLM privacy, moving beyond the narrow focus of current technical solutions and embracing interdisciplinary approaches that address the sociotechnical nature of these emerging threats. --- ### Machine Learning for Detection and Analysis of Novel LLM Jailbreaks Authors: John Hawkins, Aditya Pramar, Rodney Beard, Rohitash Chandra Published: 2025-10-02 Categories: cs.CL, cs.AI, cs.CY arXiv: https://arxiv.org/abs/2510.01644v2 Large Language Models (LLMs) suffer from a range of vulnerabilities that allow malicious users to solicit undesirable responses through manipulation of the input text. These so-called jailbreak prompts are designed to trick the LLM into circumventing the safety guardrails put in place to keep responses acceptable to the developer's policies. In this study, we analyse the ability of different machine learning models to distinguish jailbreak prompts from genuine uses, including looking at our ability to identify jailbreaks that use previously unseen strategies. Our results indicate that using current datasets the best performance is achieved by fine tuning a Bidirectional Encoder Representations from Transformers (BERT) model end-to-end for identifying jailbreaks. We visualise the keywords that distinguish jailbreak from genuine prompts and conclude that explicit reflexivity in prompt structure could be a signal of jailbreak intention. --- ### Support Basis: Fast Attention Beyond Bounded Entries Authors: Maryam Aliakbarpour, Vladimir Braverman, Junze Yin, Haochen Zhang Published: 2025-10-02 Categories: cs.LG arXiv: https://arxiv.org/abs/2510.01643v2 Large language models (LLMs) have demonstrated remarkable performance across a wide range of tasks. However, the quadratic complexity of softmax attention remains a central bottleneck that limits their scalability. Alman and Song (NeurIPS 2023a; NeurIPS 2024a) proposed sub-quadratic time algorithms for attention inference and training, respectively, but they rely on the restrictive bounded-entry assumption. We show that this assumption rarely holds in practice, which significantly limits their applicability to modern LLMs. In this paper, we introduce support-basis decomposition, a new technique for accurate and efficient attention inference and training without the bounded-entry assumption. We empirically show that the entries of the query and key matrices exhibit sub-Gaussian behavior. Leveraging this widely observed property, we perform exact computation on sparse components and polynomial approximation on dense components. Without relying on restrictive assumptions, we theoretically show that our algorithm achieves sub-quadratic runtime while matching the approximation error of prior work, and we empirically validate its computational efficiency and downstream task performance. We further generalize our method to a multi-threshold setting that eliminates all distributional assumptions, providing the first theoretical justification for the empirical success of polynomial attention. Moreover, we show that softmax attention can be closely approximated by multiple polynomial attentions with significantly smaller $\ell_p$ error. --- ### FideDiff: Efficient Diffusion Model for High-Fidelity Image Motion Deblurring Authors: Xiaoyang Liu, Zhengyan Zhou, Zihang Xu, Jiezhang Cao, Zheng Chen, Yulun Zhang Published: 2025-10-02 Categories: cs.CV arXiv: https://arxiv.org/abs/2510.01641v3 Recent advancements in image motion deblurring, driven by CNNs and transformers, have made significant progress. Large-scale pre-trained diffusion models, which are rich in real-world modeling, have shown great promise for high-quality image restoration tasks such as deblurring, demonstrating stronger generative capabilities than CNN and transformer-based methods. However, challenges such as unbearable inference time and compromised fidelity still limit the full potential of the diffusion models. To address this, we introduce FideDiff, a novel single-step diffusion model designed for high-fidelity deblurring. We reformulate motion deblurring as a diffusion-like process where each timestep represents a progressively blurred image, and we train a consistency model that aligns all timesteps to the same clean image. By reconstructing training data with matched blur trajectories, the model learns temporal consistency, enabling accurate one-step deblurring. We further enhance model performance by integrating Kernel ControlNet for blur kernel estimation and introducing adaptive timestep prediction. Our model achieves superior performance on full-reference metrics, surpassing previous diffusion-based methods and matching the performance of other state-of-the-art models. FideDiff offers a new direction for applying pre-trained diffusion models to high-fidelity image restoration tasks, establishing a robust baseline for further advancing diffusion models in real-world industrial applications. Our dataset and code will be available at https://github.com/xyLiu339/FideDiff. --- ### Joint Deblurring and 3D Reconstruction for Macrophotography Authors: Yifan Zhao, Liangchen Li, Yuqi Zhou, Kai Wang, Yan Liang, Juyong Zhang Published: 2025-10-02 Categories: cs.CV arXiv: https://arxiv.org/abs/2510.01640v1 Macro lens has the advantages of high resolution and large magnification, and 3D modeling of small and detailed objects can provide richer information. However, defocus blur in macrophotography is a long-standing problem that heavily hinders the clear imaging of the captured objects and high-quality 3D reconstruction of them. Traditional image deblurring methods require a large number of images and annotations, and there is currently no multi-view 3D reconstruction method for macrophotography. In this work, we propose a joint deblurring and 3D reconstruction method for macrophotography. Starting from multi-view blurry images captured, we jointly optimize the clear 3D model of the object and the defocus blur kernel of each pixel. The entire framework adopts a differentiable rendering method to self-supervise the optimization of the 3D model and the defocus blur kernel. Extensive experiments show that from a small number of multi-view images, our proposed method can not only achieve high-quality image deblurring but also recover high-fidelity 3D appearance. --- ### Understanding the Geospatial Reasoning Capabilities of LLMs: A Trajectory Recovery Perspective Authors: Thinh Hung Truong, Jey Han Lau, Jianzhong Qi Published: 2025-10-02 Categories: cs.AI arXiv: https://arxiv.org/abs/2510.01639v1 We explore the geospatial reasoning capabilities of Large Language Models (LLMs), specifically, whether LLMs can read road network maps and perform navigation. We frame trajectory recovery as a proxy task, which requires models to reconstruct masked GPS traces, and introduce GLOBALTRACE, a dataset with over 4,000 real-world trajectories across diverse regions and transportation modes. Using road network as context, our prompting framework enables LLMs to generate valid paths without accessing any external navigation tools. Experiments show that LLMs outperform off-the-shelf baselines and specialized trajectory recovery models, with strong zero-shot generalization. Fine-grained analysis shows that LLMs have strong comprehension of the road network and coordinate systems, but also pose systematic biases with respect to regions and transportation modes. Finally, we demonstrate how LLMs can enhance navigation experiences by reasoning over maps in flexible ways to incorporate user preferences. --- ### Towards Human-Centered RegTech: Unpacking Professionals' Strategies and Needs for Using LLMs Safely Authors: Siying Hu, Yaxing Yao, Zhicong Lu Published: 2025-10-02 Categories: cs.HC, cs.AI arXiv: https://arxiv.org/abs/2510.01638v1 Large Language Models are profoundly changing work patterns in high-risk professional domains, yet their application also introduces severe and underexplored compliance risks. To investigate this issue, we conducted semi-structured interviews with 24 highly-skilled knowledge workers from industries such as law, healthcare, and finance. The study found that these experts are commonly concerned about sensitive information leakage, intellectual property infringement, and uncertainty regarding the quality of model outputs. In response, they spontaneously adopt various mitigation strategies, such as actively distorting input data and limiting the details in their prompts. However, the effectiveness of these spontaneous efforts is limited due to a lack of specific compliance guidance and training for Large Language Models. Our research reveals a significant gap between current NLP tools and the actual compliance needs of experts. This paper positions these valuable empirical findings as foundational work for building the next generation of Human-Centered, Compliance-Driven Natural Language Processing for Regulatory Technology (RegTech), providing a critical human-centered perspective and design requirements for engineering NLP systems that can proactively support expert compliance workflows. --- ### Detecting Post-generation Edits to Watermarked LLM Outputs via Combinatorial Watermarking Authors: Liyan Xie, Muhammad Siddeek, Mohamed Seif, Andrea J. Goldsmith, Mengdi Wang Published: 2025-10-02 Categories: cs.LG arXiv: https://arxiv.org/abs/2510.01637v1 Watermarking has become a key technique for proprietary language models, enabling the distinction between AI-generated and human-written text. However, in many real-world scenarios, LLM-generated content may undergo post-generation edits, such as human revisions or even spoofing attacks, making it critical to detect and localize such modifications. In this work, we introduce a new task: detecting post-generation edits locally made to watermarked LLM outputs. To this end, we propose a combinatorial pattern-based watermarking framework, which partitions the vocabulary into disjoint subsets and embeds the watermark by enforcing a deterministic combinatorial pattern over these subsets during generation. We accompany the combinatorial watermark with a global statistic that can be used to detect the watermark. Furthermore, we design lightweight local statistics to flag and localize potential edits. We introduce two task-specific evaluation metrics, Type-I error rate and detection accuracy, and evaluate our method on open-source LLMs across a variety of editing scenarios, demonstrating strong empirical performance in edit localization. --- ### CAT: Curvature-Adaptive Transformers for Geometry-Aware Learning Authors: Ryan Y. Lin, Siddhartha Ojha, Nicholas Bai Published: 2025-10-02 Categories: cs.LG arXiv: https://arxiv.org/abs/2510.01634v1 Transformers achieve strong performance across diverse domains but implicitly assume Euclidean geometry in their attention mechanisms, limiting their effectiveness on data with non-Euclidean structure. While recent extensions to hyperbolic and spherical spaces show promise for hierarchical and cyclical patterns, respectively, they require committing to a single geometry a priori, reducing flexibility when data exhibits mixed geometric properties. We introduce the Curvature-Adaptive Transformer (CAT), a novel architecture that dynamically learns per-token routing across three geometric attention branches through a lightweight, differentiable gating mechanism. Unlike fixed-geometry approaches, CAT enables adaptive geometric specialization, routing tokens to the appropriate curvature based on their local relational structure. The routing network provides interpretable curvature preferences while each branch employs geometry-specific operations optimized for its respective manifold. On knowledge graph completion benchmarks (FB15k-237, WN18RR), CAT achieves approximately 10% improvements in MRR and Hits@10 over fixed-geometry baselines with minimal overhead (5% parameter increase, comparable inference time). These results demonstrate that learned geometric adaptation outperforms any single fixed geometry for complex relational reasoning, establishing CAT as a scalable and interpretable foundation for mixture-of-geometry architectures across language, vision, and multimodal domains. --- ### BioBlobs: Unsupervised Discovery of Functional Substructures for Protein Function Prediction Authors: Xin Wang, Kaiwen Shi, Carlos Oliver Published: 2025-10-02 Categories: q-bio.BM, cs.AI arXiv: https://arxiv.org/abs/2510.01632v3 Protein function is driven by cohesive substructures, such as catalytic triads, binding pockets, and structural motifs, that occupy only a small fraction of a protein's residues. Yet existing pipelines built on protein encoders do not model proteins at the substructure level, leaving the central biological question unanswered: which substructure of a protein is responsible for its function? We introduce BioBlobs, an encoder-agnostic, end-to-end differentiable framework that compresses a protein into a small set of cohesive substructures (blobs) and predicts function from these blobs alone, so that each blob corresponds to a candidate functional region. Across diverse protein function prediction tasks and multiple sequence- and structure-based encoders, BioBlobs matches or exceeds strong baselines while operating on only a small fraction of residues. The discovered blobs adapt their spatial scale to the task, ranging from local catalytic sites to entire structural domains. Trained only on protein-level labels, BioBlobs recovers experimentally annotated catalytic sites in the M-CSA database, demonstrating unsupervised functional substructure discovery and opening a path to large-scale functional site discovery across the unannotated proteome. --- ### Demystifying Synthetic Data in LLM Pre-training: A Systematic Study of Scaling Laws, Benefits, and Pitfalls Authors: Feiyang Kang, Newsha Ardalani, Michael Kuchnik, Youssef Emad, Mostafa Elhoushi, Shubhabrata Sengupta, Shang-Wen Li, Ramya Raghavendra, Ruoxi Jia, Carole-Jean Wu Published: 2025-10-02 Categories: cs.LG, cs.AI, cs.CL arXiv: https://arxiv.org/abs/2510.01631v1 Training data plays a crucial role in Large Language Models (LLM) scaling, yet high quality data is of limited supply. Synthetic data techniques offer a potential path toward sidestepping these limitations. We conduct a large-scale empirical investigation (>1000 LLMs with >100k GPU hours) using a unified protocol and scaling laws, comparing natural web data, diverse synthetic types (rephrased text, generated textbooks), and mixtures of natural and synthetic data. Specifically, we found pre-training on rephrased synthetic data \textit{alone} is not faster than pre-training on natural web texts; while pre-training on 1/3 rephrased synthetic data mixed with 2/3 natural web texts can speed up 5-10x (to reach the same validation loss) at larger data budgets. Pre-training on textbook-style synthetic data \textit{alone} results in notably higher loss on many downstream domains especially at small data budgets. "Good" ratios of synthetic data in training data mixtures depend on the model size and data budget, empirically converging to ~30% for rephrased synthetic data. Larger generator models do not necessarily yield better pre-training data than ~8B-param models. These results contribute mixed evidence on "model collapse" during large-scale single-round (n=1) model training on synthetic data--training on rephrased synthetic data shows no degradation in performance in foreseeable scales whereas training on mixtures of textbook-style pure-generated synthetic data shows patterns predicted by "model collapse". Our work demystifies synthetic data in pre-training, validates its conditional benefits, and offers practical guidance. --- ### Quagmires in SFT-RL Post-Training: When High SFT Scores Mislead and What to Use Instead Authors: Feiyang Kang, Michael Kuchnik, Karthik Padthe, Marin Vlastelica, Ruoxi Jia, Carole-Jean Wu, Newsha Ardalani Published: 2025-10-02 Categories: cs.LG, cs.AI, cs.CL arXiv: https://arxiv.org/abs/2510.01624v1 In post-training for reasoning Large Language Models (LLMs), the current state of practice trains LLMs in two independent stages: Supervised Fine-Tuning (SFT) and Reinforcement Learning with Verifiable Rewards (RLVR, shortened as ``RL'' below). In this work, we challenge whether high SFT scores translate to improved performance after RL. We provide extensive counter-examples where this is not true. We find high SFT scores can be biased toward simpler or more homogeneous data and are not reliably predictive of subsequent RL gains or scaled-up post-training effectiveness. In some cases, RL training on models with improved SFT performance could lead to substantially worse outcome compared to RL on the base model without SFT. We study alternative metrics and identify generalization loss on held-out reasoning examples and Pass@large k performance to provide strong proxies for the RL outcome. We trained hundreds of models up to 12B-parameter with SFT and RLVR via GRPO and ran extensive evaluations on 7 math benchmarks with up to 256 repetitions, spending $>$1M GPU hours. Experiments include models from Llama3, Mistral-Nemo, Qwen3 and multiple state-of-the-art SFT/RL datasets. Compared to directly predicting from pre-RL performance, prediction based on generalization loss and Pass@large k achieves substantial higher precision, improving $R^2$ coefficient and Spearman's rank correlation coefficient by up to 0.5 (2x). This provides strong utility for broad use cases. For example, in most experiments, we find SFT training on unique examples for a one epoch underperforms training on half examples for two epochs, either after SFT or SFT-then-RL; With the same SFT budget, training only on short examples may lead to better SFT performance, though, it often leads to worse outcome after RL compared to training on examples with varying lengths. Evaluation tool will be open-sourced. --- ### VLA-R1: Enhancing Reasoning in Vision-Language-Action Models Authors: Angen Ye, Zeyu Zhang, Boyuan Wang, Xiaofeng Wang, Dapeng Zhang, Zheng Zhu Published: 2025-10-02 Categories: cs.CV, cs.RO arXiv: https://arxiv.org/abs/2510.01623v1 Vision-Language-Action (VLA) models aim to unify perception, language understanding, and action generation, offering strong cross-task and cross-scene generalization with broad impact on embodied AI. However, current VLA models often lack explicit step-by-step reasoning, instead emitting final actions without considering affordance constraints or geometric relations. Their post-training pipelines also rarely reinforce reasoning quality, relying primarily on supervised fine-tuning with weak reward design. To address these challenges, we present VLA-R1, a reasoning-enhanced VLA that integrates Reinforcement Learning from Verifiable Rewards (RLVR) with Group Relative Policy Optimization (GRPO) to systematically optimize both reasoning and execution. Specifically, we design an RLVR-based post-training strategy with verifiable rewards for region alignment, trajectory consistency, and output formatting, thereby strengthening reasoning robustness and execution accuracy. Moreover, we develop VLA-CoT-13K, a high-quality dataset that provides chain-of-thought supervision explicitly aligned with affordance and trajectory annotations. Furthermore, extensive evaluations on in-domain, out-of-domain, simulation, and real-robot platforms demonstrate that VLA-R1 achieves superior generalization and real-world performance compared to prior VLA methods. We plan to release the model, code, and dataset following the publication of this work. Code: https://github.com/GigaAI-research/VLA-R1. Website: https://gigaai-research.github.io/VLA-R1. --- ### LLM4Rec: Large Language Models for Multimodal Generative Recommendation with Causal Debiasing Authors: Bo Ma, Hang Li, ZeHua Hu, XiaoFan Gui, LuYao Liu, Simon Lau Published: 2025-10-02 Categories: cs.IR, cs.AI, cs.CL arXiv: https://arxiv.org/abs/2510.01622v1 Contemporary generative recommendation systems face significant challenges in handling multimodal data, eliminating algorithmic biases, and providing transparent decision-making processes. This paper introduces an enhanced generative recommendation framework that addresses these limitations through five key innovations: multimodal fusion architecture, retrieval-augmented generation mechanisms, causal inference-based debiasing, explainable recommendation generation, and real-time adaptive learning capabilities. Our framework leverages advanced large language models as the backbone while incorporating specialized modules for cross-modal understanding, contextual knowledge integration, bias mitigation, explanation synthesis, and continuous model adaptation. Extensive experiments on three benchmark datasets (MovieLens-25M, Amazon-Electronics, Yelp-2023) demonstrate consistent improvements in recommendation accuracy, fairness, and diversity compared to existing approaches. The proposed framework achieves up to 2.3% improvement in NDCG@10 and 1.4% enhancement in diversity metrics while maintaining computational efficiency through optimized inference strategies. --- ### Posterior Collapse as a Phase Transition in Variational Autoencoders Authors: Zhen Li, Fan Zhang, Zheng Zhang, Yu Chen Published: 2025-10-02 Categories: cs.LG arXiv: https://arxiv.org/abs/2510.01621v2 We investigate the phenomenon of posterior collapse in variational autoencoders (VAEs) from the perspective of statistical physics, and reveal that it constitutes a phase transition governed jointly by data structure and model hyper-parameters. By analyzing the stability of the trivial solution associated with posterior collapse, we identify a critical hyper-parameter threshold. In particular, we derive an explicit criterion for the onset of collapse: posterior collapse occurs when the decoder variance exceeds the largest eigenvalue of the data covariance matrix. This critical boundary, separating meaningful latent inference from collapse, is characterized by a discontinuity in the KL divergence between the approximate posterior and the prior distribution, where the KL divergence and its derivatives exhibit clear non-analytic behavior. We validate this critical behavior on both synthetic and real-world datasets, confirming the existence of a phase transition. The experimental results align well with our theoretical predictions, demonstrating the robustness of our collapse criterion across various VAE architectures. Our stability-based analysis demonstrate that posterior collapse is not merely an optimization failure, but rather an emerging phase transition arising from the interplay between data structure and variational constraints. This perspective offers new insights into the trainability and representational capacity of deep generative models. --- ### Learning to Decide with Just Enough: Information-Theoretic Context Summarization for CMDPs Authors: Peidong Liu, Junjiang Lin, Shaowen Wang, Yao Xu, Haiqing Li, Xuhao Xie, Siyi Wu, Hao Li Published: 2025-10-02 Categories: cs.AI arXiv: https://arxiv.org/abs/2510.01620v2 Contextual Markov Decision Processes (CMDPs) offer a framework for sequential decision-making under external signals, but existing methods often fail to generalize in high-dimensional or unstructured contexts, resulting in excessive computation and unstable performance. We propose an information-theoretic summarization approach that uses large language models (LLMs) to compress contextual inputs into low-dimensional, semantically rich summaries. These summaries augment states by preserving decision-critical cues while reducing redundancy. Building on the notion of approximate context sufficiency, we provide, to our knowledge, the first regret bounds and a latency-entropy trade-off characterization for CMDPs. Our analysis clarifies how informativeness impacts computational cost. Experiments across discrete, continuous, visual, and recommendation benchmarks show that our method outperforms raw-context and non-context baselines, improving reward, success rate, and sample efficiency, while reducing latency and memory usage. These findings demonstrate that LLM-based summarization offers a scalable and interpretable solution for efficient decision-making in context-rich, resource-constrained environments. --- ### MPMAvatar: Learning 3D Gaussian Avatars with Accurate and Robust Physics-Based Dynamics Authors: Changmin Lee, Jihyun Lee, Tae-Kyun Kim Published: 2025-10-02 Categories: cs.GR, cs.CV arXiv: https://arxiv.org/abs/2510.01619v1 While there has been significant progress in the field of 3D avatar creation from visual observations, modeling physically plausible dynamics of humans with loose garments remains a challenging problem. Although a few existing works address this problem by leveraging physical simulation, they suffer from limited accuracy or robustness to novel animation inputs. In this work, we present MPMAvatar, a framework for creating 3D human avatars from multi-view videos that supports highly realistic, robust animation, as well as photorealistic rendering from free viewpoints. For accurate and robust dynamics modeling, our key idea is to use a Material Point Method-based simulator, which we carefully tailor to model garments with complex deformations and contact with the underlying body by incorporating an anisotropic constitutive model and a novel collision handling algorithm. We combine this dynamics modeling scheme with our canonical avatar that can be rendered using 3D Gaussian Splatting with quasi-shadowing, enabling high-fidelity rendering for physically realistic animations. In our experiments, we demonstrate that MPMAvatar significantly outperforms the existing state-of-the-art physics-based avatar in terms of (1) dynamics modeling accuracy, (2) rendering accuracy, and (3) robustness and efficiency. Additionally, we present a novel application in which our avatar generalizes to unseen interactions in a zero-shot manner-which was not achievable with previous learning-based methods due to their limited simulation generalizability. Our project page is at: https://KAISTChangmin.github.io/MPMAvatar/ --- ### Automated Genomic Interpretation via Concept Bottleneck Models for Medical Robotics Authors: Zijun Li, Jinchang Zhang, Ming Zhang, Guoyu Lu Published: 2025-10-02 Categories: cs.CV, q-bio.OT arXiv: https://arxiv.org/abs/2510.01618v2 We propose an automated genomic interpretation module that transforms raw DNA sequences into actionable, interpretable decisions suitable for integration into medical automation and robotic systems. Our framework combines Chaos Game Representation (CGR) with a Concept Bottleneck Model (CBM), enforcing predictions to flow through biologically meaningful concepts such as GC content, CpG density, and k mer motifs. To enhance reliability, we incorporate concept fidelity supervision, prior consistency alignment, KL distribution matching, and uncertainty calibration. Beyond accurate classification of HIV subtypes across both in-house and LANL datasets, our module delivers interpretable evidence that can be directly validated against biological priors. A cost aware recommendation layer further translates predictive outputs into decision policies that balance accuracy, calibration, and clinical utility, reducing unnecessary retests and improving efficiency. Extensive experiments demonstrate that the proposed system achieves state of the art classification performance, superior concept prediction fidelity, and more favorable cost benefit trade-offs compared to existing baselines. By bridging the gap between interpretable genomic modeling and automated decision-making, this work establishes a reliable foundation for robotic and clinical automation in genomic medicine. --- ### AMAS: Adaptively Determining Communication Topology for LLM-based Multi-Agent System Authors: Hui Yi Leong, Yuheng Li, Yuqing Wu, Wenwen Ouyang, Wei Zhu, Jiechao Gao, Wei Han Published: 2025-10-02 Categories: cs.CL arXiv: https://arxiv.org/abs/2510.01617v3 Although large language models (LLMs) have revolutionized natural language processing capabilities, their practical implementation as autonomous multi-agent systems (MAS) for industrial problem-solving encounters persistent barriers. Conventional MAS architectures are fundamentally restricted by inflexible, hand-crafted graph topologies that lack contextual responsiveness, resulting in diminished efficacy across varied academic and commercial workloads. To surmount these constraints, we introduce AMAS, a paradigm-shifting framework that redefines LLM-based MAS through a novel dynamic graph designer. This component autonomously identifies task-specific optimal graph configurations via lightweight LLM adaptation, eliminating the reliance on monolithic, universally applied structural templates. Instead, AMAS exploits the intrinsic properties of individual inputs to intelligently direct query trajectories through task-optimized agent pathways. Rigorous validation across question answering, mathematical deduction, and code generation benchmarks confirms that AMAS systematically exceeds state-of-the-art single-agent and multi-agent approaches across diverse LLM architectures. Our investigation establishes that context-sensitive structural adaptability constitutes a foundational requirement for high-performance LLM MAS deployments. --- ### Efficient Training of Robust Traditional Chinese LLaMA-1B on a Single Consumer GPU: Continual Pre-training, SFT, and DPO Authors: Yu-Cheng Chih, Ming-Tao Duan, Yong-Hao Hou Published: 2025-10-02 Categories: cs.CL arXiv: https://arxiv.org/abs/2510.01616v1 Small Language Models (SLMs) enable cost-effective, on-device and latency-sensitive AI applications, yet their deployment in Traditional Chinese (TC) remains hindered by token-level instability - models unpredictably emit non-TC characters or code-switch into other languages. We address this practical reliability gap by creating PureTC-1B, a three-stage stabilization pipeline for Llama-3.2-1B-Instruct (an open-weight, instruction-tuned model released by Meta) using parameter-efficient LoRA adapters. Our method combines Continual Pre-Training (CPT) on TC-centric corpora, Supervised Fine-Tuning (SFT) with instruction data, and Direct Preference Optimization (DPO) using TC-adherence preferences to improve monolingual robustness without full-model retraining. On a benchmark designed to simulate real-world usage, PureTC-1B achieves a 51.3% relative reduction (micro-average) in non-TC output tokens versus the base model. On a Named Entity Translation (NET) task, PureTC-1B further reduces incorrect-language tokens by 77.2% relative to Llama-3B and 57.2% relative to Qwen-1.5B, indicating that robust TC adherence is attainable even at the 1B scale. The pipeline is reproducible, adapter-only, and hardware-friendly, offering practitioners a practical recipe to enhance language stability for TC and potentially other non-English languages. --- ### RAG-BioQA: A Retrieval-Augmented Generation Framework for Long-Form Biomedical Question Answering Authors: Lovely Yeswanth Panchumarthi, Sumalatha Saleti, Sai Prasad Gudari, Atharva Negi, Praveen Raj Budime, Harsit Upadhya Published: 2025-10-02 Categories: cs.CL, cs.AI arXiv: https://arxiv.org/abs/2510.01612v3 The rapidly growth of biomedical literature creates challenges acquiring specific medical information. Current biomedical question-answering systems primarily focus on short-form answers, failing to provide comprehensive explanations necessary for clinical decision-making. We present RAG-BioQA, a retrieval-augmented generation framework for long-form biomedical question answering. Our system integrates BioBERT embeddings with FAISS indexing for retrieval and a LoRA fine-tuned FLAN-T5 model for answer generation. We train on 181k QA pairs from PubMedQA, MedDialog, and MedQuAD, and evaluate on a held-out PubMedQA test set. We compare four retrieval strategies: dense retrieval (FAISS), BM25, ColBERT, and MonoT5. Our results show that domain-adapted dense retrieval outperforms zero-shot neural re-rankers, with the best configuration achieving 0.24 BLEU-1 and 0.29 ROUGE-1. Fine-tuning improves BERTScore by 81\% over the base model. We release our framework to support reproducible biomedical QA research. --- ### PsychCounsel-Bench: Evaluating the Psychology Intelligence of Large Language Models Authors: Min Zeng Published: 2025-10-02 Categories: cs.AI, cs.CL arXiv: https://arxiv.org/abs/2510.01611v4 Large Language Models (LLMs) have demonstrated remarkable success across a wide range of industries, primarily due to their impressive generative abilities. Yet, their potential in applications requiring cognitive abilities, such as psychological counseling, remains largely untapped. This paper investigates the key question: \textit{Can LLMs be effectively applied to psychological counseling?} To determine whether an LLM can effectively take on the role of a psychological counselor, the first step is to assess whether it meets the qualifications required for such a role, namely the ability to pass the U.S. National Counselor Certification Exam (NCE). This is because, just as a human counselor must pass a certification exam to practice, an LLM must demonstrate sufficient psychological knowledge to meet the standards required for such a role. To address this, we introduce PsychCounsel-Bench, a benchmark grounded in U.S.national counselor examinations, a licensure test for professional counselors that requires about 70\% accuracy to pass. PsychCounsel-Bench comprises approximately 2,252 carefully curated single-choice questions, crafted to require deep understanding and broad enough to cover various sub-disciplines of psychology. This benchmark provides a comprehensive assessment of an LLM's ability to function as a counselor. Our evaluation shows that advanced models such as GPT-4o, Llama3.3-70B, and Gemma3-27B achieve well above the passing threshold, while smaller open-source models (e.g., Qwen2.5-7B, Mistral-7B) remain far below it. These results suggest that only frontier LLMs are currently capable of meeting counseling exam standards, highlighting both the promise and the challenges of developing psychology-oriented LLMs. We release the proposed dataset for public use: https://github.com/cloversjtu/PsychCounsel-Bench --- ### AgentRec: Next-Generation LLM-Powered Multi-Agent Collaborative Recommendation with Adaptive Intelligence Authors: Bo Ma, Hang Li, ZeHua Hu, XiaoFan Gui, LuYao Liu, Simon Lau Published: 2025-10-02 Categories: cs.AI arXiv: https://arxiv.org/abs/2510.01609v1 Interactive conversational recommender systems have gained significant attention for their ability to capture user preferences through natural language interactions. However, existing approaches face substantial challenges in handling dynamic user preferences, maintaining conversation coherence, and balancing multiple ranking objectives simultaneously. This paper introduces AgentRec, a next-generation LLM-powered multi-agent collaborative recommendation framework that addresses these limitations through hierarchical agent networks with adaptive intelligence. Our approach employs specialized LLM-powered agents for conversation understanding, preference modeling, context awareness, and dynamic ranking, coordinated through an adaptive weighting mechanism that learns from interaction patterns. We propose a three-tier learning strategy combining rapid response for simple queries, intelligent reasoning for complex preferences, and deep collaboration for challenging scenarios. Extensive experiments on three real-world datasets demonstrate that AgentRec achieves consistent improvements over state-of-the-art baselines, with 2.8\% enhancement in conversation success rate, 1.9\% improvement in recommendation accuracy (NDCG@10), and 3.2\% better conversation efficiency while maintaining comparable computational costs through intelligent agent coordination. --- ### NPN: Non-Linear Projections of the Null-Space for Imaging Inverse Problems Authors: Roman Jacome, Romario Gualdrón-Hurtado, Leon Suarez, Henry Arguello Published: 2025-10-02 Categories: cs.CV, eess.SP, math.OC arXiv: https://arxiv.org/abs/2510.01608v2 Imaging inverse problems aim to recover high-dimensional signals from undersampled, noisy measurements, a fundamentally ill-posed task with infinite solutions in the null-space of the sensing operator. To resolve this ambiguity, prior information is typically incorporated through handcrafted regularizers or learned models that constrain the solution space. However, these priors typically ignore the task-specific structure of that null-space. In this work, we propose Non-Linear Projections of the Null-Space (NPN), a novel class of regularization that, instead of enforcing structural constraints in the image domain, promotes solutions that lie in a low-dimensional projection of the sensing matrix's null-space with a neural network. Our approach has two key advantages: (1) Interpretability: by focusing on the structure of the null-space, we design sensing-matrix-specific priors that capture information orthogonal to the signal components that are fundamentally blind to the sensing process. (2) Flexibility: NPN is adaptable to various inverse problems, compatible with existing reconstruction frameworks, and complementary to conventional image-domain priors. We provide theoretical guarantees on convergence and reconstruction accuracy when used within plug-and-play methods. Empirical results across diverse sensing matrices demonstrate that NPN priors consistently enhance reconstruction fidelity in various imaging inverse problems, such as compressive sensing, deblurring, super-resolution, computed tomography, and magnetic resonance imaging, with plug-and-play methods, unrolling networks, deep image prior, and diffusion models. --- ### ActiveUMI: Robotic Manipulation with Active Perception from Robot-Free Human Demonstrations Authors: Qiyuan Zeng, Chengmeng Li, Jude St. John, Zhongyi Zhou, Junjie Wen, Guorui Feng, Yichen Zhu, Yi Xu Published: 2025-10-02 Categories: cs.RO, cs.CV arXiv: https://arxiv.org/abs/2510.01607v1 We present ActiveUMI, a framework for a data collection system that transfers in-the-wild human demonstrations to robots capable of complex bimanual manipulation. ActiveUMI couples a portable VR teleoperation kit with sensorized controllers that mirror the robot's end-effectors, bridging human-robot kinematics via precise pose alignment. To ensure mobility and data quality, we introduce several key techniques, including immersive 3D model rendering, a self-contained wearable computer, and efficient calibration methods. ActiveUMI's defining feature is its capture of active, egocentric perception. By recording an operator's deliberate head movements via a head-mounted display, our system learns the crucial link between visual attention and manipulation. We evaluate ActiveUMI on six challenging bimanual tasks. Policies trained exclusively on ActiveUMI data achieve an average success rate of 70\% on in-distribution tasks and demonstrate strong generalization, retaining a 56\% success rate when tested on novel objects and in new environments. Our results demonstrate that portable data collection systems, when coupled with learned active perception, provide an effective and scalable pathway toward creating generalizable and highly capable real-world robot policies. --- ### Bridging Collaborative Filtering and Large Language Models with Dynamic Alignment, Multimodal Fusion and Evidence-grounded Explanations Authors: Bo Ma, LuYao Liu, Simon Lau, Chandler Yuan, and XueY Cui, Rosie Zhang Published: 2025-10-02 Categories: cs.IR, cs.AI, cs.CL arXiv: https://arxiv.org/abs/2510.01606v1 Recent research has explored using Large Language Models for recommendation tasks by transforming user interaction histories and item metadata into text prompts, then having the LLM produce rankings or recommendations. A promising approach involves connecting collaborative filtering knowledge to LLM representations through compact adapter networks, which avoids expensive fine-tuning while preserving the strengths of both components. Yet several challenges persist in practice: collaborative filtering models often use static snapshots that miss rapidly changing user preferences; many real-world items contain rich visual and audio content beyond textual descriptions; and current systems struggle to provide trustworthy explanations backed by concrete evidence. Our work introduces \model{}, a framework that tackles these limitations through three key innovations. We develop an online adaptation mechanism that continuously incorporates new user interactions through lightweight modules, avoiding the need to retrain large models. We create a unified representation that seamlessly combines collaborative signals with visual and audio features, handling cases where some modalities may be unavailable. Finally, we design an explanation system that grounds recommendations in specific collaborative patterns and item attributes, producing natural language rationales users can verify. Our approach maintains the efficiency of frozen base models while adding minimal computational overhead, making it practical for real-world deployment. --- ### A Comparison of Independent and Joint Fine-tuning Strategies for Retrieval-Augmented Generation Authors: Neal Gregory Lawton, Alfy Samuel, Anoop Kumar, Daben Liu Published: 2025-10-02 Categories: cs.CL, cs.AI arXiv: https://arxiv.org/abs/2510.01600v2 A Comparison of Independent and Joint Fine-tuning Strategies for Retrieval-Augmented Generation Download PDF Neal Gregory Lawton, Alfy Samuel, Anoop Kumar, Daben Liu Published: 20 Aug 2025, Retrieval augmented generation (RAG) is a popular framework for question answering that is powered by two large language models (LLMs): an embedding model that retrieves context documents from a database that are relevant to a given question, and a generator model that uses the retrieved context to generate an answer to the question. Both the embedding and generator models can be fine-tuned to increase performance of a RAG pipeline on a new task, but multiple fine-tuning strategies exist with different costs and benefits. In this paper, we evaluate and compare several RAG fine-tuning strategies, including independent, joint, and two-phase fine-tuning. In our experiments, we observe that all of these strategies achieve about equal improvement in EM and F1 generation quality metrics, although they have significantly different computational costs. We conclude the optimal fine-tuning strategy to use depends on whether the training dataset includes context labels and whether a grid search over the learning rates for the embedding and generator models is required. --- ### Securing generative artificial intelligence with parallel magnetic tunnel junction true randomness Authors: Youwei Bao, Shuhan Yang, Hyunsoo Yang Published: 2025-10-02 Categories: cs.LG, cond-mat.mtrl-sci, physics.data-an arXiv: https://arxiv.org/abs/2510.01598v1 Deterministic pseudo random number generators (PRNGs) used in generative artificial intelligence (GAI) models produce predictable patterns vulnerable to exploitation by attackers. Conventional defences against the vulnerabilities often come with significant energy and latency overhead. Here, we embed hardware-generated true random bits from spin-transfer torque magnetic tunnel junctions (STT-MTJs) to address the challenges. A highly parallel, FPGA-assisted prototype computing system delivers megabit-per-second true random numbers, passing NIST randomness tests after in-situ operations with minimal overhead. Integrating the hardware random bits into a generative adversarial network (GAN) trained on CIFAR-10 reduces insecure outputs by up to 18.6 times compared to the low-quality random number generators (RNG) baseline. With nanosecond switching speed, high energy efficiency, and established scalability, our STT-MTJ-based system holds the potential to scale beyond 106 parallel cells, achieving gigabit-per-second throughput suitable for large language model sampling. This advancement highlights spintronic RNGs as practical security components for next-generation GAI systems. --- ### CLUE: Non-parametric Verification from Experience via Hidden-State Clustering Authors: Zhenwen Liang, Ruosen Li, Yujun Zhou, Linfeng Song, Dian Yu, Xinya Du, Haitao Mi, Dong Yu Published: 2025-10-02 Categories: cs.CL arXiv: https://arxiv.org/abs/2510.01591v1 Assessing the quality of Large Language Model (LLM) outputs presents a critical challenge. Previous methods either rely on text-level information (e.g., reward models, majority voting), which can overfit to superficial cues, or on calibrated confidence from token probabilities, which would fail on less-calibrated models. Yet both of these signals are, in fact, partial projections of a richer source of information: the model's internal hidden states. Early layers, closer to token embeddings, preserve semantic and lexical features that underpin text-based judgments, while later layers increasingly align with output logits, embedding confidence-related information. This paper explores hidden states directly as a unified foundation for verification. We show that the correctness of a solution is encoded as a geometrically separable signature within the trajectory of hidden activations. To validate this, we present Clue (Clustering and Experience-based Verification), a deliberately minimalist, non-parametric verifier. With no trainable parameters, CLUE only summarizes each reasoning trace by an hidden state delta and classifies correctness via nearest-centroid distance to ``success'' and ``failure'' clusters formed from past experience. The simplicity of this method highlights the strength of the underlying signal. Empirically, CLUE consistently outperforms LLM-as-a-judge baselines and matches or exceeds modern confidence-based methods in reranking candidates, improving both top-1 and majority-vote accuracy across AIME 24/25 and GPQA. As a highlight, on AIME 24 with a 1.5B model, CLUE boosts accuracy from 56.7% (majority@64) to 70.0% (top-maj@16). --- ### Enhancing Noise Robustness of Parkinson's Disease Telemonitoring via Contrastive Feature Augmentation Authors: Ziming Tang, Chengbin Hou, Tianyu Zhang, Bangxu Tian, Jinbao Wang, Hairong Lv Published: 2025-10-02 Categories: cs.LG, cs.AI arXiv: https://arxiv.org/abs/2510.01588v1 Parkinson's disease (PD) is one of the most common neurodegenerative disorder. PD telemonitoring emerges as a novel assessment modality enabling self-administered at-home tests of Unified Parkinson's Disease Rating Scale (UPDRS) scores, enhancing accessibility for PD patients. However, three types of noise would occur during measurements: (1) patient-induced measurement inaccuracies, (2) environmental noise, and (3) data packet loss during transmission, resulting in higher prediction errors. To address these challenges, NoRo, a noise-robust UPDRS prediction framework is proposed. First, the original speech features are grouped into ordered bins, based on the continuous values of a selected feature, to construct contrastive pairs. Second, the contrastive pairs are employed to train a multilayer perceptron encoder for generating noise-robust features. Finally, these features are concatenated with the original features as the augmented features, which are then fed into the UPDRS prediction models. Notably, we further introduces a novel evaluation approach with customizable noise injection module, and extensive experiments show that NoRo can successfully enhance the noise robustness of UPDRS prediction across various downstream prediction models under different noisy environments. --- ### AdvEvo-MARL: Shaping Internalized Safety through Adversarial Co-Evolution in Multi-Agent Reinforcement Learning Authors: Zhenyu Pan, Yiting Zhang, Zhuo Liu, Yolo Yunlong Tang, Zeliang Zhang, Haozheng Luo, Yuwei Han, Jianshu Zhang, Dennis Wu, Hong-Yu Chen, Haoran Lu, Haoyang Fang et al. Published: 2025-10-02 Categories: cs.AI arXiv: https://arxiv.org/abs/2510.01586v1 LLM-based multi-agent systems excel at planning, tool use, and role coordination, but their openness and interaction complexity also expose them to jailbreak, prompt-injection, and adversarial collaboration. Existing defenses fall into two lines: (i) self-verification that asks each agent to pre-filter unsafe instructions before execution, and (ii) external guard modules that police behaviors. The former often underperforms because a standalone agent lacks sufficient capacity to detect cross-agent unsafe chains and delegation-induced risks; the latter increases system overhead and creates a single-point-of-failure-once compromised, system-wide safety collapses, and adding more guards worsens cost and complexity. To solve these challenges, we propose AdvEvo-MARL, a co-evolutionary multi-agent reinforcement learning framework that internalizes safety into task agents. Rather than relying on external guards, AdvEvo-MARL jointly optimizes attackers (which synthesize evolving jailbreak prompts) and defenders (task agents trained to both accomplish their duties and resist attacks) in adversarial learning environments. To stabilize learning and foster cooperation, we introduce a public baseline for advantage estimation: agents within the same functional group share a group-level mean-return baseline, enabling lower-variance updates and stronger intra-group coordination. Across representative attack scenarios, AdvEvo-MARL consistently keeps attack-success rate (ASR) below 20%, whereas baselines reach up to 38.33%, while preserving-and sometimes improving-task accuracy (up to +3.67% on reasoning tasks). These results show that safety and utility can be jointly improved without relying on extra guard agents or added system overhead. --- ### ReSSFormer: A Recursive Sparse Structured Transformer for Scalable and Long-Context Reasoning Authors: Haochen You, Baojing Liu Published: 2025-10-02 Categories: cs.CL, cs.NI arXiv: https://arxiv.org/abs/2510.01585v1 While Transformer architectures have demonstrated impressive scalability across domains, they continue to face challenges in long-context reasoning, computational efficiency, and structural generalization - largely due to rigid layer stacking, dense attention, and reliance on positional encodings. We present ReSSFormer, a Recursive Sparse Structured Transformer that integrates three complementary innovations: Recurrent Reasoning & Memory Unit (R2MU) for iterative reasoning with bounded depth, Adaptive Sparse Attention Module (ASAM) for efficient and focused context selection, and Self-Organizing Encoder Structure (SOES) for position-free structure induction. ReSSFormer replaces conventional depth stacking with recurrent inference, substitutes full attention with token- and expert-level sparsity, and models latent token topology directly from content. Across language modeling, multi-hop QA, and structure-sensitive tasks, ReSSFormer consistently outperforms strong baselines under comparable FLOPs and parameter budgets, highlighting its scalability, efficiency, and structural flexibility. --- ### ImageNet-Think-250K: A Large-Scale Synthetic Dataset for Multimodal Reasoning for Vision Language Models Authors: Krishna Teja Chitty-Venkata, Murali Emani Published: 2025-10-02 Categories: cs.CV, cs.LG arXiv: https://arxiv.org/abs/2510.01582v1 We develop ImageNet-Think, a multimodal reasoning dataset designed to aid the development of Vision Language Models (VLMs) with explicit reasoning capabilities. Our dataset is built on 250,000 images from ImageNet21k dataset, providing structured thinking tokens and corresponding answers. Our synthetic dataset is generated by two state-of-the-art VLMs: GLM-4.1V-9B-Thinking and Kimi-VL-A3B-Thinking-2506. Each image is accompanied by two pairs of thinking-answer sequences, creating a resource for training and evaluating multimodal reasoning models. We capture the step-by-step reasoning process of VLMs and the final descriptive answers. Our goal with this dataset is to enable the development of more robust VLMs while contributing to the broader understanding of multimodal reasoning mechanisms. The dataset and evaluation benchmarks will be publicly available to aid research in reasoning/thinking multimodal VLMs. --- ### Think Right: Learning to Mitigate Under-Over Thinking via Adaptive, Attentive Compression Authors: Joykirat Singh, Justin Chih-Yao Chen, Archiki Prasad, Elias Stengel-Eskin, Akshay Nambi, Mohit Bansal Published: 2025-10-02 Categories: cs.LG, cs.AI, cs.CL arXiv: https://arxiv.org/abs/2510.01581v2 Recent thinking models are capable of solving complex reasoning tasks by scaling test-time compute, but this scaling should be allocated in line with task difficulty. On one hand, short reasoning (underthinking) leads to errors on harder problems that require extended reasoning steps; but, excessively long reasoning (overthinking) can be token-inefficient by generating unnecessary steps even after reaching a correct intermediate solution. We refer to this as under-adaptivity, where the model fails to modulate its response length appropriately given problems of varying difficulty. To address under-adaptivity and strike a balance between under- and overthinking, we propose TRAAC (Think Right with Adaptive, Attentive Compression), an online post-training RL method that leverages the model's self-attention to identify key steps and prune redundant ones. TRAAC also estimates difficulty and incorporates it into training rewards, thereby learning to allocate a reasoning budget commensurate with example difficulty. Across a variety of tasks (AIME, AMC, GPQA-D, BBEH), TRAAC (Qwen3-4B) achieves an average absolute accuracy gain of 8.4% with a relative reduction in reasoning length of 36.8% compared to the base model, and a 7.9% accuracy gain paired with a 29.4% length drop compared to the best RL baseline. TRAAC generalizes well, with accuracy and efficiency gains on out-of-distribution non-math datasets like GPQA-D, BBEH, and OptimalThinkingBench. Our analysis shows that TRAAC learns to adjust its thinking budget based on difficulty and that a combination of task-difficulty calibration and attention-based compression yields gains across diverse tasks. --- ### Gradient Shaping Beyond Clipping: A Functional Perspective on Update Magnitude Control Authors: Haochen You, Baojing Liu Published: 2025-10-02 Categories: cs.LG arXiv: https://arxiv.org/abs/2510.01578v1 Gradient clipping is widely used to stabilize deep network training, but its formulation as a hard, fixed threshold limits flexibility and ignores gradient distribution dynamics. We propose SPAMP (Statistical Per-layer Adaptive Modulation and Projection), a unified framework that generalizes clipping into smooth, per-layer gradient shaping. SPAMP tracks local gradient statistics, dynamically estimates thresholds, and applies power-based transformations to modulate update magnitudes in a differentiable manner. This perspective recasts clipping and warmup as dual mechanisms for controlling the effective update scale $η_t \|g_t\|$, offering a principled alternative to rigid heuristics. Extensive experiments across image and language tasks demonstrate that SPAMP improves stability, convergence, and robustness over existing methods. --- ### Guiding Multimodal Large Language Models with Blind and Low Vision People Visual Questions for Proactive Visual Interpretations Authors: Ricardo Gonzalez Penuela, Felipe Arias-Russi, Victor Capriles Published: 2025-10-02 Categories: cs.CV, cs.AI, cs.HC arXiv: https://arxiv.org/abs/2510.01576v1 Multimodal large language models (MLLMs) have been integrated into visual interpretation applications to support Blind and Low Vision (BLV) users because of their accuracy and ability to provide rich, human-like interpretations. However, these applications often default to comprehensive, lengthy descriptions regardless of context. This leads to inefficient exchanges, as users must go through irrelevant details rather than receiving the specific information they are likely to seek. To deliver more contextually-relevant information, we developed a system that draws on historical BLV users questions. When given an image, our system identifies similar past visual contexts from the VizWiz-LF dataset and uses the associated questions to guide the MLLM generate descriptions more relevant to BLV users. An evaluation with three human labelers who revised 92 context-aware and context-free descriptions showed that context-aware descriptions anticipated and answered users' questions in 76.1% of cases (70 out of 92) and were preferred in 54.4% of comparisons (50 out of 92). Our paper reviews, and data analysis are publicly available in a Github repository at https://github.com/rgonzalezp/guiding-multimodal-large-language-models-with-blind-and-low-vision-people-visual-questions . --- ### Synthetic Prefixes to Mitigate Bias in Real-Time Neural Query Autocomplete Authors: Adithya Rajan, Xiaoyu Liu, Prateek Verma, Vibhu Arora Published: 2025-10-02 Categories: cs.IR, cs.AI, cs.CL, cs.LG arXiv: https://arxiv.org/abs/2510.01574v1 We introduce a data-centric approach for mitigating presentation bias in real-time neural query autocomplete systems through the use of synthetic prefixes. These prefixes are generated from complete user queries collected during regular search sessions where autocomplete was not active. This allows us to enrich the training data for learning to rank models with more diverse and less biased examples. This method addresses the inherent bias in engagement signals collected from live query autocomplete interactions, where model suggestions influence user behavior. Our neural ranker is optimized for real-time deployment under strict latency constraints and incorporates a rich set of features, including query popularity, seasonality, fuzzy match scores, and contextual signals such as department affinity, device type, and vertical alignment with previous user queries. To support efficient training, we introduce a task-specific simplification of the listwise loss, reducing computational complexity from $O(n^2)$ to $O(n)$ by leveraging the query autocomplete structure of having only one ground-truth selection per prefix. Deployed in a large-scale e-commerce setting, our system demonstrates statistically significant improvements in user engagement, as measured by mean reciprocal rank and related metrics. Our findings show that synthetic prefixes not only improve generalization but also provide a scalable path toward bias mitigation in other low-latency ranking tasks, including related searches and query recommendations. --- ### From Supervision to Exploration: What Does Protein Language Model Learn During Reinforcement Learning? Authors: Hanqun Cao, Hongrui Zhang, Junde Xu, Zhou Zhang, Lingdong Shen, Minghao Sun, Ge Liu, Jinbo Xu, Wu-Jun Li, Jinren Ni, Cesar de la Fuente-Nunez, Tianfan Fu et al. Published: 2025-10-02 Categories: cs.LG, cs.AI, q-bio.BM arXiv: https://arxiv.org/abs/2510.01571v1 Protein language models (PLMs) have advanced computational protein science through large-scale pretraining and scalable architectures. In parallel, reinforcement learning (RL) has broadened exploration and enabled precise multi-objective optimization in protein design. Yet whether RL can push PLMs beyond their pretraining priors to uncover latent sequence-structure-function rules remains unclear. We address this by pairing RL with PLMs across four domains: antimicrobial peptide design, kinase variant optimization, antibody engineering, and inverse folding. Using diverse RL algorithms and model classes, we ask if RL improves sampling efficiency and, more importantly, if it reveals capabilities not captured by supervised learning. Across benchmarks, RL consistently boosts success rates and sample efficiency. Performance follows a three-factor interaction: task headroom, reward fidelity, and policy capacity jointly determine gains. When rewards are accurate and informative, policies have sufficient capacity, and tasks leave room beyond supervised baselines, improvements scale; when rewards are noisy or capacity is constrained, gains saturate despite exploration. This view yields practical guidance for RL in protein design: prioritize reward modeling and calibration before scaling policy size, match algorithm and regularization strength to task difficulty, and allocate capacity where marginal gains are largest. Implementation is available at https://github.com/chq1155/RL-PLM. --- ### InvThink: Premortem Reasoning for Safer Language Models Authors: Yubin Kim, Taehan Kim, Eugene Park, Chunjong Park, Cynthia Breazeal, Daniel McDuff, Hae Won Park Published: 2025-10-02 Categories: cs.AI, cs.CL arXiv: https://arxiv.org/abs/2510.01569v3 We present InvThink, a training and prompting framework that requires the model to enumerate, analyze, and constrain potential failures before generating its final response. Unlike existing safety alignment methods that optimize only for safe final responses, InvThink structures generation into three steps: (1) enumerate potential harms, (2) analyze their consequences, (3) generate the response under explicit mitigation constraints. We observe three findings: (i) InvThink shows higher safety scores at larger model sizes, compared to existing safety prompting and alignment baselines. (ii) InvThink mitigates the safety tax. Models trained with INVTHINK preserve their reasoning capability on standard benchmarks. (iii) beyond general safety tasks, InvThink also reduces harmful behavior in professional ethics domains (medicine, finance, law) and in agentic misalignment scenarios, achieving up to 32% reduction in harmfulness over zero-shot baselines and 16% over SafetyPrompt. We extend InvThink with supervised fine-tuning, and GRPO-based reinforcement learning across three LLM families. --- ### TetriServe: Efficiently Serving Mixed DiT Workloads Authors: Runyu Lu, Shiqi He, Wenxuan Tan, Shenggui Li, Ruofan Wu, Jeff J. Ma, Ang Chen, Mosharaf Chowdhury Published: 2025-10-02 Categories: cs.LG, cs.DC arXiv: https://arxiv.org/abs/2510.01565v4 Diffusion Transformer (DiT) models excel at generating high-quality images through iterative denoising steps, but serving them under strict Service Level Objectives (SLOs) is challenging due to their high computational cost, particularly at larger resolutions. Existing serving systems use fixed-degree sequence parallelism, which is inefficient for heterogeneous workloads with mixed resolutions and deadlines, leading to poor GPU utilization and low SLO attainment. In this paper, we propose step-level sequence parallelism to dynamically adjust the degree of parallelism of individual requests according to their deadlines. We present TetriServe, a DiT serving system that implements this strategy for highly efficient image generation. Specifically, TetriServe introduces a novel round-based scheduling mechanism that improves SLO attainment by (1) discretizing time into fixed rounds to make deadline-aware scheduling tractable, (2) adapting parallelism at the step level and minimizing GPU hour consumption, and (3) jointly packing requests to minimize late completions. Extensive evaluation on state-of-the-art DiT models shows that TetriServe achieves up to 32% higher SLO attainment compared to existing solutions without degrading image quality. ---