# Research Papers — 2026-03 · bundle page 227 of 243 > Full-text markdown of papers #11301–11350 of 12123 published in 2026-03, newest first. AI-readable mirrors of the full 12123-paper 2026-03 archive (50 papers per file). Canonical page: https://aipod.dev/papers/archive/2026-03/page/227 [← Previous bundle](https://aipod.dev/papers/archive/2026-03/page/226.md) [Next bundle →](https://aipod.dev/papers/archive/2026-03/page/228.md) ### VisionCreator: A Native Visual-Generation Agentic Model with Understanding, Thinking, Planning and Creation Authors: Jinxiang Lai, Zexin Lu, Jiajun He, Rongwei Quan, Wenzhe Zhao, Qinyu Yang, Qi Chen, Qin Lin, Chuyue Li, Tao Gao, Yuhao Shan, Shuai Shao et al. Published: 2026-03-03 Categories: cs.CV arXiv: https://arxiv.org/abs/2603.02681v1 Visual content creation tasks demand a nuanced understanding of design conventions and creative workflows-capabilities challenging for general models, while workflow-based agents lack specialized knowledge for autonomous creative planning. To overcome these challenges, we propose VisionCreator, a native visual-generation agentic model that unifies Understanding, Thinking, Planning, and Creation (UTPC) capabilities within an end-to-end learnable framework. Our work introduces four key contributions: (i) VisGenData-4k and its construction methodology using metacognition-based VisionAgent to generate high-quality creation trajectories with explicit UTPC structures; (ii) The VisionCreator agentic model, optimized through Progressive Specialization Training (PST) and Virtual Reinforcement Learning (VRL) within a high-fidelity simulated environment, enabling stable and efficient acquisition of UTPC capabilities for complex creation tasks; (iii) VisGenBench, a comprehensive benchmark featuring 1.2k test samples across diverse scenarios for standardized evaluation of multi-step visual creation capabilities; (iv) Remarkably, our VisionCreator-8B/32B models demonstrate superior performance over larger closed-source models across multiple evaluation dimensions. Overall, this work provides a foundation for future research in visual-generation agentic systems. --- ### LLMs for High-Frequency Decision-Making: Normalized Action Reward-Guided Consistency Policy Optimization Authors: Yang Zhao, Zihao Li, Zhiyu Jiang, Dandan Ma, Ganchao Liu, Wenzhe Zhao Published: 2026-03-03 Categories: cs.AI arXiv: https://arxiv.org/abs/2603.02680v1 While Large Language Models (LLMs) form the cornerstone of sequential decision-making agent development, they have inherent limitations in high-frequency decision tasks. Existing research mainly focuses on discrete embodied decision scenarios with low-frequency and significant semantic differences in state space (e.g., household planning). These methods suffer from limited performance in high-frequency decision-making tasks, since high-precision numerical state information in such tasks undergoes frequent updates with minimal fluctuations, and exhibiting policy misalignment between the learned sub-tasks and composite tasks. To address these issues, this paper proposes Normalized Action Reward guided Consistency Policy Optimization (NAR-CP). 1) Our method first acquires predefined dense rewards from environmental feedback of candidate actions via reward functions, then completes reward shaping through normalization, and theoretically verifies action reward normalization does not impair optimal policy. 2) To reduce policy misalignment in composite tasks, we use LLMs to infer sub-observation candidate actions and generate joint policies, with consistency loss ensuring precise alignment between global semantic policies and sub-semantic policies. Experiments on UAV pursuit, a typical high-frequency task, show our method delivers superior performance on independent and composite tasks with excellent generalization to unseen tasks. --- ### Causal Discovery Should Embrace the Wisdom of the Crowd Authors: Ryan Feng Lin, Yuantao Wei, Huiling Liao, Xiaoning Qian, Shuai Huang Published: 2026-03-03 Categories: cs.LG, cs.ET, cs.HC, stat.ME, stat.ML arXiv: https://arxiv.org/abs/2603.02678v3 This paper argues for recognizing an emerging paradigm of causal learning by wisdom of the crowd. Recent developments in government, industry, and research point to the rise of decentralized and crowd-based approaches within causal modeling, where causal knowledge distributed across many contributors can be systematically elicited and integrated with causal learning workflows. In this paradigm, causal learning becomes a distributed decision-making problem: each participant contributes partial and potentially noisy knowledge, while collective contributions help construct a global causal structure. This direction is enabled by advances in crowdsourcing platforms, expert knowledge elicitation, aggregation techniques, and large language model (LLM)-augmented information acquisition. Its promise is increasingly visible in early research and emerging real-world practices. Building on this momentum, we outline a framework for crowd-based causal learning spanning elicitation, modeling, aggregation, and optimization. We further discuss the opportunities and challenges introduced by this paradigm and call for interdisciplinary collaboration across causal learning, collective intelligence, human-AI interaction, and decision science. --- ### ITLC at SemEval-2026 Task 11: Normalization and Deterministic Parsing for Formal Reasoning in LLMs Authors: Wicaksono Leksono Muhamad, Joanito Agili Lopo, Tack Hwa Wong, Muhammad Ravi Shulthan Habibi, Samuel Cahyawijaya Published: 2026-03-03 Categories: cs.CL, cs.AI arXiv: https://arxiv.org/abs/2603.02676v2 Large language models suffer from content effects in reasoning tasks, particularly in multi-lingual contexts. We introduce a novel method that reduces these biases through explicit structural abstraction that transforms syllogisms into canonical logical representations and applies deterministic parsing to determine validity. Evaluated on the SemEval-2026 Task 11 multilingual benchmark, our approach achieves top-5 rankings across all subtasks while substantially reducing content effects and offering a competitive alternative to complex fine-tuning or activation-level interventions. --- ### From Shallow to Deep: Pinning Semantic Intent via Causal GRPO Authors: Shuyi Zhou, Zeen Song, Wenwen Qiang, Jiyan Sun, Yao Zhou, Yinlong Liu, Wei Ma Published: 2026-03-03 Categories: cs.LG arXiv: https://arxiv.org/abs/2603.02675v1 Large Language Models remain vulnerable to adversarial prefix attacks (e.g., ``Sure, here is'') despite robust standard safety. We diagnose this vulnerability as Shallow Safety Alignment, stemming from a pathology we term semantic representation decay: as the model generates compliant prefixes, its internal malicious intent signal fades. To address this, we propose Two-Stage Causal-GRPO (TSC-GRPO), a framework designed to achieve intent pinning. First, grounded in causal identifiability theory, we train a causal intent probe to disentangle invariant intent from stylistic perturbations. Second, we internalize this causal awareness into the policy via Group Relative Policy Optimization. By employing a cumulative causal penalty within ``fork-in-the-road'' training scenarios, we force the model to learn that accumulating harmful tokens monotonically decreases reward, enabling robust late-stage refusals. Experiments show that TSC-GRPO significantly outperforms baselines in defending against jailbreak attacks while preserving general utility. --- ### Exact Functional ANOVA Decomposition for Categorical Inputs Models Authors: Baptiste Ferrere, Nicolas Bousquet, Fabrice Gamboa, Jean-Michel Loubes, Joseph Muré Published: 2026-03-03 Categories: stat.ML, cs.LG arXiv: https://arxiv.org/abs/2603.02673v2 Functional ANOVA offers a principled framework for interpretability by decomposing a model's prediction into main effects and higher-order interactions. For independent features, this decomposition is well-defined, strongly linked with SHAP values, and serves as a cornerstone of additive explainability. However, the lack of an explicit closed-form expression for general dependent distributions has forced practitioners to rely on costly sampling-based approximations. We completely resolve this limitation for categorical inputs. By bridging functional analysis with the extension of discrete Fourier analysis, we derive a closed-form decomposition without any assumption. Our formulation is computationally very efficient. It seamlessly recovers the classical independent case and extends to arbitrary dependence structures, including distributions with non-rectangular support. Furthermore, leveraging the intrinsic link between SHAP and ANOVA under independence, our framework yields a natural generalization of SHAP values for the general categorical setting. --- ### SorryDB: Can AI Provers Complete Real-World Lean Theorems? Authors: Austin Letson, Leopoldo Sarra, Auguste Poiroux, Oliver Dressler, Paul Lezeau, Dhyan Aranha, Frederick Pu, Aaron Hill, Miguel Corredera Hidalgo, Julian Berman, George Tsoukalas, Lenny Taelman Published: 2026-03-03 Categories: cs.AI, cs.LG arXiv: https://arxiv.org/abs/2603.02668v2 We present SorryDB, a dynamically-updating benchmark of open Lean tasks drawn from 78 real world formalization projects on GitHub. Unlike existing static benchmarks, often composed of competition problems, hillclimbing the SorryDB benchmark will yield tools that are aligned to the community needs, more usable by mathematicians, and more capable of understanding complex dependencies. Moreover, by providing a continuously updated stream of tasks, SorryDB mitigates test-set contamination and offers a robust metric for an agent's ability to contribute to novel formal mathematics projects. We evaluate a collection of approaches, including generalist large language models, agentic approaches, and specialized symbolic provers, over a selected snapshot of 1000 tasks from SorryDB. We show that current approaches are complementary: even though an agentic approach based on Gemini Flash is the most performant, it is not strictly better than other off-the-shelf large-language models, specialized provers, or even a curated list of Lean tactics. --- ### Unifying Contrastive and Generative Objectives for Visual Understanding and Text-to-Image Generation Authors: Chao Li, Tianhong Li, Sai Vidyaranya Nuthalapati, Hong-You Chen, Satya Narayan Shukla, Jianpeng Cheng, Yonghuan Yang, Jun Xiao, Xiangjun Fan, Aashu Singh, Dina Katabi, Shlok Kumar Mishra Published: 2026-03-03 Categories: cs.CV, cs.LG arXiv: https://arxiv.org/abs/2603.02667v2 Unifying text-image contrastive learning and text-to-image (T2I) generation in a single end-to-end model is challenging because the two objectives demand opposing masking regimes: contrastive alignment needs near-complete visible tokens, while masked generative modeling needs heavy corruption. We introduce DREAM, a unified framework that resolves this conflict through Masking Warmup, a schedule that shifts the center of the masking distribution over training, so low and high masking ratios coexist at every step. This co-exposure lets a single jointly-trained encoder serve both objectives. The resulting stable optimization unlocks Semantically Aligned Decoding at inference: the text encoder, trained against visual embeddings at all masking ratios, can score partially generated images and select the best trajectory with as little as 12.5% of the image decoded, improving both FID and throughput. DREAM outperforms its single-objective baselines, CLIP and FLUID: on ImageNet linear-probing (+1.1%), 5-shot transfer (+4.1%), ADE20K segmentation (+1.9%), and NYU depth estimation (+6.25%) over CLIP, and on CC12M FID (+6.2%) over FLUID while maintaining CLIP Score. Together, these gains show that text-image contrastive and generative objectives, when properly unified, are synergistic rather than competing. --- ### Evaluating Cross-Modal Reasoning Ability and Problem Characteristics with Multimodal Item Response Theory Authors: Shunki Uebayashi, Kento Masui, Kyohei Atarashi, Han Bao, Hisashi Kashima, Naoto Inoue, Mayu Otani, Koh Takeuchi Published: 2026-03-03 Categories: cs.CL, cs.CV arXiv: https://arxiv.org/abs/2603.02663v1 Multimodal Large Language Models (MLLMs) have recently emerged as general architectures capable of reasoning over diverse modalities. Benchmarks for MLLMs should measure their ability for cross-modal integration. However, current benchmarks are filled with shortcut questions, which can be solved using only a single modality, thereby yielding unreliable rankings. For example, in vision-language cases, we can find the correct answer without either the image or the text. These low-quality questions unnecessarily increase the size and computational requirements of benchmarks. We introduce a multi-modal and multidimensional item response theory framework (M3IRT) that extends classical IRT by decomposing both model ability and item difficulty into image-only, text-only, and cross-modal components. M3IRT estimates cross-modal ability of MLLMs and each question's cross-modal difficulty, enabling compact, high-quality subsets that better reflect multimodal reasoning. Across 24 VLMs on three benchmarks, M3IRT prioritizes genuinely cross-modal questions over shortcuts and preserves ranking fidelity even when 50% of items are artificially generated low-quality questions, thereby reducing evaluation cost while improving reliability. M3IRT thus offers a practical tool for assessing cross-modal reasoning and refining multimodal benchmarks. --- ### OmniFashion: Towards Generalist Fashion Intelligence via Multi-Task Vision-Language Learning Authors: Zhengwei Yang, Andi Long, Hao Li, Zechao Hu, Kui Jiang, Zheng Wang Published: 2026-03-03 Categories: cs.CV arXiv: https://arxiv.org/abs/2603.02658v1 Fashion intelligence spans multiple tasks, i.e., retrieval, recommendation, recognition, and dialogue, yet remains hindered by fragmented supervision and incomplete fashion annotations. These limitations jointly restrict the formation of consistent visual-semantic structures, preventing recent vision-language models (VLMs) from serving as a generalist fashion brain that unifies understanding and reasoning across tasks. Therefore, we construct FashionX, a million-scale dataset that exhaustively annotates visible fashion items within an outfit and organizes attributes from global to part-level. Built upon this foundation, we propose OmniFashion, a unified vision-language framework that bridges diverse fashion tasks under a unified fashion dialogue paradigm, enabling both multi-task reasoning and interactive dialogue. Experiments on multi-subtasks and retrieval benchmarks show that OmniFashion achieves strong task-level accuracy and cross-task generalization, highlighting its offering of a scalable path toward universal, dialogue-oriented fashion intelligence. --- ### Real-Time Generation of Game Video Commentary with Multimodal LLMs: Pause-Aware Decoding Approaches Authors: Anum Afzal, Yuki Saito, Hiroya Takamura, Katsuhito Sudoh, Shinnosuke Takamichi, Graham Neubig, Florian Matthes, Tatsuya Ishigaki Published: 2026-03-03 Categories: cs.CL, cs.AI arXiv: https://arxiv.org/abs/2603.02655v1 Real-time video commentary generation provides textual descriptions of ongoing events in videos. It supports accessibility and engagement in domains such as sports, esports, and livestreaming. Commentary generation involves two essential decisions: what to say and when to say it. While recent prompting-based approaches using multimodal large language models (MLLMs) have shown strong performance in content generation, they largely ignore the timing aspect. We investigate whether in-context prompting alone can support real-time commentary generation that is both semantically relevant and well-timed. We propose two prompting-based decoding strategies: 1) a fixed-interval approach, and 2) a novel dynamic interval-based decoding approach that adjusts the next prediction timing based on the estimated duration of the previous utterance. Both methods enable pause-aware generation without any fine-tuning. Experiments on Japanese and English datasets of racing and fighting games show that the dynamic interval-based decoding can generate commentary more closely aligned with human utterance timing and content using prompting alone. We release a multilingual benchmark dataset, trained models, and implementations to support future research on real-time video commentary generation. --- ### AlphaFree: Recommendation Free from Users, IDs, and GNNs Authors: Minseo Jeon, Junwoo Jung, Daewon Gwak, Jinhong Jung Published: 2026-03-03 Categories: cs.IR, cs.AI arXiv: https://arxiv.org/abs/2603.02653v1 Can we design effective recommender systems free from users, IDs, and GNNs? Recommender systems are central to personalized content delivery across domains, with top-K item recommendation being a fundamental task to retrieve the most relevant items from historical interactions. Existing methods rely on entrenched design conventions, often adopted without reconsideration, such as storing per-user embeddings (user-dependent), initializing features from raw IDs (ID-dependent), and employing graph neural networks (GNN-dependent). These dependencies incur several limitations, including high memory costs, cold-start and over-smoothing issues, and poor generalization to unseen interactions. In this work, we propose AlphaFree, a novel recommendation method free from users, IDs, and GNNs. Our main ideas are to infer preferences on-the-fly without user embeddings (user-free), replace raw IDs with language representations (LRs) from pre-trained language models (ID-free), and capture collaborative signals through augmentation with similar items and contrastive learning, without GNNs (GNN-free). Extensive experiments on various real-world datasets show that AlphaFree consistently outperforms its competitors, achieving up to around 40% improvements over non-LR-based methods and up to 5.7% improvements over LR-based methods, while significantly reducing GPU memory usage by up to 69% under high-dimensional LRs. --- ### Improving Diffusion Planners by Self-Supervised Action Gating with Energies Authors: Yuan Lu, Dongqi Han, Yansen Wang, Dongsheng Li Published: 2026-03-03 Categories: cs.LG, cs.AI, cs.RO arXiv: https://arxiv.org/abs/2603.02650v2 Diffusion planners are a strong approach for offline reinforcement learning, but they can fail when value-guided selection favours trajectories that score well yet are locally inconsistent with the environment dynamics, resulting in brittle execution. We propose Self-supervised Action Gating with Energies (SAGE), an inference-time re-ranking method that penalises dynamically inconsistent plans using a latent consistency signal. SAGE trains a Joint-Embedding Predictive Architecture (JEPA) encoder on offline state sequences and an action-conditioned latent predictor for short horizon transitions. At test time, SAGE assigns each sampled candidate an energy given by its latent prediction error and combines this feasibility score with value estimates to select actions. SAGE can integrate into existing diffusion planning pipelines that can sample trajectories and select actions via value scoring; it requires no environment rollouts and no policy re-training. Across locomotion, navigation, and manipulation benchmarks, SAGE improves the performance and robustness of diffusion planners. --- ### HomeAdam: Adam and AdamW Algorithms Sometimes Go Home to Obtain Better Provable Generalization Authors: Feihu Huang, Guanyi Zhang, Songcan Chen Published: 2026-03-03 Categories: cs.LG, math.OC, stat.ML arXiv: https://arxiv.org/abs/2603.02649v1 Adam and AdamW are a class of default optimizers for training deep learning models in machine learning. These adaptive algorithms converge faster but generalize worse compared to SGD. In fact, their proved generalization error $O(\frac{1}{\sqrt{N}})$ also is larger than $O(\frac{1}{N})$ of SGD, where $N$ denotes training sample size. Recently, although some variants of Adam have been proposed to improve its generalization, their improved generalizations are still unexplored in theory. To fill this gap, in the paper, we restudy generalization of Adam and AdamW via algorithmic stability, and first prove that Adam and AdamW without square-root (i.e., Adam(W)-srf) have a generalization error $O(\frac{\hatρ^{-2T}}{N})$, where $T$ denotes iteration number and $\hatρ>0$ denotes the smallest element of second-order momentum plus a small positive number. To improve generalization, we propose a class of efficient clever Adam (i.e., HomeAdam(W)) algorithms via sometimes returning momentum-based SGD. Moreover, we prove that our HomeAdam(W) have a smaller generalization error $O(\frac{1}{N})$ than $O(\frac{\hatρ^{-2T}}{N})$ of Adam(W)-srf, since $\hatρ$ is generally very small. In particular, it is also smaller than the existing $O(\frac{1}{\sqrt{N}})$ of Adam(W). Meanwhile, we prove our HomeAdam(W) have a faster convergence rate of $O(\frac{1}{T^{1/4}})$ than $O(\frac{\breveρ^{-1}}{T^{1/4}})$ of the Adam(W)-srf, where $\breveρ\leq\hatρ$ also is very small. Extensive numerical experiments demonstrate efficiency of our HomeAdam(W) algorithms. --- ### SEP-YOLO: Fourier-Domain Feature Representation for Transparent Object Instance Segmentation Authors: Fengming Zhang, Tao Yan, Jianchao Huang Published: 2026-03-03 Categories: cs.CV arXiv: https://arxiv.org/abs/2603.02648v1 Transparent object instance segmentation presents significant challenges in computer vision, due to the inherent properties of transparent objects, including boundary blur, low contrast, and high dependence on background context. Existing methods often fail as they depend on strong appearance cues and clear boundaries. To address these limitations, we propose SEP-YOLO, a novel framework that integrates a dual-domain collaborative mechanism for transparent object instance segmentation. Our method incorporates a Frequency Domain Detail Enhancement Module, which separates and enhances weak highfrequency boundary components via learnable complex weights. We further design a multi-scale spatial refinement stream, which consists of a Content-Aware Alignment Neck and a Multi-scale Gated Refinement Block, to ensure precise feature alignment and boundary localization in deep semantic features. We also provide high-quality instance-level annotations for the Trans10K dataset, filling the critical data gap in transparent object instance segmentation. Extensive experiments on the Trans10K and GVD datasets show that SEP-YOLO achieves state-of-the-art (SOTA) performance. --- ### Credibility Governance: A Social Mechanism for Collective Self-Correction under Weak Truth Signals Authors: Wanying He, Yanxi Lin, Ziheng Zhou, Xue Feng, Min Peng, Qianqian Xie, Zilong Zheng, Yipeng Kang Published: 2026-03-03 Categories: cs.CY, cs.AI, cs.CL, cs.MA, cs.SI arXiv: https://arxiv.org/abs/2603.02640v1 Online platforms increasingly rely on opinion aggregation to allocate real-world attention and resources, yet common signals such as engagement votes or capital-weighted commitments are easy to amplify and often track visibility rather than reliability. This makes collective judgments brittle under weak truth signals, noisy or delayed feedback, early popularity surges, and strategic manipulation. We propose Credibility Governance (CG), a mechanism that reallocates influence by learning which agents and viewpoints consistently track evolving public evidence. CG maintains dynamic credibility scores for both agents and opinions, updates opinion influence via credibility-weighted endorsements, and updates agent credibility based on the long-run performance of the opinions they support, rewarding early and persistent alignment with emerging evidence while filtering short-lived noise. We evaluate CG in POLIS, a socio-physical simulation environment that models coupled belief dynamics and downstream feedback under uncertainty. Across settings with initial majority misalignment, observation noise and contamination, and misinformation shocks, CG outperforms vote-based, stake-weighted, and no-governance baselines, yielding faster recovery to the true state, reduced lock-in and path dependence, and improved robustness under adversarial pressure. Our implementation and experimental scripts are publicly available at https://github.com/Wanying-He/Credibility_Governance. --- ### Convex and Non-convex Federated Learning with Stale Stochastic Gradients: Diminishing Step Size is All You Need Authors: Xinran Zheng, Tara Javidi, Behrouz Touri Published: 2026-03-03 Categories: math.OC, cs.LG arXiv: https://arxiv.org/abs/2603.02639v1 We propose a general framework for distributed stochastic optimization under delayed gradient models. In this setting, $n$ local agents leverage their own data and computation to assist a central server in minimizing a global objective composed of agents' local cost functions. Each agent is allowed to transmit stochastic-potentially biased and delayed-estimates of its local gradient. While a prior work has advocated delay-adaptive step sizes for stochastic gradient descent (SGD) in the presence of delays, we demonstrate that a pre-chosen diminishing step size is sufficient and matches the performance of the adaptive scheme. Moreover, our analysis establishes that diminishing step sizes recover the optimal SGD rates for nonconvex and strongly convex objectives. --- ### The Vienna 4G/5G Drive-Test Dataset Authors: Wilfried Wiedner, Lukas Eller, Mariam Mussbah, Dominik Rössler, Valerian Maresch, Philipp Svoboda, Markus Rupp Published: 2026-03-03 Categories: eess.SP, cs.AI, cs.LG arXiv: https://arxiv.org/abs/2603.02638v2 Machine learning for mobile network analysis, planning, and optimization is often limited by the lack of large, comprehensive real-world datasets. This paper introduces the Vienna 4G/5G Drive-Test Dataset, a city-scale open dataset of georeferenced Long Term Evolution (LTE) and 5G New Radio (NR) measurements collected across Vienna, Austria. The dataset combines passive wideband scanner observations with active handset logs, providing complementary network-side and user-side views of deployed radio access networks. The measurements cover diverse urban and suburban settings and are aligned with time and location information to support consistent evaluation. For a representative subset of base stations (BSs), we provide inferred deployment descriptors, including estimated BS locations, sector azimuths, and antenna heights. The release further includes high-resolution building and terrain models, enabling geometry-conditioned learning and calibration of deterministic approaches such as ray tracing. To facilitate practical reuse, the data are organized into scanner, handset, estimated cell information, and city-model components, and the accompanying documentation describes the available fields and intended joins between them. The dataset enables reproducible benchmarking across environment-aware learning, propagation modeling, coverage analysis, and ray-tracing calibration workflows. --- ### The Vienna 4G/5G Drive-Test Dataset Authors: Wilfried Wiedner, Lukas Eller, Mariam Mussbah, Dominik Rössler, Valerian Maresch, Philipp Svoboda, Markus Rupp Published: 2026-03-03 Categories: eess.SP, cs.AI, cs.LG arXiv: https://arxiv.org/abs/2603.02638v1 Machine learning for mobile network analysis, planning, and optimization is often limited by the lack of large, comprehensive real-world datasets. This paper introduces the Vienna 4G/5G Drive-Test Dataset, a city-scale open dataset of georeferenced Long Term Evolution (LTE) and 5G New Radio (NR) measurements collected across Vienna, Austria. The dataset combines passive wideband scanner observations with active handset logs, providing complementary network-side and user-side views of deployed radio access networks. The measurements cover diverse urban and suburban settings and are aligned with time and location information to support consistent evaluation. For a representative subset of base stations (BSs), we provide inferred deployment descriptors, including estimated BS locations, sector azimuths, and antenna heights. The release further includes high-resolution building and terrain models, enabling geometry-conditioned learning and calibration of deterministic approaches such as ray tracing. To facilitate practical reuse, the data are organized into scanner, handset, estimated cell information, and city-model components, and the accompanying documentation describes the available fields and intended joins between them. The dataset enables reproducible benchmarking across environment-aware learning, propagation modeling, coverage analysis, and ray-tracing calibration workflows. --- ### StitchCUDA: An Automated Multi-Agents End-to-End GPU Programing Framework with Rubric-based Agentic Reinforcement Learning Authors: Shiyang Li, Zijian Zhang, Winson Chen, Yuebo Luo, Mingyi Hong, Caiwen Ding Published: 2026-03-03 Categories: cs.MA, cs.CL, cs.PL arXiv: https://arxiv.org/abs/2603.02637v2 Modern machine learning (ML) workloads increasingly rely on GPUs, yet achieving high end-to-end performance remains challenging due to dependencies on both GPU kernel efficiency and host-side settings. Although LLM-based methods show promise on automated GPU kernel generation, prior works mainly focus on single-kernel optimization and do not extend to end-to-end programs, hindering practical deployment. To address the challenge, in this work, we propose StitchCUDA, a multi-agent framework for end-to-end GPU program generation, with three specialized agents: a Planner to orchestrate whole system design, a Coder dedicated to implementing it step-by-step, and a Verifier for correctness check and performance profiling using Nsys/NCU. To fundamentally improve the Coder's ability in end-to-end GPU programming, StitchCUDA integrates rubric-based agentic reinforcement learning over two atomic skills, task-to-code generation and feedback-driven code optimization, with combined rubric reward and rule-based reward from real executions. Therefore, the Coder learns how to implement advanced CUDA programming techniques (e.g., custom kernel fusion, cublas epilogue), and we also effectively prevent Coder's reward hacking (e.g., just copy PyTorch code or hardcoding output) during benchmarking. Experiments on KernelBench show that StitchCUDA achieves nearly 100% success rate on end-to-end GPU programming tasks, with 1.72x better speedup over the multi-agent baseline and 2.73x than the RL model baselines. Code of the STITCHCUDA framework is avalaible at https://github.com/UMN-APEX-Lab/StitchCUDA. --- ### StitchCUDA: An Automated Multi-Agents End-to-End GPU Programing Framework with Rubric-based Agentic Reinforcement Learning Authors: Shiyang Li, Zijian Zhang, Winson Chen, Yuebo Luo, Mingyi Hong, Caiwen Ding Published: 2026-03-03 Categories: cs.MA, cs.CL, cs.PL arXiv: https://arxiv.org/abs/2603.02637v1 Modern machine learning (ML) workloads increasingly rely on GPUs, yet achieving high end-to-end performance remains challenging due to dependencies on both GPU kernel efficiency and host-side settings. Although LLM-based methods show promise on automated GPU kernel generation, prior works mainly focus on single-kernel optimization and do not extend to end-to-end programs, hindering practical deployment. To address the challenge, in this work, we propose StitchCUDA, a multi-agent framework for end-to-end GPU program generation, with three specialized agents: a Planner to orchestrate whole system design, a Coder dedicated to implementing it step-by-step, and a Verifier for correctness check and performance profiling using Nsys/NCU. To fundamentally improve the Coder's ability in end-to-end GPU programming, StitchCUDA integrates rubric-based agentic reinforcement learning over two atomic skills, task-to-code generation and feedback-driven code optimization, with combined rubric reward and rule-based reward from real executions. Therefore, the Coder learns how to implement advanced CUDA programming techniques (e.g., custom kernel fusion, cublas epilogue), and we also effectively prevent Coder's reward hacking (e.g., just copy PyTorch code or hardcoding output) during benchmarking. Experiments on KernelBench show that StitchCUDA achieves nearly 100% success rate on end-to-end GPU programming tasks, with 1.72x better speedup over the multi-agent baseline and 2.73x than the RL model baselines. --- ### SaFeR-ToolKit: Structured Reasoning via Virtual Tool Calling for Multimodal Safety Authors: Zixuan Xu, Tiancheng He, Huahui Yi, Kun Wang, Xi Chen, Gongli Xi, Qiankun Li, Kang Li, Yang Liu, Zhigang Zeng Published: 2026-03-03 Categories: cs.LG arXiv: https://arxiv.org/abs/2603.02635v1 Vision-language models remain susceptible to multimodal jailbreaks and over-refusal because safety hinges on both visual evidence and user intent, while many alignment pipelines supervise only the final response. To address this, we present SaFeR-ToolKit, which formalizes safety decision-making as a checkable protocol. Concretely, a planner specifies a persona, a Perception $\to$ Reasoning $\to$ Decision tool set, and a constrained transition graph, while a responder outputs a typed key-value tool trace before the final answer. To make the protocol reliably followed in practice, we train a single policy with a three-stage curriculum (SFT $\to$ DPO $\to$ GRPO), where GRPO directly supervises tool usage beyond answer-level feedback. Our contributions are two-fold: I. Dataset. The first tool-based safety reasoning dataset, comprising 31,654 examples (SFT 6k, DPO 18.6k, GRPO 6k) plus 1k held-out evaluation. II. Experiments. On Qwen2.5-VL, SaFeR-ToolKit significantly improves Safety/Helpfulness/Reasoning Rigor on 3B (29.39/45.04/4.98 $\to$ 84.40/71.13/78.87) and 7B (53.21/52.92/19.26 $\to$ 86.34/80.79/85.34), while preserving general capabilities (3B: 58.67 $\to$ 59.21; 7B: 66.39 $\to$ 66.81). Codes are available at https://github.com/Duebassx/SaFeR_ToolKit. --- ### Robust Heterogeneous Analog-Digital Computing for Mixture-of-Experts Models with Theoretical Generalization Guarantees Authors: Mohammed Nowaz Rabbani Chowdhury, Hsinyu Tsai, Geoffrey W. Burr, Kaoutar El Maghraoui, Liu Liu, Meng Wang Published: 2026-03-03 Categories: cs.LG, cs.AI arXiv: https://arxiv.org/abs/2603.02633v1 Sparse Mixture-of-Experts (MoE) models enable efficient scalability by activating only a small sub-set of experts per input, yet their massive parameter counts lead to substantial memory and energy inefficiency during inference. Analog in-memory computing (AIMC) offers a promising solution by eliminating frequent data movement between memory and compute units. However, mitigating hardware nonidealities of AIMC typically requires noise-aware retraining, which is infeasible for large MoE models. In this paper, we propose a retraining-free heterogeneous computation framework in which noise-sensitive experts, which are provably identifiable by their maximum neuron norm, are computed digitally while the majority of the experts are executed on AIMC hardware. We further assign densely activated modules, such as attention layers, to digital computation due to their high noise sensitivity despite comprising a small fraction of parameters. Extensive experiments on large MoE language models, including DeepSeekMoE and OLMoE, across multiple benchmark tasks validate the robustness of our approach in maintaining accuracy under analog nonidealities. --- ### Cross-Family Speculative Prefill: Training-Free Long-Context Compression with Small Draft Models Authors: Shubhangi Upasani, Ravi Shanker Raju, Bo Li, Mengmeng Ji, John Long, Chen Wu, Urmish Thakker, Guangtao Wang Published: 2026-03-03 Categories: cs.CL arXiv: https://arxiv.org/abs/2603.02631v3 Prompt length is a major bottleneck in agentic large language model (LLM) workloads, where repeated inference steps and multi-call loops incur substantial prefill cost. Recent work on speculative prefill demonstrates that attention-based token importance estimation can enable training-free prompt compression, but this assumes the existence of a draft model that shares the same tokenizer as the target model. In practice, however, agentic pipelines frequently employ models without any smaller in-family draft model. In this work, we study cross-family speculative prefill, where a lightweight draft model from one model family is used to perform prompt compression for a target model from a different family. Using the same speculative prefill mechanism as prior work, we evaluate a range of cross-family draft-target combinations, including Qwen, LLaMA, and DeepSeek models. Across a broad diversity of tasks, we find that attention-based token importance estimation transfers reliably across different model families despite differences in model architectures and tokenizers between draft and target models. Cross-model prompt compression largely retains 90~100% of full-prompt baseline performance and, in some cases, slightly improves accuracy due to denoising effects, while delivering substantial reductions in time to first token (TTFT). These results suggest that speculative prefill depends mainly on task priors and semantic structure, thus serving as a generalizable prompt compression primitive. We discuss the implications of our findings for agentic systems, where repeated long-context inference and heterogeneous model stacks make cross-model prompt compression both necessary and practical. --- ### MASPOB: Bandit-Based Prompt Optimization for Multi-Agent Systems with Graph Neural Networks Authors: Zhi Hong, Qian Zhang, Jiahang Sun, Zhiwei Shang, Mingze Kong, Xiangyi Wang, Yao Shu, Zhongxiang Dai Published: 2026-03-03 Categories: cs.LG, cs.AI arXiv: https://arxiv.org/abs/2603.02630v2 Large Language Models (LLMs) have achieved great success in many real-world applications, especially the one serving as the cognitive backbone of Multi-Agent Systems (MAS) to orchestrate complex workflows in practice. Since many deployment scenarios preclude MAS workflow modifications and its performance is highly sensitive to the input prompts, prompt optimization emerges as a more natural approach to improve its performance. However, real-world prompt optimization for MAS is impeded by three key challenges: (1) the need of sample efficiency due to prohibitive evaluation costs, (2) topology-induced coupling among prompts, and (3) the combinatorial explosion of the search space. To address these challenges, we introduce MASPOB (Multi-Agent System Prompt Optimization via Bandits), a novel sample-efficient framework based on bandits. By leveraging Upper Confidence Bound (UCB) to quantify uncertainty, the bandit framework balances exploration and exploitation, maximizing gains within a strictly limited budget. To handle topology-induced coupling, MASPOB integrates Graph Neural Networks (GNNs) to capture structural priors, learning topology-aware representations of prompt semantics. Furthermore, it employs coordinate ascent to decompose the optimization into univariate sub-problems, reducing search complexity from exponential to linear. Extensive experiments across diverse benchmarks demonstrate that MASPOB achieves state-of-the-art performance, consistently outperforming existing baselines. --- ### Towards an Incremental Unified Multimodal Anomaly Detection: Augmenting Multimodal Denoising From an Information Bottleneck Perspective Authors: Kaifang Long, Lianbo Ma, Jiaqi Liu, Liming Liu, Guoyang Xie Published: 2026-03-03 Categories: cs.CV arXiv: https://arxiv.org/abs/2603.02629v1 The quest for incremental unified multimodal anomaly detection seeks to empower a single model with the ability to systematically detect anomalies across all categories and support incremental learning to accommodate emerging objects/categories. Central to this pursuit is resolving the catastrophic forgetting dilemma, which involves acquiring new knowledge while preserving prior learned knowledge. Despite some efforts to address this dilemma, a key oversight persists: ignoring the potential impact of spurious and redundant features on catastrophic forgetting. In this paper, we delve into the negative effect of spurious and redundant features on this dilemma in incremental unified frameworks, and reveal that under similar conditions, the multimodal framework developed by naive aggregation of unimodal architectures is more prone to forgetting. To address this issue, we introduce a novel denoising framework called IB-IUMAD, which exploits the complementary benefits of the Mamba decoder and information bottleneck fusion module: the former dedicated to disentangle inter-object feature coupling, preventing spurious feature interference between objects; the latter serves to filter out redundant features from the fused features, thus explicitly preserving discriminative information. A series of theoretical analyses and experiments on MVTec 3D-AD and Eyecandies datasets demonstrates the effectiveness and competitive performance of IB-IUMAD. --- ### Post Hoc Extraction of Pareto Fronts for Continuous Control Authors: Raghav Thakar, Gaurav Dixit, Kagan Tumer Published: 2026-03-03 Categories: cs.LG arXiv: https://arxiv.org/abs/2603.02628v1 Agents in the real world must often balance multiple objectives, such as speed, stability, and energy efficiency in continuous control. To account for changing conditions and preferences, an agent must ideally learn a Pareto frontier of policies representing multiple optimal trade-offs. Recent advances in multi-policy multi-objective reinforcement learning (MORL) enable learning a Pareto front directly, but require full multi-objective consideration from the start of training. In practice, multi-objective preferences often arise after a policy has already been trained on a single specialised objective. Existing MORL methods cannot leverage these pre-trained `specialists' to learn Pareto fronts and avoid incurring the sample costs of retraining. We introduce Mixed Advantage Pareto Extraction (MAPEX), an offline MORL method that constructs a frontier of policies by reusing pre-trained specialist policies, critics, and replay buffers. MAPEX combines evaluations from specialist critics into a mixed advantage signal, and weights a behaviour cloning loss with it to train new policies that balance multiple objectives. MAPEX's post hoc Pareto front extraction preserves the simplicity of single-objective off-policy RL, and avoids retrofitting these algorithms into complex MORL frameworks. We formally describe the MAPEX procedure and evaluate MAPEX on five multi-objective MuJoCo environments. Given the same starting policies, MAPEX produces comparable fronts at $0.001\%$ the sample cost of established baselines. --- ### See and Remember: A Multimodal Agent for Web Traversal Authors: Xinjun Wang, Shengyao Wang, Aimin Zhou, Hao Hao Published: 2026-03-03 Categories: cs.AI arXiv: https://arxiv.org/abs/2603.02626v1 Autonomous web navigation requires agents to perceive complex visual environments and maintain long-term context, yet current Large Language Model (LLM) based agents often struggle with spatial disorientation and navigation loops. In this paper, we propose generally applicable V-GEMS(Visual Grounding and Explicit Memory System), a robust multimodal agent architecture designed for precise and resilient web traversal. Our agent integrates visual grounding to resolve ambiguous interactive elements and introduces an explicit memory stack with state tracking. This dual mechanism allows the agent to maintain a structured map of its traversal path, enabling valid backtracking and preventing cyclical failures in deep navigation tasks. We also introduce an updatable dynamic benchmark to rigorously evaluate adaptability. Experiments show V-GEMS significantly dominates the WebWalker baseline, achieving a substantial 28.7% performance gain. Code is available at https://github.com/Vaultttttttttttt/V-GEMS. --- ### Uni-Skill: Building Self-Evolving Skill Repository for Generalizable Robotic Manipulation Authors: Senwei Xie, Yuntian Zhang, Ruiping Wang, Xilin Chen Published: 2026-03-03 Categories: cs.RO, cs.LG arXiv: https://arxiv.org/abs/2603.02623v1 While skill-centric approaches leverage foundation models to enhance generalization in compositional tasks, they often rely on fixed skill libraries, limiting adaptability to new tasks without manual intervention. To address this, we propose Uni-Skill, a Unified Skill-centric framework that supports skill-aware planning and facilitates automatic skill evolution. Unlike prior methods that restrict planning to predefined skills, Uni-Skill requests for new skill implementations when existing ones are insufficient, ensuring adaptable planning with self-augmented skill library. To support automatic implementation of diverse skills requested by the planning module, we construct SkillFolder, a VerbNet-inspired repository derived from large-scale unstructured robotic videos. SkillFolder introduces a hierarchical skill taxonomy that captures diverse skill descriptions at multiple levels of abstraction. By populating this taxonomy with large-scale, automatically annotated demonstrations, Uni-Skill shifts the paradigm of skill acquisition from inefficient manual annotation to efficient offline structural retrieval. Retrieved examples provide semantic supervision over behavior patterns and fine-grained references for spatial trajectories, enabling few-shot skill inference without deployment-time demonstrations. Comprehensive experiments in both simulation and real-world settings verify the state-of-the-art performance of Uni-Skill over existing VLM-based skill-centric approaches, highlighting its advanced reasoning capabilities and strong zero-shot generalization across a wide range of novel tasks. --- ### Implicit Bias in Deep Linear Discriminant Analysis Authors: Jiawen Li Published: 2026-03-03 Categories: cs.LG, stat.ML arXiv: https://arxiv.org/abs/2603.02622v2 While the Implicit Bias(or Implicit Regularization) of standard loss functions has been studied, the optimization geometry induced by discriminative metric-learning objectives remains largely unexplored.To the best of our knowledge, this paper presents an initial theoretical analysis of the implicit regularization induced by the Deep LDA,a scale invariant objective designed to minimize intraclass variance and maximize interclass distance. By analyzing the gradient flow of the loss on a L-layer diagonal linear network, we prove that under balanced initialization, the network architecture transforms standard additive gradient updates into multiplicative weight updates, which demonstrates an automatic conservation of the (2/L) quasi-norm. --- ### Same Error, Different Function: The Optimizer as an Implicit Prior in Financial Time Series Authors: Federico Vittorio Cortesi, Giuseppe Iannone, Giulia Crippa, Tomaso Poggio, Pierfrancesco Beneventano Published: 2026-03-03 Categories: cs.LG, q-fin.CP arXiv: https://arxiv.org/abs/2603.02620v1 Neural networks applied to financial time series operate in a regime of underspecification, where model predictors achieve indistinguishable out-of-sample error. Using large-scale volatility forecasting for S$\&$P 500 stocks, we show that different model-training-pipeline pairs with identical test loss learn qualitatively different functions. Across architectures, predictive accuracy remains unchanged, yet optimizer choice reshapes non-linear response profiles and temporal dependence differently. These divergences have material consequences for decisions: volatility-ranked portfolios trace a near-vertical Sharpe-turnover frontier, with nearly $3\times$ turnover dispersion at comparable Sharpe ratios. We conclude that in underspecified settings, optimization acts as a consequential source of inductive bias, thus model evaluation should extend beyond scalar loss to encompass functional and decision-level implications. --- ### Direct Reward Fine-Tuning on Poses for Single Image to 3D Human in the Wild Authors: Seunguk Do, Minwoo Huh, Joonghyuk Shin, Jaesik Park Published: 2026-03-03 Categories: cs.CV arXiv: https://arxiv.org/abs/2603.02619v1 Single-view 3D human reconstruction has achieved remarkable progress through the adoption of multi-view diffusion models, yet the recovered 3D humans often exhibit unnatural poses. This phenomenon becomes pronounced when reconstructing 3D humans with dynamic or challenging poses, which we attribute to the limited scale of available 3D human datasets with diverse poses. To address this limitation, we introduce DrPose, Direct Reward fine-tuning algorithm on Poses, which enables post-training of a multi-view diffusion model on diverse poses without requiring expensive 3D human assets. DrPose trains a model using only human poses paired with single-view images, employing a direct reward fine-tuning to maximize PoseScore, which is our proposed differentiable reward that quantifies consistency between a generated multi-view latent image and a ground-truth human pose. This optimization is conducted on DrPose15K, a novel dataset that was constructed from an existing human motion dataset and a pose-conditioned video generative model. Constructed from abundant human pose sequence data, DrPose15K exhibits a broader pose distribution compared to existing 3D human datasets. We validate our approach through evaluation on conventional benchmark datasets, in-the-wild images, and a newly constructed benchmark, with a particular focus on assessing performance on challenging human poses. Our results demonstrate consistent qualitative and quantitative improvements across all benchmarks. Project page: https://seunguk-do.github.io/drpose. --- ### Mind the Way You Select Negative Texts: Pursuing the Distance Consistency in OOD Detection with VLMs Authors: Zhikang Xu, Qianqian Xu, Zitai Wang, Cong Hua, Sicong Li, Zhiyong Yang, Qingming Huang Published: 2026-03-03 Categories: cs.CV arXiv: https://arxiv.org/abs/2603.02618v3 Out-of-distribution (OOD) detection seeks to identify samples from unknown classes, a critical capability for deploying machine learning models in open-world scenarios. Recent research has demonstrated that Vision-Language Models (VLMs) can effectively leverage their multi-modal representations for OOD detection. However, current methods often incorporate intra-modal distance during OOD detection, such as comparing negative texts with ID labels or comparing test images with image proxies. This design paradigm creates an inherent inconsistency against the inter-modal distance that CLIP-like VLMs are optimized for, potentially leading to suboptimal performance. To address this limitation, we propose InterNeg, a simple yet effective framework that systematically utilizes consistent inter-modal distance enhancement from textual and visual perspectives. From the textual perspective, we devise an inter-modal criterion for selecting negative texts. From the visual perspective, we dynamically identify high-confidence OOD images and invert them into the textual space, generating extra negative text embeddings guided by inter-modal distance. Extensive experiments across multiple benchmarks demonstrate the superiority of our approach. Notably, our InterNeg achieves state-of-the-art performance compared to existing works, with a 3.47% reduction in FPR95 on the large-scale ImageNet benchmark and a 5.50% improvement in AUROC on the challenging Near-OOD benchmark. --- ### Detecting Structural Heart Disease from Electrocardiograms via a Generalized Additive Model of Interpretable Foundation-Model Predictors Authors: Ya Zhou, Zhaohong Sun, Tianxiang Hao, Xiangjie Li Published: 2026-03-03 Categories: stat.AP, cs.AI, cs.LG, stat.ME, stat.ML arXiv: https://arxiv.org/abs/2603.02616v1 Structural heart disease (SHD) is a prevalent condition with many undiagnosed cases, and early detection is often limited by the high cost and accessibility constraints of echocardiography (ECHO). Recent studies show that artificial intelligence (AI)-based analysis of electrocardiograms (ECGs) can detect SHD, offering a scalable alternative. However, existing methods are fully black-box models, limiting interpretability and clinical adoption. To address these challenges, we propose an interpretable and effective framework that integrates clinically meaningful ECG foundation-model predictors within a generalized additive model, enabling transparent risk attribution while maintaining strong predictive performance. Using the EchoNext benchmark of over 80,000 ECG-ECHO pairs, the method demonstrates relative improvements of +0.98% in AUROC, +1.01% in AUPRC, and +1.41% in F1 score over the latest state-of-the-art deep-learning baseline, while achieving slightly better performance even with only 30% of the training data. Subgroup analyses confirm robust performance across heterogeneous populations, and the estimated entry-wise functions provide interpretable insights into the relationships between risks of traditional ECG diagnoses and SHD. This work illustrates a complementary paradigm between classical statistical modeling and modern AI, offering a pathway to interpretable, high-performing, and clinically actionable ECG-based SHD screening. --- ### Think, But Don't Overthink: Reproducing Recursive Language Models Authors: Daren Wang Published: 2026-03-03 Categories: cs.CL arXiv: https://arxiv.org/abs/2603.02615v1 This project reproduces and extends the recently proposed ``Recursive Language Models'' (RLMs) framework by Zhang et al. (2026). This framework enables Large Language Models (LLMs) to process near-infinite contexts by offloading the prompt into an external REPL environment. While the original paper relies on a default recursion depth of 1 and suggests deeper recursion as a future direction, this study specifically investigates the impact of scaling the recursion depth. Using state-of-the-art open-source agentic models (DeepSeek v3.2 and Kimi K2), I evaluated pure LLM, RLM (depth=1), and RLM (depth=2) on the S-NIAH and OOLONG benchmarks. The findings reveal a compelling phenomenon: Deeper recursion causes models to ``overthink''. While depth-1 RLMs effectively boost accuracy on complex reasoning tasks, applying deeper recursion (depth=2) or using RLMs on simple retrieval tasks paradoxically degrades performance and exponentially inflates execution time (e.g., from 3.6s to 344.5s) and token costs. Code and data are available at: https://github.com/drbillwang/rlm-reproduction --- ### Real-Time Generative Policy via Langevin-Guided Flow Matching for Autonomous Driving Authors: Tianze Zhu, Yinuo Wang, Wenjun Zou, Tianyi Zhang, Likun Wang, Letian Tao, Feihong Zhang, Yao Lyu, Shengbo Eben Li Published: 2026-03-03 Categories: cs.LG, cs.RO arXiv: https://arxiv.org/abs/2603.02613v1 Reinforcement learning (RL) is a fundamental methodology in autonomous driving systems, where generative policies exhibit considerable potential by leveraging their ability to model complex distributions to enhance exploration. However, their inherent high inference latency severely impedes their deployment in real-time decision-making and control. To address this issue, we propose diffusion actor-critic with entropy regulator via flow matching (DACER-F) by introducing flow matching into online RL, enabling the generation of competitive actions in a single inference step. By leveraging Langevin dynamics and gradients of the Q-function, DACER-F dynamically optimizes actions from experience replay toward a target distribution that balances high Q-value information with exploratory behavior. The flow policy is then trained to efficiently learn a mapping from a simple prior distribution to this dynamic target. In complex multi-lane and intersection simulations, DACER-F outperforms baselines diffusion actor-critic with entropy regulator (DACER) and distributional soft actor-critic (DSAC), while maintaining an ultra-low inference latency. DACER-F further demonstrates its scalability on standard RL benchmark DeepMind Control Suite (DMC), achieving a score of 775.8 in the humanoid-stand task and surpassing prior methods. Collectively, these results establish DACER-F as a high-performance and computationally efficient RL algorithm. --- ### VLMFusionOcc3D: VLM Assisted Multi-Modal 3D Semantic Occupancy Prediction Authors: A. Enes Doruk, Hasan F. Ates Published: 2026-03-03 Categories: cs.CV, cs.RO arXiv: https://arxiv.org/abs/2603.02609v1 This paper introduces VLMFusionOcc3D, a robust multimodal framework for dense 3D semantic occupancy prediction in autonomous driving. Current voxel-based occupancy models often struggle with semantic ambiguity in sparse geometric grids and performance degradation under adverse weather conditions. To address these challenges, we leverage the rich linguistic priors of Vision-Language Models (VLMs) to anchor ambiguous voxel features to stable semantic concepts. Our framework initiates with a dual-branch feature extraction pipeline that projects multi-view images and LiDAR point clouds into a unified voxel space. We propose Instance-driven VLM Attention (InstVLM), which utilizes gated cross-attention and LoRA-adapted CLIP embeddings to inject high-level semantic and geographic priors directly into the 3D voxels. Furthermore, we introduce Weather-Aware Adaptive Fusion (WeathFusion), a dynamic gating mechanism that utilizes vehicle metadata and weather-conditioned prompts to re-weight sensor contributions based on real-time environmental reliability. To ensure structural consistency, a Depth-Aware Geometric Alignment (DAGA) loss is employed to align dense camera-derived geometry with sparse, spatially accurate LiDAR returns. Extensive experiments on the nuScenes and SemanticKITTI datasets demonstrate that our plug-and-play modules consistently enhance the performance of state-of-the-art voxel-based baselines. Notably, our approach achieves significant improvements in challenging weather scenarios, offering a scalable and robust solution for complex urban navigation. --- ### Combinatorial Sparse PCA Beyond the Spiked Identity Model Authors: Syamantak Kumar, Purnamrita Sarkar, Kevin Tian, Peiyuan Zhang Published: 2026-03-03 Categories: stat.ML, cs.DS, cs.LG, math.OC arXiv: https://arxiv.org/abs/2603.02607v2 Sparse PCA is one of the most well-studied problems in high-dimensional statistics. In this problem, we are given samples from a distribution with covariance $Σ$, whose top eigenvector $v \in R^d$ is $s$-sparse. Existing sparse PCA algorithms can be broadly categorized into (1) combinatorial algorithms (e.g., diagonal or elementwise covariance thresholding) and (2) SDP-based algorithms. While combinatorial algorithms are much simpler, they are typically only analyzed under the spiked identity model (where $Σ= I_d + γvv^\top$ for some $γ> 0$), whereas SDP-based algorithms require no additional assumptions on $Σ$. We demonstrate explicit counterexample covariances $Σ$ against the success of standard combinatorial algorithms for sparse PCA, when moving beyond the spiked identity model. In light of this discrepancy, we give the first combinatorial method for sparse PCA that provably succeeds for general $Σ$ using $s^2 \cdot \mathrm{polylog}(d)$ samples and $d^2 \cdot \mathrm{poly}(s, \log(d))$ time, by providing a global convergence guarantee on a variant of the truncated power method of Yuan and Zhang (2013). We provide a natural generalization of our method to recovering a vector in a sparse leading eigenspace. Finally, we evaluate our method on synthetic and real-world sparse PCA datasets. --- ### Heterogeneous Agent Collaborative Reinforcement Learning Authors: Zhixia Zhang, Zixuan Huang, Gongxun Li, Huaiyang Wang, Chengyi Yuan, Xin Xia, Deqing Wang, Fuzhen Zhuang, Shuai Ma, Ning Ding, Yaodong Yang, Jianxin Li et al. Published: 2026-03-03 Categories: cs.LG arXiv: https://arxiv.org/abs/2603.02604v2 We introduce Heterogeneous Agent Collaborative Reinforcement Learning (HACRL), a new Reinforcement Learning from Verifiable Reward (RLVR) problem that addresses the inefficiencies of isolated multi-agent on-policy optimization. HACRL enables collaborative optimization with independent execution: heterogeneous agents share verified rollouts during training to mutually improve, while operating independently at inference time. Unlike LLM-based multi-agent reinforcement learning (MARL), HACRL does not require coordinated deployment, and unlike on-/off-policy distillation, it enables bidirectional mutual learning among heterogeneous agents rather than one-directional homogeneous teacher-to-student transfer. Building on this problem, we propose HACPO, a collaborative RL algorithm that enables principled rollout sharing to maximize sample utilization and cross-agent knowledge transfer. To mitigate capability discrepancies and policy distribution shifts, HACPO introduces four tailored mechanisms with theoretical guarantees on unbiased advantage estimation. Extensive experiments across diverse heterogeneous model combinations and reasoning benchmarks show that HACPO consistently improves all participating agents, outperforming GSPO with double rollouts by an average of 3.6% while using only half the rollout cost. --- ### AgentAssay: Token-Efficient Regression Testing for Non-Deterministic AI Agent Workflows Authors: Varun Pratap Bhardwaj Published: 2026-03-03 Categories: cs.AI, cs.SE arXiv: https://arxiv.org/abs/2603.02601v1 Autonomous AI agents are deployed at unprecedented scale, yet no principled methodology exists for verifying that an agent has not regressed after changes to its prompts, tools, models, or orchestration logic. We present AgentAssay, the first token-efficient framework for regression testing non-deterministic AI agent workflows, achieving 78-100% cost reduction while maintaining rigorous statistical guarantees. Our contributions include: (1) stochastic three-valued verdicts (PASS/FAIL/INCONCLUSIVE) grounded in hypothesis testing; (2) five-dimensional agent coverage metrics; (3) agent-specific mutation testing operators; (4) metamorphic relations for agent workflows; (5) CI/CD deployment gates as statistical decision procedures; (6) behavioral fingerprinting that maps execution traces to compact vectors, enabling multivariate regression detection; (7) adaptive budget optimization calibrating trial counts to behavioral variance; and (8) trace-first offline analysis enabling zero-cost testing on production traces. Experiments across 5 models (GPT-5.2, Claude Sonnet 4.6, Mistral-Large-3, Llama-4-Maverick, Phi-4), 3 scenarios, and 7,605 trials demonstrate that behavioral fingerprinting achieves 86% detection power where binary testing has 0%, SPRT reduces trials by 78%, and the full pipeline achieves 100% cost savings through trace-first analysis. Implementation: 20,000+ lines of Python, 751 tests, 10 framework adapters. --- ### SUN: Shared Use of Next-token Prediction for Efficient Multi-LLM Disaggregated Serving Authors: Sunghyeon Woo, Ahreum Seo, Jaegwang Lee, Jaeeun Kil, Hanbae Seo, Joonghoon Kim, Baeseong Park, Se Jung Kwon, Dongsoo Lee Published: 2026-03-03 Categories: cs.AI, cs.LG arXiv: https://arxiv.org/abs/2603.02599v1 In multi-model LLM serving, decode execution remains inefficient due to model-specific resource partitioning: since cross-model batching is not possible, memory-bound decoding often suffers from severe GPU underutilization, especially under skewed workloads. We propose Shared Use of Next-token Prediction (SUN), the first approach that enables cross-model sharing of decode execution in disaggregated multi-LLM serving. SUN decomposes a decoder-only Transformer into a prefill module and a decode module, and fine-tunes only the task-specific prefill module, enabling a frozen decode module to be shared across models. This design enables a model-agnostic decode routing policy that balances decode requests across shared workers to maximize utilization. Across diverse tasks and model families, SUN achieves accuracy comparable to full fine-tuning while maintaining system throughput with fewer decode workers. In particular, SUN improves throughput per GPU by up to 2.0x over conventional disaggregation while keeping time-per-output-token (TPOT) within 5%. SUN inherently enables and facilitates low-bit decoding; with Quantized SUN (QSUN), it achieves a 45% speedup with comparable accuracy to SUN while preserving the benefits of shared decoding. --- ### Synthetic-Child: An AIGC-Based Synthetic Data Pipeline for Privacy-Preserving Child Posture Estimation Authors: Taowen Zeng Published: 2026-03-03 Categories: cs.CV arXiv: https://arxiv.org/abs/2603.02598v1 Accurate child posture estimation is critical for AI-powered study companion devices, yet collecting large-scale annotated datasets of children is both expensive and ethically prohibitive due to privacy concerns. We present Synthetic-Child, an AIGC-based synthetic data pipeline that produces photorealistic child posture training images with ground-truth-projected keypoint annotations, requiring zero real child photographs. The pipeline comprises four stages: (1) a programmable 3D child body model (SMPL-X) in Blender generates diverse desk-study poses with IK-constrained anatomical plausibility and automatic COCO-format ground-truth export; (2) a custom PoseInjectorNode feeds 3D-derived skeletons into a dual ControlNet (pose + depth) conditioned on FLUX-1 Dev, synthesizing 12,000 photorealistic images across 10 posture categories with low annotation drift; (3) ViTPose-based confidence filtering and targeted augmentation remove generation failures and improve robustness; (4) RTMPose-M (13.6M params) is fine-tuned on the synthetic data and paired with geometric feature engineering and a lightweight MLP for posture classification, then quantized to INT8 for real-time edge deployment. On a real-child test set (n~300), the FP16 model achieves 71.2 AP -- a +12.5 AP improvement over the COCO-pretrained adult-data baseline at identical model capacity. After INT8 quantization the model retains 70.4 AP while running at 22 FPS on a 0.8-TOPS Rockchip RK3568 NPU. In a single-subject controlled comparison with a commercial posture corrector, our system achieves substantially higher recognition rates across most tested categories and responds ~1.8x faster on average. These results demonstrate that carefully designed AIGC pipelines can substantially reduce dependence on real child imagery while achieving deployment-ready accuracy, with potential applications to other privacy-sensitive domains. --- ### GPUTOK: GPU Accelerated Byte Level BPE Tokenization Authors: Venu Gopal Kadamba, Kanishkha Jaisankar Published: 2026-03-03 Categories: cs.CL, cs.AI, cs.DC, cs.LG arXiv: https://arxiv.org/abs/2603.02597v1 As large language models move toward million-token context windows, CPU tokenizers become a major slowdown because they process text one step at a time while powerful GPUs sit unused. We built a GPU-based byte-level BPE tokenizer that follows GPT-2's merge rules. It includes a basic BlockBPE-style kernel and a faster, optimized version that uses cuCollections static map, CUB reductions, and a pybind11 interface for Python. On WikiText103 sequences up to 131k tokens, the optimized GPU tokenizer produces the same tokens as a CPU version and, for the longest inputs, is about 1.7x faster than tiktoken and about 7.6x faster than the HuggingFace GPT-2 tokenizer. Nsight profiling shows that 70-80% of CUDA API time goes to memory allocation, so adding memory pooling should give the biggest speed boost next. Tests on generation tasks using WikiText103 prompts show that our GPU tokenizer's outputs stay within about one percentage point of tiktoken and HuggingFace GPT-2 on similarity and overlap metrics, meaning it keeps output quality while making long-context inference more practical. --- ### Low-Degree Method Fails to Predict Robust Subspace Recovery Authors: He Jia, Aravindan Vijayaraghavan Published: 2026-03-03 Categories: stat.ML, cs.CC, cs.DS, cs.LG arXiv: https://arxiv.org/abs/2603.02594v1 The low-degree polynomial framework has been highly successful in predicting computational versus statistical gaps for high-dimensional problems in average-case analysis and machine learning. This success has led to the low-degree conjecture, which posits that this method captures the power and limitations of efficient algorithms for a wide class of high-dimensional statistical problems. We identify a natural and basic hypothesis testing problem in $\mathbb{R}^n$ which is polynomial time solvable, but for which the low-degree polynomial method fails to predict its computational tractability even up to degree $k=n^{Ω(1)}$. Moreover, the low-degree moments match exactly up to degree $k=O(\sqrt{\log n/\log\log n})$. Our problem is a special case of the well-studied robust subspace recovery problem. The lower bounds suggest that there is no polynomial time algorithm for this problem. In contrast, we give a simple and robust polynomial time algorithm that solves the problem (and noisy variants of it), leveraging anti-concentration properties of the distribution. Our results suggest that the low-degree method and low-degree moments fail to capture algorithms based on anti-concentration, challenging their universality as a predictor of computational barriers. --- ### Maximizing Generalization: The Effect of Different Augmentation Techniques on Lightweight Vision Transformer for Bengali Character Classification Authors: Rafi Hassan Chowdhury, Naimul Haque, Kaniz Fatiha Published: 2026-03-03 Categories: cs.CV arXiv: https://arxiv.org/abs/2603.02591v1 Deep learning models have proven to be highly effective in computer vision, with deep convolutional neural networks achieving impressive results across various computer vision tasks. However, these models rely heavily on large datasets to avoid overfitting. When a model learns features with either low or high variance, it can lead to underfitting or overfitting on the training data. Unfortunately, large-scale datasets may not be available in many domains, particularly for resource-limited languages such as Bengali. In this experiment, a series of tests were conducted in the field of image data augmentation as an approach to addressing the limited data problem for Bengali handwritten characters. The study also provides an in-depth analysis of the performance of different augmentation techniques. Data augmentation refers to a set of techniques applied to data to increase its size and diversity, making it more suitable for training deep learning models. The image augmentation techniques evaluated in this study include CLAHE, Random Rotation, Random Affine, Color Jitter, and their combinations. The study further explores the use of augmentation methods with a lightweight model such as EfficientViT. Among the different augmentation strategies, the combination of Random Affine and Color Jitter produced the best accuracy on the Ekush (1) and AIBangla (2) datasets, achieving accuracies of 97.48% and 97.57%, respectively. This combination outperformed all other individual and combined augmentation techniques. Overall, this analysis presents a thorough examination of the impact of image data augmentation in resource-scarce languages, particularly in the context of Bengali handwritten character recognition using lightweight models. --- ### ExpGuard: LLM Content Moderation in Specialized Domains Authors: Minseok Choi, Dongjin Kim, Seungbin Yang, Subin Kim, Youngjun Kwak, Juyoung Oh, Jaegul Choo, Jungmin Son Published: 2026-03-03 Categories: cs.CL arXiv: https://arxiv.org/abs/2603.02588v1 With the growing deployment of large language models (LLMs) in real-world applications, establishing robust safety guardrails to moderate their inputs and outputs has become essential to ensure adherence to safety policies. Current guardrail models predominantly address general human-LLM interactions, rendering LLMs vulnerable to harmful and adversarial content within domain-specific contexts, particularly those rich in technical jargon and specialized concepts. To address this limitation, we introduce ExpGuard, a robust and specialized guardrail model designed to protect against harmful prompts and responses across financial, medical, and legal domains. In addition, we present ExpGuardMix, a meticulously curated dataset comprising 58,928 labeled prompts paired with corresponding refusal and compliant responses, from these specific sectors. This dataset is divided into two subsets: ExpGuardTrain, for model training, and ExpGuardTest, a high-quality test set annotated by domain experts to evaluate model robustness against technical and domain-specific content. Comprehensive evaluations conducted on ExpGuardTest and eight established public benchmarks reveal that ExpGuard delivers competitive performance across the board while demonstrating exceptional resilience to domain-specific adversarial attacks, surpassing state-of-the-art models such as WildGuard by up to 8.9% in prompt classification and 15.3% in response classification. To encourage further research and development, we open-source our code, data, and model, enabling adaptation to additional domains and supporting the creation of increasingly robust guardrail models. --- ### LiveAgentBench: Comprehensive Benchmarking of Agentic Systems Across 104 Real-World Challenges Authors: Hao Li, Huan Wang, Jinjie Gu, Wenjie Wang, Chenyi Zhuang, Sikang Bian Published: 2026-03-03 Categories: cs.AI arXiv: https://arxiv.org/abs/2603.02586v1 As large language models grow more capable, general AI agents have become increasingly prevalent in practical applications. However, existing benchmarks face significant limitations, failing to represent real-world user tasks accurately. To address this gap, we present LiveAgentBench, a comprehensive benchmark with 104 scenarios that reflect real user requirements. It is constructed from publicly sourced questions on social media and real-world products. Central to our approach is the Social Perception-Driven Data Generation (SPDG) method, a novel process we developed to ensure each question's real-world relevance, task complexity, and result verifiability. We evaluate various models, frameworks, and commercial products using LiveAgentBench, revealing their practical performance and identifying areas for improvement. This release includes 374 tasks, with 125 for validation and 249 for testing. The SPDG process enables continuous updates with fresh queries from real-world interactions. --- ### Neural Electromagnetic Fields for High-Resolution Material Parameter Reconstruction Authors: Zhe Chen, Peilin Zheng, Wenshuo Chen, Xiucheng Wang, Yutao Yue, Nan Cheng Published: 2026-03-03 Categories: cs.CV, eess.SP arXiv: https://arxiv.org/abs/2603.02582v1 Creating functional Digital Twins, simulatable 3D replicas of the real world, is a central challenge in computer vision. Current methods like NeRF produce visually rich but functionally incomplete twins. The key barrier is the lack of underlying material properties (e.g., permittivity, conductivity). Acquiring this information for every point in a scene via non-contact, non-invasive sensing is a primary goal, but it demands solving a notoriously ill-posed physical inversion problem. Standard remote signals, like images and radio frequencies (RF), deeply entangle the unknown geometry, ambient field, and target materials. We introduce NEMF, a novel framework for dense, non-invasive physical inversion designed to build functional digital twins. Our key insight is a systematic disentanglement strategy. NEMF leverages high-fidelity geometry from images as a powerful anchor, which first enables the resolution of the ambient field. By constraining both geometry and field using only non-invasive data, the original ill-posed problem transforms into a well-posed, physics-supervised learning task. This transformation unlocks our core inversion module: a decoder. Guided by ambient RF signals and a differentiable layer incorporating physical reflection models, it learns to explicitly output a continuous, spatially-varying field of the scene's underlying material parameters. We validate our framework on high-fidelity synthetic datasets. Experiments show our non-invasive inversion reconstructs these material maps with high accuracy, and the resulting functional twin enables high-fidelity physical simulation. This advance moves beyond passive visual replicas, enabling the creation of truly functional and simulatable models of the physical world. --- ### ATD: Improved Transformer with Adaptive Token Dictionary for Image Restoration Authors: Leheng Zhang, Wei Long, Yawei Li, Xingyu Zhou, Xiaorui Zhao, Shuhang Gu Published: 2026-03-03 Categories: cs.CV arXiv: https://arxiv.org/abs/2603.02581v1 Recently, Transformers have gained significant popularity in image restoration tasks such as image super-resolution and denoising, owing to their superior performance. However, balancing performance and computational burden remains a long-standing problem for transformer-based architectures. Due to the quadratic complexity of self-attention, existing methods often restrict attention to local windows, resulting in limited receptive field and suboptimal performance. To address this issue, we propose Adaptive Token Dictionary (ATD), a novel transformer-based architecture for image restoration that enables global dependency modeling with linear complexity relative to image size. The ATD model incorporates a learnable token dictionary, which summarizes external image priors (i.e., typical image structures) during the training process. To utilize this information, we introduce a token dictionary cross-attention (TDCA) mechanism that enhances the input features via interaction with the learned dictionary. Furthermore, we exploit the category information embedded in the TDCA attention maps to group input features into multiple categories, each representing a cluster of similar features across the image and serving as an attention group. We also integrate the learned category information into the feed-forward network to further improve feature fusion. ATD and its lightweight version ATD-light, achieve state-of-the-art performance on multiple image super-resolution benchmarks. Moreover, we develop ATD-U, a multi-scale variant of ATD, to address other image restoration tasks, including image denoising and JPEG compression artifacts removal. Extensive experiments demonstrate the superiority of out proposed models, both quantitatively and qualitatively. --- ### Joint Optimization of Model Partitioning and Resource Allocation for Anti-Jamming Collaborative Inference Systems Authors: Mengru Wu, Jiawei Li, Jiaqi Wei, Bin Lyu, Kai-Kit Wong, Hyundong Shin Published: 2026-03-03 Categories: cs.LG, eess.SY arXiv: https://arxiv.org/abs/2603.02579v1 With the increasing computational demands of deep neural network (DNN) inference on resource-constrained devices, DNN partitioning-based device-edge collaborative inference has emerged as a promising paradigm. However, the transmission of intermediate feature data is vulnerable to malicious jamming, which significantly degrades the overall inference performance. To counter this threat, this letter focuses on an anti-jamming collaborative inference system in the presence of a malicious jammer. In this system, a DNN model is partitioned into two distinct segments, which are executed by wireless devices and edge servers, respectively. We first analyze the effects of jamming and DNN partitioning on inference accuracy via data regression. Based on this, our objective is to maximize the system's revenue of delay and accuracy (RDA) under inference accuracy and computing resource constraints by jointly optimizing computation resource allocation, devices' transmit power, and DNN partitioning. To address the mixed-integer nonlinear programming problem, we propose an efficient alternating optimization-based algorithm, which decomposes the problem into three subproblems that are solved via Karush-Kuhn-Tucker conditions, convex optimization methods, and a quantum genetic algorithm, respectively. Extensive simulations demonstrate that our proposed scheme outperforms baselines in terms of RDA. ---