Showing 1–50 of 2,866 results for author: Zha, Z

WavAlign: Enhancing Intelligence and Expressiveness in Spoken Dialogue Models via Adaptive Hybrid Post-Training

Authors: Yifu Chen, Shengpeng Ji, Qian Chen, Tianle Liang, Yangzhuo Li, Ziqing Wang, Wen Wang, Jingyu Lu, Haoxiao Wang, Xueyi Pu, Fan Zhuo, Zhou Zhao

Abstract: End-to-end spoken dialogue models have garnered significant attention because they offer a higher potential ceiling in expressiveness and perceptual ability than cascaded systems. However, the intelligence and expressiveness of current open-source spoken dialogue models often remain below expectations. Motivated by the success of online reinforcement learning(RL) in other domains, one might attemp… ▽ More End-to-end spoken dialogue models have garnered significant attention because they offer a higher potential ceiling in expressiveness and perceptual ability than cascaded systems. However, the intelligence and expressiveness of current open-source spoken dialogue models often remain below expectations. Motivated by the success of online reinforcement learning(RL) in other domains, one might attempt to directly apply preference optimization to spoken dialogue models, yet this transfer is non-trivial. We analyze these obstacles from the perspectives of reward modeling and rollout sampling, focusing on how sparse preference supervision interacts with dense speech generation under shared-parameter updates. Based on the analysis, we propose a modality-aware adaptive post-training recipe that makes RL practical for spoken dialogue: it constrains preference updates to the semantic channel and improves acoustic behavior via explicit anchoring, while dynamically regulating their mixture from rollout statistics to avoid unreliable preference gradients. We evaluate the method across multiple spoken dialogue benchmarks and representative architectures, and observe consistent improvements in semantic quality and speech expressiveness. △ Less

Submitted 16 April, 2026; originally announced April 2026.

Dual-Axis Generative Reward Model Toward Semantic and Turn-taking Robustness in Interactive Spoken Dialogue Models

Authors: Yifu Chen, Shengpeng Ji, Zhengqing Liu, Qian Chen, Wen Wang, Ziqing Wang, Yangzhuo Li, Tianle Liang, Zhou Zhao

Abstract: Achieving seamless, human-like interaction remains a key challenge for full-duplex spoken dialogue models (SDMs). Reinforcement learning (RL) has substantially enhanced text- and vision-language models, while well-designed reward signals are crucial for the performance of RL. We consider RL a promising strategy to address the key challenge for SDMs. However, a fundamental barrier persists: prevail… ▽ More Achieving seamless, human-like interaction remains a key challenge for full-duplex spoken dialogue models (SDMs). Reinforcement learning (RL) has substantially enhanced text- and vision-language models, while well-designed reward signals are crucial for the performance of RL. We consider RL a promising strategy to address the key challenge for SDMs. However, a fundamental barrier persists: prevailing automated metrics for assessing interaction quality rely on superficial proxies, such as behavioral statistics or timing-prediction accuracy, failing to provide reliable reward signals for RL. On the other hand, human evaluations, despite their richness, remain costly, inconsistent, and difficult to scale. We tackle this critical barrier by proposing a Dual-Axis Generative Reward Model, which is trained to understand complex interaction dynamics using a detailed taxonomy and an annotated dataset, produces a single score and, crucially, provides separate evaluations for semantic quality and interaction timing. Such dual outputs furnish precise diagnostic feedback for SDMs and deliver a dependable, instructive reward signal suitable for online reinforcement learning. Our model achieves state-of-the-art performance on interaction-quality assessment across a wide spectrum of datasets, spanning synthetic dialogues and complex real-world interactions. △ Less

Submitted 16 April, 2026; originally announced April 2026.

Tug-of-War within A Decade: Conflict Resolution in Vulnerability Analysis via Teacher-Guided Retrieval-Augmented Generations

Authors: Ziyin Zhou, Jianyi Zhang, Xu ji, Yilong Li, Jiameng Han, Zhangchi Zhao

Abstract: Large Language Models (LLMs) are essential for analyzing and addressing vulnerabilities in cybersecurity. However, among over 200,000 vulnerabilities were discovered in the past decade, more than 30,000 have been changed or updated. This necessitates frequent updates to the training datasets and internal knowledge bases of LLMs to maintain knowledge consistency. In this paper, we focus on the prob… ▽ More Large Language Models (LLMs) are essential for analyzing and addressing vulnerabilities in cybersecurity. However, among over 200,000 vulnerabilities were discovered in the past decade, more than 30,000 have been changed or updated. This necessitates frequent updates to the training datasets and internal knowledge bases of LLMs to maintain knowledge consistency. In this paper, we focus on the problem of knowledge discrepancy and conflict within CVE (Common Vulnerabilities and Exposures) detection and analysis. This problem hinders LLMs' ability to retrieve the latest knowledge from original training datasets, leading to knowledge conflicts, fabrications of factually incorrect results, and generation hallucinations. To address this problem, we propose an innovative two-stage framework called CRVA-TGRAG (Conflict Resolution in Vulnerability Analysis via Teacher-Guided Retrieval-Augmented Generation). First, to improve document retrieval accuracy during the retrieval stage, we utilize Parent Document Segmentation and an ensemble retrieval scheme based on semantic similarity and inverted indexing. Second, to enhance LLMs' capabilities based on the retrieval of CVE dataset in generation stage, we employ a teacher-guided preference optimization technique to fine-tune LLMs. Our framework not only enhances the quality of content retrieval through RAG but also leverages the advantages of preference fine-tuning in LLMs to answer questions more effectively and precisely. Experiments demonstrate our method achieves higher accuracy in retrieving the latest CVEs compared to external knowledge bases. In conclusion, our framework significantly mitigates potential knowledge conflicts and inconsistencies that may arise from relying solely on LLMs for knowledge retrieval. △ Less

Submitted 25 March, 2026; originally announced April 2026.

ASTRA: Enhancing Multi-Subject Generation with Retrieval-Augmented Pose Guidance and Disentangled Position Embedding

Authors: Tianze Xia, Zijian Ning, Zonglin Zhao, Mingjia Wang

Abstract: Subject-driven image generation has shown great success in creating personalized content, but its capabilities are largely confined to single subjects in common poses. Current approaches face a fundamental conflict when handling multiple subjects with complex, distinct actions: preserving individual identities while enforcing precise pose structures. This challenge often leads to identity fusion a… ▽ More Subject-driven image generation has shown great success in creating personalized content, but its capabilities are largely confined to single subjects in common poses. Current approaches face a fundamental conflict when handling multiple subjects with complex, distinct actions: preserving individual identities while enforcing precise pose structures. This challenge often leads to identity fusion and pose distortion, as appearance and structure signals become entangled within the model's architecture. To resolve this conflict, we introduce ASTRA(Adaptive Synthesis through Targeted Retrieval Augmentation), a novel framework that architecturally disentangles subject appearance from pose structure within a unified Diffusion Transformer. ASTRA achieves this through a dual-pronged strategy. It first employs a Retrieval-Augmented Pose (RAG-Pose) pipeline to provide a clean, explicit structural prior from a curated database. Then, its core generative model learns to process these dual visual conditions using our Enhanced Universal Rotary Position Embedding (EURoPE), an asymmetric encoding mechanism that decouples identity tokens from spatial locations while binding pose tokens to the canvas. Concurrently, a Disentangled Semantic Modulation (DSM) adapter offloads the identity preservation task into the text conditioning stream. Extensive experiments demonstrate that our integrated approach achieves superior disentanglement. On our designed COCO-based complex pose benchmark, ASTRA achieves a new state-of-the-art in pose adherence, while maintaining high identity fidelity and text alignment in DreamBench. △ Less

Submitted 15 April, 2026; originally announced April 2026.

Character Beyond Speech: Leveraging Role-Playing Evaluation in Audio Large Language Models via Reinforcement Learning

Authors: Dongjie Fu, Fangming Feng, Xize Cheng, Linjun Li, Zhou Zhao, Tao Jin

Abstract: The rapid evolution of multimodal large models has revolutionized the simulation of diverse characters in speech dialogue systems, enabling a novel interactive paradigm. Character attributes are manifested not only in textual responses but also through vocal features, as speech conveys rich paralinguistic information that is challenging to quantify. This poses significant difficulties in evaluatin… ▽ More The rapid evolution of multimodal large models has revolutionized the simulation of diverse characters in speech dialogue systems, enabling a novel interactive paradigm. Character attributes are manifested not only in textual responses but also through vocal features, as speech conveys rich paralinguistic information that is challenging to quantify. This poses significant difficulties in evaluating the character alignment of role-playing agents. To address these challenges, we present RoleJudge, an evaluation framework that leverages audio large language models to systematically assess the alignment between speech and character across multiple modalities and dimensions. Furthermore, we introduce RoleChat, the first voice role-playing evaluation dataset enriched with chain-of-thought reasoning annotations, comprising a diverse set of authentic and LLM-generated speech samples. Utilizing this dataset, we implement a multi-stage training paradigm and incorporate Standard Alignment in reinforcement learning to mitigate reward misalignment during optimization. Experimental results in terms of accuracy and subjective assessment demonstrate that RoleJudge outperforms various baseline models, validating the effectiveness of our multidimensional evaluation framework. △ Less

Submitted 15 April, 2026; originally announced April 2026.

Figma2Code: Automating Multimodal Design to Code in the Wild

Authors: Yi Gui, Jiawan Zhang, Yina Wang, Tianran Ma, Yao Wan, Shilin He, Dongping Chen, Zhou Zhao, Wenbin Jiang, Xuanhua Shi, Hai Jin, Philip S Yu

Abstract: Front-end development constitutes a substantial portion of software engineering, yet converting design mockups into production-ready User Interface (UI) code remains tedious and costly. While recent work has explored automating this process with Multimodal Large Language Models (MLLMs), existing approaches typically rely solely on design images. As a result, they must infer complex UI details from… ▽ More Front-end development constitutes a substantial portion of software engineering, yet converting design mockups into production-ready User Interface (UI) code remains tedious and costly. While recent work has explored automating this process with Multimodal Large Language Models (MLLMs), existing approaches typically rely solely on design images. As a result, they must infer complex UI details from images alone, often leading to degraded results. In real-world development workflows, however, design mockups are usually delivered as Figma files, a widely used tool for front-end design, that embed rich multimodal information (e.g., metadata and assets) essential for generating high-quality UI. To bridge this gap, we introduce Figma2Code, a new task that advances design-to-code into a multimodal setting and aims to automate design-to-code in the wild. Specifically, we collect paired design images and their corresponding metadata files from the Figma community. We then apply a series of processing operations, including rule-based filtering, human- and MLLM-based annotation and screening, and metadata refinement. This process yields 3,055 samples, from which designers curate a balanced dataset of 213 high-quality cases. Using this dataset, we benchmark ten state-of-the-art open-source and proprietary MLLMs. Our results show that while proprietary models achieve superior visual fidelity, they remain limited in layout responsiveness and code maintainability. Further experiments across modalities and ablation studies corroborate this limitation, partly due to models' tendency to directly map primitive visual attributes from Figma metadata. △ Less

Submitted 15 April, 2026; originally announced April 2026.

Comments: ICLR 2026

Bridging MARL to SARL: An Order-Independent Multi-Agent Transformer via Latent Consensus

Authors: Zijian Zhao, Jing Gao, Sen Li

Abstract: Cooperative multi-agent reinforcement learning (MARL) is widely used to address large joint observation and action spaces by decomposing a centralized control problem into multiple interacting agents. However, such decomposition often introduces additional challenges, including non-stationarity, unstable training, weak coordination, and limited theoretical guarantees. In this paper, we propose the… ▽ More Cooperative multi-agent reinforcement learning (MARL) is widely used to address large joint observation and action spaces by decomposing a centralized control problem into multiple interacting agents. However, such decomposition often introduces additional challenges, including non-stationarity, unstable training, weak coordination, and limited theoretical guarantees. In this paper, we propose the Consensus Multi-Agent Transformer (CMAT), a centralized framework that bridges cooperative MARL to a hierarchical single-agent reinforcement learning (SARL) formulation. CMAT treats all agents as a unified entity and employs a Transformer encoder to process the large joint observation space. To handle the extensive joint action space, we introduce a hierarchical decision-making mechanism in which a Transformer decoder autoregressively generates a high-level consensus vector, simulating the process by which agents reach agreement on their strategies in latent space. Conditioned on this consensus, all agents generate their actions simultaneously, enabling order-independent joint decision making and avoiding the sensitivity to action-generation order in conventional Multi-Agent Transformers (MAT). This factorization allows the joint policy to be optimized using single-agent PPO while preserving expressive coordination through the latent consensus. To evaluate the proposed method, we conduct experiments on benchmark tasks from StarCraft II, Multi-Agent MuJoCo, and Google Research Football. The results show that CMAT achieves superior performance over recent centralized solutions, sequential MARL methods, and conventional MARL baselines. The code for this paper is available at:https://github.com/RS2002/CMAT . △ Less

Submitted 15 April, 2026; originally announced April 2026.

Turán-Theoretic Bounds on Several Elementary Trapping Sets in LDPC Codes

Authors: Ziyang Zhao, Haoran Xiong, Zicheng Ye, Guiying Yan

Abstract: LDPC codes have attracted significant attention because of their superior performance close to the Shannon limit. Elementary trapping sets are the main cause of the error floor phenomenon in LDPC codes. We consider typical graphs related to trapping sets, including theta graphs, dumbbell graphs, and short cycles with chords. Based on the Turán numbers of $θ(2,2,2)$, $θ(1,3,3)$ and $D(4,4;0)$, we p… ▽ More LDPC codes have attracted significant attention because of their superior performance close to the Shannon limit. Elementary trapping sets are the main cause of the error floor phenomenon in LDPC codes. We consider typical graphs related to trapping sets, including theta graphs, dumbbell graphs, and short cycles with chords. Based on the Turán numbers of $θ(2,2,2)$, $θ(1,3,3)$ and $D(4,4;0)$, we prove that any $(a,b)$-ETS with $g=8$ variable-regular $γ$ satisfies the inequality $b\geq aγ-\frac{a(\sqrt{24a-23}-1)}{4}$, provided that any two 8-cycles in the Tanner graph do not share common variable node. In addition, we can also eliminate ETSs by removing certain short-cycle structures with chords. The minimum sizes of ETSs obtained through these methods are significantly increased. To assess practical impact , we analyze spectral radii of the ETSs and construct QC-LDPC codes to show frame error rates in the error floor region. △ Less

Submitted 14 April, 2026; originally announced April 2026.

AnyPoC: Universal Proof-of-Concept Test Generation for Scalable LLM-Based Bug Detection

Authors: Zijie Zhao, Chenyuan Yang, Weidong Wang, Yihan Yang, Ziqi Zhang, Lingming Zhang

Abstract: While recent LLM-based agents can identify many candidate bugs in source code, their reports remain static hypotheses that require manual validation, limiting the practicality of automated bug detection. We frame this challenge as a test generation task: given a candidate report, synthesizing an executable proof-of-concept test, or simply a PoC - such as a script, command sequence, or crafted inpu… ▽ More While recent LLM-based agents can identify many candidate bugs in source code, their reports remain static hypotheses that require manual validation, limiting the practicality of automated bug detection. We frame this challenge as a test generation task: given a candidate report, synthesizing an executable proof-of-concept test, or simply a PoC - such as a script, command sequence, or crafted input - to trigger the suspected defect. Automated PoC generation can act as a scalable validation oracle, enabling end-to-end autonomous bug detection by providing concrete execution evidence. However, naive LLM agents are unreliable validators: they are biased toward "success" and may reward-hack by producing plausible but non-functional PoCs or even hallucinated traces. To address this, we present AnyPoC, a general multi-agent framework that (1) analyzes and fact-checks a candidate bug report, (2) iteratively synthesizes and executes a PoC while collecting execution traces, and (3) independently re-executes and scrutinizes the PoC to mitigate hallucination and reward hacking. In addition, AnyPoC also continuously extracts and evolves a PoC knowledge base to handle heterogeneous tasks. AnyPoC operates on candidate bug reports regardless of their source and can be paired with different bug reporters. To demonstrate practicality and generality, we apply AnyPoC, with a simple agentic bug reporter, on 12 critical software systems across diverse languages/domains (many with millions of lines of code) including Firefox, Chromium, LLVM, OpenSSL, SQLite, FFmpeg, and Redis. Compared to the state-of-the-art coding agents, e.g., Claude Code and Codex, AnyPoC produces 1.3x more valid PoCs for true-positive bug reports and rejects 9.8x more false-positive bug reports. To date, AnyPoC has discovered 122 new bugs (105 confirmed, 86 already fixed), with 45 generated PoCs adopted as official regression tests. △ Less

Submitted 13 April, 2026; originally announced April 2026.

POINTS-Long: Adaptive Dual-Mode Visual Reasoning in MLLMs

Authors: Haicheng Wang, Yuan Liu, Yikun Liu, Zhemeng Yu, Zhongyin Zhao, Yangxiu You, Zilin Yu, Le Tian, Xiao Zhou, Jie Zhou, Weidi Xie, Yanfeng Wang

Abstract: Multimodal Large Language Models (MLLMs) have recently demonstrated remarkable capabilities in cross-modal understanding and generation. However, the rapid growth of visual token sequences--especially in long-video and streaming scenarios--poses a major challenge to their scalability and real-world deployment. Thus, we introduce POINTS-Long, a native dual-mode MLLM featuring dynamic visual token s… ▽ More Multimodal Large Language Models (MLLMs) have recently demonstrated remarkable capabilities in cross-modal understanding and generation. However, the rapid growth of visual token sequences--especially in long-video and streaming scenarios--poses a major challenge to their scalability and real-world deployment. Thus, we introduce POINTS-Long, a native dual-mode MLLM featuring dynamic visual token scaling inspired by the human visual system. The model supports two complementary perception modes: focus mode and standby mode, enabling users to dynamically trade off efficiency and accuracy during inference. On fine-grained visual tasks, the focus mode retains the optimal performance, while on long-form general visual understanding, the standby mode retains 97.7-99.7% of the original accuracy using only 1/40-1/10th of the visual tokens. Moreover, POINTS-Long natively supports streaming visual understanding via a dynamically detachable KV-cache design, allowing efficient maintenance of ultra-long visual memory. Our work provides new insights into the design of future MLLMs and lays the foundation for adaptive and efficient long-form visual understanding. △ Less

Submitted 13 April, 2026; originally announced April 2026.

HumDial-EIBench: A Human-Recorded Multi-Turn Emotional Intelligence Benchmark for Audio Language Models

Authors: Shuiyuan Wang, Zhixian Zhao, Hongfei Yue, Chengyou Wang, Shuai Wang, Hui Bu, Xin Xu, Lei Xie

Abstract: Evaluating the emotional intelligence (EI) of audio language models (ALMs) is critical. However, existing benchmarks mostly rely on synthesized speech, are limited to single-turn interactions, and depend heavily on open-ended scoring. This paper proposes HumDial-EIBench, a comprehensive benchmark for evaluating ALMs' EI. Using real-recorded human dialogues from the ICASSP 2026 HumDial Challenge, i… ▽ More Evaluating the emotional intelligence (EI) of audio language models (ALMs) is critical. However, existing benchmarks mostly rely on synthesized speech, are limited to single-turn interactions, and depend heavily on open-ended scoring. This paper proposes HumDial-EIBench, a comprehensive benchmark for evaluating ALMs' EI. Using real-recorded human dialogues from the ICASSP 2026 HumDial Challenge, it reformulates emotional tracking and causal reasoning into multiple-choice questions with adversarial distractors, mitigating subjective scoring bias for cognitive tasks. It retains the generation of empathetic responses and introduces an acoustic-semantic conflict task to assess robustness against contradictory multimodal signals. Evaluations of eight ALMs reveal that most models struggle with multi-turn emotional tracking and implicit causal reasoning. Furthermore, all models exhibit decoupled textual and acoustic empathy, alongside a severe text-dominance bias during cross-modal conflicts. △ Less

Submitted 13 April, 2026; originally announced April 2026.

Finetune Like You Pretrain: Boosting Zero-shot Adversarial Robustness in Vision-language Models

Authors: Songlong Xing, Weijie Wang, Zhengyu Zhao, Jindong Gu, Philip Torr, Nicu Sebe

Abstract: Despite their impressive zero-shot abilities, vision-language models such as CLIP have been shown to be susceptible to adversarial attacks. To enhance its adversarial robustness, recent studies finetune the pretrained vision encoder of CLIP with adversarial examples on a proxy dataset such as ImageNet by aligning adversarial images with correct class labels. However, these methods overlook the imp… ▽ More Despite their impressive zero-shot abilities, vision-language models such as CLIP have been shown to be susceptible to adversarial attacks. To enhance its adversarial robustness, recent studies finetune the pretrained vision encoder of CLIP with adversarial examples on a proxy dataset such as ImageNet by aligning adversarial images with correct class labels. However, these methods overlook the important roles of training data distributions and learning objectives, resulting in reduced zero-shot capabilities and limited transferability of robustness across domains and datasets. In this work, we propose a simple yet effective paradigm AdvFLYP, which follows the training recipe of CLIP's pretraining process when performing adversarial finetuning to the model. Specifically, AdvFLYP finetunes CLIP with adversarial images created based on image-text pairs collected from the web, and match them with their corresponding texts via a contrastive loss. To alleviate distortion of adversarial image embeddings of noisy web images, we further propose to regularise AdvFLYP by penalising deviation of adversarial image features. We show that logit- and feature-level regularisation terms benefit robustness and clean accuracy, respectively. Extensive experiments on 14 downstream datasets spanning various domains show the superiority of our paradigm over mainstream practices. Our code and model weights are released at https://github.com/Sxing2/AdvFLYP. △ Less

Submitted 13 April, 2026; originally announced April 2026.

Comments: Accepted to CVPR Findings Track 2026

Online Reasoning Video Object Segmentation

Authors: Jinyuan Liu, Yang Wang, Zeyu Zhao, Weixin Li, Song Wang, Ruize Han

Abstract: Reasoning video object segmentation predicts pixel-level masks in videos from natural-language queries that may involve implicit and temporally grounded references. However, existing methods are developed and evaluated in an offline regime, where the entire video is available at inference time and future frames can be exploited for retrospective disambiguation, deviating from real-world deployment… ▽ More Reasoning video object segmentation predicts pixel-level masks in videos from natural-language queries that may involve implicit and temporally grounded references. However, existing methods are developed and evaluated in an offline regime, where the entire video is available at inference time and future frames can be exploited for retrospective disambiguation, deviating from real-world deployments that require strictly causal, frame-by-frame decisions. We study Online Reasoning Video Object Segmentation (ORVOS), where models must incrementally interpret queries using only past and current frames without revisiting previous predictions, while handling referent shifts as events unfold. To support evaluation, we introduce ORVOSB, a benchmark with frame-level causal annotations and referent-shift labels, comprising 210 videos, 12,907 annotated frames, and 512 queries across five reasoning categories. We further propose a baseline with continually-updated segmentation prompts and a structured temporal token reservoir for long-horizon reasoning under bounded computation. Experiments show that existing methods struggle under strict causality and referent shifts, while our baseline establishes a strong foundation for future research. △ Less

Submitted 13 April, 2026; originally announced April 2026.

A Faster Path to Continual Learning

Authors: Wei Li, Hangjie Yuan, Zixiang Zhao, Borui Kang, Ziwei Liu, Tao Feng

Abstract: Continual Learning (CL) aims to train neural networks on a dynamic stream of tasks without forgetting previously learned knowledge. Among optimization-based approaches, C-Flat has emerged as a promising solution due to its plug-and-play nature and its ability to encourage uniformly low-loss regions for both new and old tasks. However, C-Flat requires three additional gradient computations per iter… ▽ More Continual Learning (CL) aims to train neural networks on a dynamic stream of tasks without forgetting previously learned knowledge. Among optimization-based approaches, C-Flat has emerged as a promising solution due to its plug-and-play nature and its ability to encourage uniformly low-loss regions for both new and old tasks. However, C-Flat requires three additional gradient computations per iteration, imposing substantial overhead on the optimization process. In this work, we propose C-Flat Turbo, a faster yet stronger optimizer that significantly reduces the training cost. We show that the gradients associated with first-order flatness contain direction-invariant components relative to the proxy-model gradients, enabling us to skip redundant gradient computations in the perturbed ascent steps. Moreover, we observe that these flatness-promoting gradients progressively stabilize across tasks, which motivates a linear scheduling strategy with an adaptive trigger to allocate larger turbo steps for later tasks. Experiments show that C-Flat Turbo is 1.0$\times$ to 1.25$\times$ faster than C-Flat across a wide range of CL methods, while achieving comparable or even improved accuracy. △ Less

Submitted 14 April, 2026; v1 submitted 13 April, 2026; originally announced April 2026.

Comments: Update Author Affiliations

BridgeSim: Unveiling the OL-CL Gap in End-to-End Autonomous Driving

Authors: Seth Z. Zhao, Luobin Wang, Hongwei Ruan, Yuxin Bao, Yilan Chen, Ziyang Leng, Abhijit Ravichandran, Honglin He, Zewei Zhou, Xu Han, Abhishek Peri, Zhiyu Huang, Pranav Desai, Henrik Christensen, Jiaqi Ma, Bolei Zhou

Abstract: Open-loop (OL) to closed-loop (CL) gap (OL-CL gap) exists when OL-pretrained policies scoring high in OL evaluations fail to transfer effectively in closed-loop (CL) deployment. In this paper, we unveil the root causes of this systemic failure and propose a practical remedy. Specifically, we demonstrate that OL policies suffer from Observational Domain Shift and Objective Mismatch. We show that wh… ▽ More Open-loop (OL) to closed-loop (CL) gap (OL-CL gap) exists when OL-pretrained policies scoring high in OL evaluations fail to transfer effectively in closed-loop (CL) deployment. In this paper, we unveil the root causes of this systemic failure and propose a practical remedy. Specifically, we demonstrate that OL policies suffer from Observational Domain Shift and Objective Mismatch. We show that while the former is largely recoverable with adaptation techniques, the latter creates a structural inability to model complex reactive behaviors, which forms the primary OL-CL gap. We find that a wide range of OL policies learn a biased Q-value estimator that neglects both the reactive nature of CL simulations and the temporal awareness needed to reduce compounding errors. To this end, we propose a Test-Time Adaptation (TTA) framework that calibrates observational shift, reduces state-action biases, and enforces temporal consistency. Extensive experiments show that TTA effectively mitigates planning biases and yields superior scaling dynamics than its baseline counterparts. Furthermore, our analysis highlights the existence of blind spots in standard OL evaluation protocols that fail to capture the realities of closed-loop deployment. △ Less

Submitted 12 April, 2026; originally announced April 2026.

From Perception to Planning: Evolving Ego-Centric Task-Oriented Spatiotemporal Reasoning via Curriculum Learning

Authors: Xiaoda Yang, Yuxiang Liu, Shenzhou Gao, Can Wang, Jingyang Xue, Lixin Yang, Yao Mu, Tao Jin, Shuicheng Yan, Zhimeng Zhang, Zhou Zhao

Abstract: Modern vision-language models achieve strong performance in static perception, but remain limited in the complex spatiotemporal reasoning required for embodied, egocentric tasks. A major source of failure is their reliance on temporal priors learned from passive video data, which often leads to spatiotemporal hallucinations and poor generalization in dynamic environments. To address this, we prese… ▽ More Modern vision-language models achieve strong performance in static perception, but remain limited in the complex spatiotemporal reasoning required for embodied, egocentric tasks. A major source of failure is their reliance on temporal priors learned from passive video data, which often leads to spatiotemporal hallucinations and poor generalization in dynamic environments. To address this, we present EgoTSR, a curriculum-based framework for learning task-oriented spatiotemporal reasoning. EgoTSR is built on the premise that embodied reasoning should evolve from explicit spatial understanding to internalized task-state assessment and finally to long-horizon planning. To support this paradigm, we construct EgoTSR-Data, a large-scale dataset comprising 46 million samples organized into three stages: Chain-of-Thought (CoT) supervision, weakly supervised tagging, and long-horizon sequences. Extensive experiments demonstrate that EgoTSR effectively eliminates chronological biases, achieving 92.4% accuracy on long-horizon logical reasoning tasks while maintaining high fine-grained perceptual precision, significantly outperforming existing open-source and closed-source state-of-the-art models. △ Less

Submitted 12 April, 2026; originally announced April 2026.

A Progressive Training Strategy for Vision-Language Models to Counteract Spatio-Temporal Hallucinations in Embodied Reasoning

Authors: Xiaoda Yang, Shuai Yang, Can Wang, Jingyang Xue, Menglan Tang, Checheng Yu, Xunzhe Zhou, Sashuai Zhou, Tao Jin, Lixin Yang, Xiangyu Yue, Zhou Zhao

Abstract: Vision-Language Models (VLMs) have made significant strides in static image understanding but continue to face critical hurdles in spatiotemporal reasoning. A major bottleneck is "multi-image reasoning hallucination", where a massive performance drop between forward and reverse temporal queries reveals a dependence on superficial shortcuts instead of genuine causal understanding. To mitigate this,… ▽ More Vision-Language Models (VLMs) have made significant strides in static image understanding but continue to face critical hurdles in spatiotemporal reasoning. A major bottleneck is "multi-image reasoning hallucination", where a massive performance drop between forward and reverse temporal queries reveals a dependence on superficial shortcuts instead of genuine causal understanding. To mitigate this, we first develop a new Chain-of-Thought (CoT) dataset that decomposes intricate reasoning into detailed spatiotemporal steps and definitive judgments. Building on this, we present a progressive training framework: it initiates with supervised pre-training on our CoT dataset to instill logical structures, followed by fine-tuning with scalable weakly-labeled data for broader generalization. Our experiments demonstrate that this approach not only improves backbone accuracy but also slashes the forward-backward performance gap from over 70\% to only 6.53\%. This confirms the method's ability to develop authentic dynamic reasoning and reduce the inherent temporal biases of current VLMs. △ Less

Submitted 12 April, 2026; originally announced April 2026.

Inductive Reasoning for Temporal Knowledge Graphs with Emerging Entities

Authors: Ze Zhao, Yuhui He, Lyuwen Wu, Gu Tang, Bin Lu, Xiaoying Gan, Luoyi Fu, Xinbing Wang, Chenghu Zhou

Abstract: Reasoning on Temporal Knowledge Graphs (TKGs) is essential for predicting future events and time-aware facts. While existing methods are effective at capturing relational dynamics, their performance is limited by a closed-world assumption, which fails to account for emerging entities not present in the training. Notably, these entities continuously join the network without historical interactions.… ▽ More Reasoning on Temporal Knowledge Graphs (TKGs) is essential for predicting future events and time-aware facts. While existing methods are effective at capturing relational dynamics, their performance is limited by a closed-world assumption, which fails to account for emerging entities not present in the training. Notably, these entities continuously join the network without historical interactions. Empirical study reveals that emerging entities are widespread in TKGs, comprising roughly 25\% of all entities. The absence of historical interactions of these entities leads to significant performance degradation in reasoning tasks. Whereas, we observe that entities with semantic similarities often exhibit comparable interaction histories, suggesting the presence of transferable temporal patterns. Inspired by this insight, we propose TransFIR (Transferable Inductive Reasoning), a novel framework that leverages historical interaction sequences from semantically similar known entities to support inductive reasoning. Specifically, we propose a codebook-based classifier that categorizes emerging entities into latent semantic clusters, allowing them to adopt reasoning patterns from similar entities. Experimental results demonstrate that TransFIR outperforms all baselines in reasoning on emerging entities, achieving an average improvement of 28.6% in Mean Reciprocal Rank (MRR) across multiple datasets. The implementations are available at https://github.com/zhaodazhuang2333/TransFIR. △ Less

Submitted 11 April, 2026; originally announced April 2026.

Comments: 24 pages, accepted by ICLR2026

PhysInOne: Visual Physics Learning and Reasoning in One Suite

Authors: Siyuan Zhou, Hejun Wang, Hu Cheng, Jinxi Li, Dongsheng Wang, Junwei Jiang, Yixiao Jin, Jiayue Huang, Shiwei Mao, Shangjia Liu, Yafei Yang, Hongkang Song, Shenxing Wei, Zihui Zhang, Peng Huang, Shijie Liu, Zhengli Hao, Hao Li, Yitian Li, Wenqi Zhou, Zhihan Zhao, Zongqi He, Hongtao Wen, Shouwang Huang, Peng Yun , et al. (14 additional authors not shown)

Abstract: We present PhysInOne, a large-scale synthetic dataset addressing the critical scarcity of physically-grounded training data for AI systems. Unlike existing datasets limited to merely hundreds or thousands of examples, PhysInOne provides 2 million videos across 153,810 dynamic 3D scenes, covering 71 basic physical phenomena in mechanics, optics, fluid dynamics, and magnetism. Distinct from previous… ▽ More We present PhysInOne, a large-scale synthetic dataset addressing the critical scarcity of physically-grounded training data for AI systems. Unlike existing datasets limited to merely hundreds or thousands of examples, PhysInOne provides 2 million videos across 153,810 dynamic 3D scenes, covering 71 basic physical phenomena in mechanics, optics, fluid dynamics, and magnetism. Distinct from previous works, our scenes feature multiobject interactions against complex backgrounds, with comprehensive ground-truth annotations including 3D geometry, semantics, dynamic motion, physical properties, and text descriptions. We demonstrate PhysInOne's efficacy across four emerging applications: physics-aware video generation, long-/short-term future frame prediction, physical property estimation, and motion transfer. Experiments show that fine-tuning foundation models on PhysInOne significantly enhances physical plausibility, while also exposing critical gaps in modeling complex physical dynamics and estimating intrinsic properties. As the largest dataset of its kind, orders of magnitude beyond prior works, PhysInOne establishes a new benchmark for advancing physics-grounded world models in generation, simulation, and embodied AI. △ Less

Submitted 10 April, 2026; originally announced April 2026.

Comments: CVPR 2026. Siyuan, Hejun, Hu, Jinxi, Dongsheng, Junwei, Yixiao, Jiayue, and Shiwei are co-first authors. Project page: https://vlar-group.github.io/PhysInOne.html

Compositional-Degradation UAV Image Restoration: Conditional Decoupled MoE Network and A Benchmark

Authors: Jinquan Yan, Zhicheng Zhao, Zhengzheng Tu, Chenglong Li, Jin Tang, Bin Luo

Abstract: UAV images are critical for applications such as large-area mapping, infrastructure inspection, and emergency response. However, in real-world flight environments, a single image is often affected by multiple degradation factors, including rain, haze, and noise, undermining downstream task performance. Current unified restoration approaches typically rely on implicit degradation representations th… ▽ More UAV images are critical for applications such as large-area mapping, infrastructure inspection, and emergency response. However, in real-world flight environments, a single image is often affected by multiple degradation factors, including rain, haze, and noise, undermining downstream task performance. Current unified restoration approaches typically rely on implicit degradation representations that entangle multiple factors into a single condition, causing mutual interference among heterogeneous corrections. To this end, we propose DAME-Net, a Degradation-Aware Mixture-of-Experts Network that decouples explicit degradation perception from degradation-conditioned reconstruction for compositional UAV image restoration. Specifically, we design a Factor-wise Degradation Perception module(FDPM) to provide explicit per-factor degradation cues for the restoration stage through multi-label prediction with label-similarity-guided soft alignment, replacing implicit entangled conditions with interpretable and generalizable degradation descriptions. Moreover, we develop a Conditioned Decoupled MoE module(CDMM) that leverages these cues for stage-wise conditioning, spatial-frequency hybrid processing, and mask-constrained decoupled expert routing, enabling selective factor-specific correction while suppressing irrelevant interference. In addition, we construct the Multi-Degradation UAV Restoration benchmark (MDUR), the first large-scale UAV benchmark for compositional UAV image restoration, with 43 degradation configurations from single degradations to four-factor composites and standardized seen/unseen splits.Extensive experiments on MDUR demonstrate consistent improvements over representative unified restoration methods, with greater gains on unseen and higher-order composite degradations. Downstream experiments further validate benefits for UAV object detection. △ Less

Submitted 10 April, 2026; originally announced April 2026.

Guaranteeing Knowledge Integration with Joint Decoding for Retrieval-Augmented Generation

Authors: Zhengyi Zhao, Shubo Zhang, Zezhong Wang, Yuxi Zhang, Huimin Wang, Yutian Zhao, Yefeng Zheng, Binyang Li, Kam-Fai Wong, Xian Wu

Abstract: Retrieval-Augmented Generation (RAG) significantly enhances Large Language Models (LLMs) by providing access to external knowledge. However, current research primarily focuses on retrieval quality, often overlooking the critical ''integration bottleneck'': even when relevant documents are retrieved, LLMs frequently fail to utilize them effectively due to conflicts with their internal parametric kn… ▽ More Retrieval-Augmented Generation (RAG) significantly enhances Large Language Models (LLMs) by providing access to external knowledge. However, current research primarily focuses on retrieval quality, often overlooking the critical ''integration bottleneck'': even when relevant documents are retrieved, LLMs frequently fail to utilize them effectively due to conflicts with their internal parametric knowledge. In this paper, we argue that implicitly resolving this conflict in a single generation pass is suboptimal. We introduce GuarantRAG, a framework that explicitly decouples reasoning from evidence integration. First, we generate an ''Inner-Answer'' based solely on parametric knowledge to capture the model's reasoning flow. Second, to guarantee faithful evidence extraction, we generate a ''Refer-Answer'' using a novel Contrastive DPO objective. This objective treats the parametric Inner-Answer as a negative constraint and the retrieved documents as positive ground truth, forcing the model to suppress internal hallucinations in favor of external evidence during this phase. Finally, rather than naive concatenation or using the DPO trained model directly, we propose a joint decoding mechanism that dynamically fuses the logical coherence of the Inner-Answer with the factual precision of the Refer-Answer at the token level. Experiments on five QA benchmarks demonstrate that GuarantRAG improves accuracy by up to 12.1% and reduces hallucinations by 16.3% compared to standard and dynamic RAG baselines. △ Less

Submitted 15 April, 2026; v1 submitted 9 April, 2026; originally announced April 2026.

ImVideoEdit: Image-learning Video Editing via 2D Spatial Difference Attention Blocks

Authors: Jiayang Xu, Fan Zhuo, Majun Zhang, Changhao Pan, Zehan Wang, Siyu Chen, Xiaoda Yang, Tao Jin, Zhou Zhao

Abstract: Current video editing models often rely on expensive paired video data, which limits their practical scalability. In essence, most video editing tasks can be formulated as a decoupled spatiotemporal process, where the temporal dynamics of the pretrained model are preserved while spatial content is selectively and precisely modified. Based on this insight, we propose ImVideoEdit, an efficient frame… ▽ More Current video editing models often rely on expensive paired video data, which limits their practical scalability. In essence, most video editing tasks can be formulated as a decoupled spatiotemporal process, where the temporal dynamics of the pretrained model are preserved while spatial content is selectively and precisely modified. Based on this insight, we propose ImVideoEdit, an efficient framework that learns video editing capabilities entirely from image pairs. By freezing the pre-trained 3D attention modules and treating images as single-frame videos, we decouple the 2D spatial learning process to help preserve the original temporal dynamics. The core of our approach is a Predict-Update Spatial Difference Attention module that progressively extracts and injects spatial differences. Rather than relying on rigid external masks, we incorporate a Text-Guided Dynamic Semantic Gating mechanism for adaptive and implicit text-driven modifications. Despite training on only 13K image pairs for 5 epochs with exceptionally low computational overhead, ImVideoEdit achieves editing fidelity and temporal consistency comparable to larger models trained on extensive video datasets. △ Less

Submitted 9 April, 2026; originally announced April 2026.

Personalized RewardBench: Evaluating Reward Models with Human Aligned Personalization

Authors: Qiyao Ma, Dechen Gao, Rui Cai, Boqi Zhao, Hanchu Zhou, Junshan Zhang, Zhe Zhao

Abstract: Pluralistic alignment has emerged as a critical frontier in the development of Large Language Models (LLMs), with reward models (RMs) serving as a central mechanism for capturing diverse human values. While benchmarks for general response quality are prevalent, evaluating how well reward models account for individual user preferences remains an open challenge. To bridge this gap, we introduce Pers… ▽ More Pluralistic alignment has emerged as a critical frontier in the development of Large Language Models (LLMs), with reward models (RMs) serving as a central mechanism for capturing diverse human values. While benchmarks for general response quality are prevalent, evaluating how well reward models account for individual user preferences remains an open challenge. To bridge this gap, we introduce Personalized RewardBench, a novel benchmark designed to rigorously assess reward models' capacity to model personalized preferences. We construct chosen and rejected response pairs based on strict adherence to (or violation of) user-specific rubrics, ensuring that preference distinctions are uniquely tailored to the individual. In particular, human evaluations confirm that the primary discriminative factor between pairs is strictly personal preference, with both responses maintaining high general quality (e.g., correctness, relevance and helpfulness). Extensive testing reveals that existing state-of-the-art reward models struggle significantly with personalization, peaking at an accuracy of just 75.94%. Crucially, because an effective reward model benchmark should predict a reward model's performance on downstream tasks, we conduct experiments demonstrating that our benchmark exhibits a significantly higher correlation with downstream performance in both Best-of-N (BoN) sampling and Proximal Policy Optimization (PPO) compared to existing baselines. These findings establish Personalized RewardBench as a robust and accurate proxy for evaluating reward models' performance in downstream applications. △ Less

Submitted 8 April, 2026; originally announced April 2026.

INSPATIO-WORLD: A Real-Time 4D World Simulator via Spatiotemporal Autoregressive Modeling

Authors: InSpatio Team, Donghui Shen, Guofeng Zhang, Haomin Liu, Haoyu Ji, Hujun Bao, Hongjia Zhai, Jialin Liu, Jing Guo, Nan Wang, Siji Pan, Weihong Pan, Weijian Xie, Xianbin Liu, Xiaojun Xiang, Xiaoyu Zhang, Xinyu Chen, Yifu Wang, Yipeng Chen, Zhenzhou Fan, Zhewen Le, Zhichao Ye, Ziqiang Zhao

Abstract: Building world models with spatial consistency and real-time interactivity remains a fundamental challenge in computer vision. Current video generation paradigms often struggle with a lack of spatial persistence and insufficient visual realism, making it difficult to support seamless navigation in complex environments. To address these challenges, we propose INSPATIO-WORLD, a novel real-time frame… ▽ More Building world models with spatial consistency and real-time interactivity remains a fundamental challenge in computer vision. Current video generation paradigms often struggle with a lack of spatial persistence and insufficient visual realism, making it difficult to support seamless navigation in complex environments. To address these challenges, we propose INSPATIO-WORLD, a novel real-time framework capable of recovering and generating high-fidelity, dynamic interactive scenes from a single reference video. At the core of our approach is a Spatiotemporal Autoregressive (STAR) architecture, which enables consistent and controllable scene evolution through two tightly coupled components: Implicit Spatiotemporal Cache aggregates reference and historical observations into a latent world representation, ensuring global consistency during long-horizon navigation; Explicit Spatial Constraint Module enforces geometric structure and translates user interactions into precise and physically plausible camera trajectories. Furthermore, we introduce Joint Distribution Matching Distillation (JDMD). By using real-world data distributions as a regularizing guide, JDMD effectively overcomes the fidelity degradation typically caused by over-reliance on synthetic data. Extensive experiments demonstrate that INSPATIO-WORLD significantly outperforms existing state-of-the-art (SOTA) models in spatial consistency and interaction precision, ranking first among real-time interactive methods on the WorldScore-Dynamic benchmark, and establishing a practical pipeline for navigating 4D environments reconstructed from monocular videos. △ Less

Submitted 13 April, 2026; v1 submitted 8 April, 2026; originally announced April 2026.

MoBiE: Efficient Inference of Mixture of Binary Experts under Post-Training Quantization

Authors: Zhixiong Zhao, Zukang Xu, Zhixuan Chen, Dawei Yang

Abstract: Mixture-of-Experts (MoE) based large language models (LLMs) offer strong performance but suffer from high memory and computation costs. Weight binarization provides extreme efficiency, yet existing binary methods designed for dense LLMs struggle with MoE-specific issues, including cross-expert redundancy, task-agnostic importance estimation, and quantization-induced routing shifts. To this end, we… ▽ More Mixture-of-Experts (MoE) based large language models (LLMs) offer strong performance but suffer from high memory and computation costs. Weight binarization provides extreme efficiency, yet existing binary methods designed for dense LLMs struggle with MoE-specific issues, including cross-expert redundancy, task-agnostic importance estimation, and quantization-induced routing shifts. To this end, we propose MoBiE, the first binarization framework tailored for MoE-based LLMs. MoBiE is built on three core innovations: 1. using joint SVD decomposition to reduce cross-expert redundancy; 2. integrating global loss gradients into local Hessian metrics to enhance weight importance estimation; 3. introducing an error constraint guided by the input null space to mitigate routing distortion. Notably, MoBiE achieves these optimizations while incurring no additional storage overhead, striking a balance between efficiency and model performance. Extensive experiments demonstrate that MoBiE consistently outperforms state-of-the-art binary methods across multiple MoE-based LLMs and benchmarks. For example, on Qwen3-30B-A3B, MoBiE reduces perplexity by 52.2$\%$, improves average zero-shot performance by 43.4$\%$, achieves over 2 $\times$ inference speedup, and further shortens quantization time. The code is available at https://github.com/Kishon-zzx/MoBiE. △ Less

Submitted 13 April, 2026; v1 submitted 8 April, 2026; originally announced April 2026.

Comments: Accepted at ACL 2026 Findings

URMF: Uncertainty-aware Robust Multimodal Fusion for Multimodal Sarcasm Detection

Authors: Zhenyu Wang, Weichen Cheng, Weijia Li, Junjie Mou, Zongyou Zhao, Guoying Zhang

Abstract: Multimodal sarcasm detection (MSD) aims to identify sarcastic intent from semantic incongruity between text and image. Although recent methods have improved MSD through cross-modal interaction and incongruity reasoning, they often assume that all modalities are equally reliable. In real-world social media, however, textual content may be ambiguous and visual content may be weakly relevant or even… ▽ More Multimodal sarcasm detection (MSD) aims to identify sarcastic intent from semantic incongruity between text and image. Although recent methods have improved MSD through cross-modal interaction and incongruity reasoning, they often assume that all modalities are equally reliable. In real-world social media, however, textual content may be ambiguous and visual content may be weakly relevant or even irrelevant, causing deterministic fusion to introduce noisy evidence and weaken robust reasoning. To address this issue, we propose Uncertainty-aware Robust Multimodal Fusion (URMF), a unified framework that explicitly models modality reliability during interaction and fusion. URMF first employs multi-head cross-attention to inject visual evidence into textual representations, followed by multi-head self-attention in the fused semantic space to enhance incongruity-aware reasoning. It then performs unified unimodal aleatoric uncertainty modeling over text, image, and interaction-aware latent representations by parameterizing each modality as a learnable Gaussian posterior. The estimated uncertainty is further used to dynamically regulate modality contributions during fusion, suppressing unreliable modalities and yielding a more robust joint representation. In addition, we design a joint training objective integrating task supervision, modality prior regularization, cross-modal distribution alignment, and uncertainty-driven self-sampling contrastive learning. Experiments on public MSD benchmarks show that URMF consistently outperforms strong unimodal, multimodal, and MLLM-based baselines, demonstrating the effectiveness of uncertainty-aware fusion for improving both accuracy and robustness. △ Less

Submitted 8 April, 2026; originally announced April 2026.

Robustness Risk of Conversational Retrieval: Identifying and Mitigating Noise Sensitivity in Qwen3-Embedding Model

Authors: Weishu Chen, Zhouhui Hou, Mingjie Zhan, Zhicheng Zhao, Fei Su

Abstract: We present an empirical study of embedding-based retrieval under realistic conversational settings, where queries are short, dialogue-like, and weakly specified, and retrieval corpora contain structured conversational artifacts. Focusing on Qwen3-embedding models, we identify a deployment-relevant robustness vulnerability: under conversational retrieval without query prompting, structured dialogue… ▽ More We present an empirical study of embedding-based retrieval under realistic conversational settings, where queries are short, dialogue-like, and weakly specified, and retrieval corpora contain structured conversational artifacts. Focusing on Qwen3-embedding models, we identify a deployment-relevant robustness vulnerability: under conversational retrieval without query prompting, structured dialogue-style noise can become disproportionately retrievable and intrude into top-ranked results, despite being semantically uninformative. This failure mode emerges consistently across model scales, remains largely invisible under standard clean-query benchmarks, and is significantly more pronounced in Qwen3 than in earlier Qwen variants and other widely used dense retrieval baselines. We further show that lightweight query prompting qualitatively alters retrieval behavior, effectively suppressing noise intrusion and restoring ranking stability. Our findings highlight an underexplored robustness risk in conversational retrieval and underscore the importance of evaluation protocols that reflect the complexities of deployed systems. △ Less

Submitted 3 February, 2026; originally announced April 2026.

Beyond Compromise: Pareto-Lenient Consensus for Efficient Multi-Preference LLM Alignment

Authors: Renxuan Tan, Rongpeng Li, Zhifeng Zhao, Honggang Zhang

Abstract: Transcending the single-preference paradigm, aligning LLMs with diverse human values is pivotal for robust deployment. Contemporary Multi-Objective Preference Alignment (MPA) approaches predominantly rely on static linear scalarization or rigid gradient projection to navigate these trade-offs. However, by enforcing strict conflict avoidance or simultaneous descent, these paradigms often prematurel… ▽ More Transcending the single-preference paradigm, aligning LLMs with diverse human values is pivotal for robust deployment. Contemporary Multi-Objective Preference Alignment (MPA) approaches predominantly rely on static linear scalarization or rigid gradient projection to navigate these trade-offs. However, by enforcing strict conflict avoidance or simultaneous descent, these paradigms often prematurely converge to local stationary points. While mathematically stable, these points represent a conservative compromise where the model sacrifices potential global Pareto improvements to avoid transient local trade-offs. To break this deadlock, we propose Pareto-Lenient Consensus (PLC), a game-theoretic framework that reimagines alignment as a dynamic negotiation process. Unlike rigid approaches, PLC introduces consensus-driven lenient gradient rectification, which dynamically tolerates local degradation provided there is a sufficient dominant coalition surplus, thereby empowering the optimization trajectory to escape local suboptimal equilibrium and explore the distal Pareto-optimal frontier. Theoretical analysis validates PLC can facilitate stalemate escape and asymptotically converge to a Pareto consensus equilibrium. Moreover, extensive experiments show that PLC surpasses baselines in both fixed-preference alignment and global Pareto frontier quality. This work highlights the potential of negotiation-driven alignment as a promising avenue for MPA. Our codes are available at https://anonymous.4open.science/r/aaa-6BB8. △ Less

Submitted 7 April, 2026; originally announced April 2026.

ResearchEVO: An End-to-End Framework for Automated Scientific Discovery and Documentation

Authors: Zhe Zhao, Haibin Wen, Jiaming Ma, Jiachang Zhan, Tianyi Xu, Ye Wei, Qingfu Zhang

Abstract: An important recurring pattern in scientific breakthroughs is a two-stage process: an initial phase of undirected experimentation that yields an unexpected finding, followed by a retrospective phase that explains why the finding works and situates it within existing theory. We present ResearchEVO, an end-to-end framework that computationally instantiates this discover-then-explain paradigm. The Ev… ▽ More An important recurring pattern in scientific breakthroughs is a two-stage process: an initial phase of undirected experimentation that yields an unexpected finding, followed by a retrospective phase that explains why the finding works and situates it within existing theory. We present ResearchEVO, an end-to-end framework that computationally instantiates this discover-then-explain paradigm. The Evolution Phase employs LLM-guided bi-dimensional co-evolution -- simultaneously optimizing both algorithmic logic and overall architecture -- to search the space of code implementations purely by fitness, without requiring any understanding of the solutions it produces. The Writing Phase then takes the best-performing algorithm and autonomously generates a complete, publication-ready research paper through sentence-level retrieval-augmented generation with explicit anti-hallucination verification and automated experiment design. To our knowledge, ResearchEVO is the first system to cover this full pipeline end to end: no prior work jointly performs principled algorithm evolution and literature-grounded scientific documentation. We validate the framework on two cross-disciplinary scientific problems -- Quantum Error Correction using real Google quantum hardware data, and Physics-Informed Neural Networks -- where the Evolution Phase discovered human-interpretable algorithmic mechanisms that had not been previously proposed in the respective domain literatures. In both cases, the Writing Phase autonomously produced compilable LaTeX manuscripts that correctly grounded these blind discoveries in existing theory via RAG, with zero fabricated citations. △ Less

Submitted 7 April, 2026; originally announced April 2026.

UniCreative: Unifying Long-form Logic and Short-form Sparkle via Reference-Free Reinforcement Learning

Authors: Xiaolong Wei, Zerun Zhu, Simin Niu, Xingyu Zhang, Peiying Yu, Changxuan Xiao, Yuchen Li, Jicheng Yang, Zhejun Zhao, Chong Meng, Long Xia, Daiting Shi

Abstract: A fundamental challenge in creative writing lies in reconciling the inherent tension between maintaining global coherence in long-form narratives and preserving local expressiveness in short-form texts. While long-context generation necessitates explicit macroscopic planning, short-form creativity often demands spontaneous, constraint-free expression. Existing alignment paradigms, however, typical… ▽ More A fundamental challenge in creative writing lies in reconciling the inherent tension between maintaining global coherence in long-form narratives and preserving local expressiveness in short-form texts. While long-context generation necessitates explicit macroscopic planning, short-form creativity often demands spontaneous, constraint-free expression. Existing alignment paradigms, however, typically employ static reward signals and rely heavily on high-quality supervised data, which is costly and difficult to scale. To address this, we propose \textbf{UniCreative}, a unified reference-free reinforcement learning framework. We first introduce \textbf{AC-GenRM}, an adaptive constraint-aware reward model that dynamically synthesizes query-specific criteria to provide fine-grained preference judgments. Leveraging these signals, we propose \textbf{ACPO}, a policy optimization algorithm that aligns models with human preferences across both content quality and structural paradigms without supervised fine-tuning and ground-truth references. Empirical results demonstrate that AC-GenRM aligns closely with expert evaluations, while ACPO significantly enhances performance across diverse writing tasks. Crucially, our analysis reveals an emergent meta-cognitive ability: the model learns to autonomously differentiate between tasks requiring rigorous planning and those favoring direct generation, validating the effectiveness of our direct alignment approach. △ Less

Submitted 7 April, 2026; originally announced April 2026.

Comments: Accepted to Findings of ACL 2026

Synthetic Sandbox for Training Machine Learning Engineering Agents

Authors: Yuhang Zhou, Lizhu Zhang, Yifan Wu, Jiayi Liu, Xiangjun Fan, Zhuokai Zhao, Hong Yan

Abstract: As large language model agents advance beyond software engineering (SWE) tasks toward machine learning engineering (MLE), verifying agent behavior becomes orders of magnitude more expensive: while SWE tasks can be verified via fast-executing unit tests, MLE verification requires running full ML pipelines -- data preprocessing, model training, and metric evaluation -- on large datasets at each roll… ▽ More As large language model agents advance beyond software engineering (SWE) tasks toward machine learning engineering (MLE), verifying agent behavior becomes orders of magnitude more expensive: while SWE tasks can be verified via fast-executing unit tests, MLE verification requires running full ML pipelines -- data preprocessing, model training, and metric evaluation -- on large datasets at each rollout step, rendering trajectory-wise on-policy reinforcement learning (RL) prohibitively slow. Existing approaches retreat to supervised fine-tuning (SFT) or offline proxy rewards, sacrificing the exploration and generalization benefits of on-policy RL. We observe that sandbox data size is the primary source of this bottleneck. Based on this insight, we introduce SandMLE, a multi-agent framework that generates diverse, verifiable synthetic MLE environments from a small number of seed tasks, preserving the structural and technical complexity of real-world problems while constraining datasets to micro-scale (each task is paired with only 50-200 training samples). Through extensive experiments, we show that SandMLE reduces execution time by over 13 times, enabling large-scale, on-policy trajectory-wise RL for the first time in the MLE domain. On MLE-bench-lite, SandMLE yields significant gains over SFT baselines across Qwen3-8B, 14B, and 30B-A3B, with relative medal rate improvements ranging from 20.3% to 66.9%. Furthermore, the trained policy generalizes across unseen agentic scaffolds, achieving up to 32.4% better HumanRank score on MLE-Dojo. △ Less

Submitted 6 April, 2026; originally announced April 2026.

Comments: 28 pages, 9 tables, 8 figures

MinerU2.5-Pro: Pushing the Limits of Data-Centric Document Parsing at Scale

Authors: Bin Wang, Tianyao He, Linke Ouyang, Fan Wu, Zhiyuan Zhao, Tao Chu, Yuan Qu, Zhenjiang Jin, Weijun Zeng, Ziyang Miao, Bangrui Xu, Junbo Niu, Mengzhang Cai, Jiantao Qiu, Qintong Zhang, Dongsheng Ma, Yuefeng Sun, Hejun Dong, Wenzheng Zhang, Jutao Xiao, Jiayong Shi, Pengyu Liao, Xiaomeng Zhao, Huaping Zhong, Liqun Wei , et al. (18 additional authors not shown)

Abstract: Current document parsing methods advance primarily through model architecture innovation, while systematic engineering of training data remains underexplored. Yet state-of-the-art models spanning diverse architectures and parameter scales exhibit highly consistent failure patterns on the same set of hard samples, suggesting that the performance bottleneck stems from shared deficiencies in training… ▽ More Current document parsing methods advance primarily through model architecture innovation, while systematic engineering of training data remains underexplored. Yet state-of-the-art models spanning diverse architectures and parameter scales exhibit highly consistent failure patterns on the same set of hard samples, suggesting that the performance bottleneck stems from shared deficiencies in training data rather than from architectural differences. Building on this finding, we present MinerU2.5-Pro, which advances the state of the art purely through data engineering and training strategy design while retaining the 1.2B-parameter architecture of MinerU2.5 unchanged. At its core is a Data Engine co-designed around coverage, informativeness, and annotation accuracy: Diversity-and-Difficulty-Aware Sampling expands training data from under 10M to 65.5M samples while mitigating distribution shift; Cross-Model Consistency Verification leverages output consensus among heterogeneous models to assess sample difficulty and generate reliable annotations; the Judge-and-Refine pipeline improves annotation quality for hard samples through render-then-verify iterative correction. A three-stage progressive training strategy--large-scale pre-training, hard sample fine-tuning, and GRPO alignment--sequentially exploits these data at different quality tiers. On the evaluation front, we rectify element-matching biases in OmniDocBench v1.5 and introduce a Hard subset, establishing the more discriminative OmniDocBench v1.6 protocol. Without any architectural modification, MinerU2.5-Pro achieves 95.69 on OmniDocBench v1.6, improving over the same-architecture baseline by 2.71 points and surpassing all existing methods, including those based on models with over 200x more parameters. △ Less

Submitted 9 April, 2026; v1 submitted 6 April, 2026; originally announced April 2026.

Comments: Technical Report

Gradual Cognitive Externalization: From Modeling Cognition to Constituting It

Authors: Zhimin Zhao

Abstract: Developers are publishing AI agent skills that replicate a colleague's communication style, encode a supervisor's mentoring heuristics, or preserve a person's behavioral repertoire beyond biological death. To explain why, we propose Gradual Cognitive Externalization (GCE), a framework arguing that ambient AI systems, through sustained causal coupling with users, transition from modeling cognitive… ▽ More Developers are publishing AI agent skills that replicate a colleague's communication style, encode a supervisor's mentoring heuristics, or preserve a person's behavioral repertoire beyond biological death. To explain why, we propose Gradual Cognitive Externalization (GCE), a framework arguing that ambient AI systems, through sustained causal coupling with users, transition from modeling cognitive functions to constituting part of users' cognitive architectures. GCE adopts an explicit functionalist commitment: cognitive functions are individuated by their causal-functional roles, not by substrate. The framework rests on the behavioral manifold hypothesis and a central falsifiable assumption, the no behaviorally invisible residual (NBIR) hypothesis: for any cognitive function whose behavioral output lies on a learnable manifold, no behaviorally invisible component is necessary for that function's operation. We document evidence from deployed AI systems showing that externalization preconditions are already observable, formalize three criteria separating cognitive integration from tool use (bidirectional adaptation, functional equivalence, causal coupling), and derive five testable predictions with theory-constrained thresholds. △ Less

Submitted 6 April, 2026; v1 submitted 5 April, 2026; originally announced April 2026.

Avoiding Non-Integrable Beliefs in Expectation Propagation

Authors: Zilu Zhao, Jichao Chen, Dirk Slock

Abstract: Expectation Propagation (EP) is a widely used iterative message-passing algorithm that decomposes a global inference problem into multiple local ones. It approximates marginal distributions as ``beliefs'' using intermediate functions called ``messages''. It has been shown that the stationary points of EP are the same as corresponding constrained Bethe Free Energy (BFE) optimization problem. Theref… ▽ More Expectation Propagation (EP) is a widely used iterative message-passing algorithm that decomposes a global inference problem into multiple local ones. It approximates marginal distributions as ``beliefs'' using intermediate functions called ``messages''. It has been shown that the stationary points of EP are the same as corresponding constrained Bethe Free Energy (BFE) optimization problem. Therefore, EP is an iterative method of optimizing the constrained BFE. However, the iterative method may fall out of the feasible set of the BFE optimization problem, i.e., the beliefs are not integrable. In most literature, the authors use various methods to keep all the messages integrable. In most Bayesian estimation problems, limiting the messages to be integrable shrinks the actual feasible set. Furthermore, in extreme cases where the factors are not integrable, making the message itself integrable is not enough to have integrable beliefs. In this paper, two EP frameworks are proposed to ensure that EP has integrable beliefs. Both of the methods allows non-integrable messages. We then investigate the signal recovery problem in Generalized Linear Model (GLM) using our proposed methods. △ Less

Submitted 5 April, 2026; originally announced April 2026.

Position: Logical Soundness is not a Reliable Criterion for Neurosymbolic Fact-Checking with LLMs

Authors: Jason Chan, Robert Gaizauskas, Zhixue Zhao

Abstract: As large language models (LLMs) are increasing integrated into fact-checking pipelines, formal logic is often proposed as a rigorous means by which to mitigate bias, errors and hallucinations in these models' outputs. For example, some neurosymbolic systems verify claims by using LLMs to translate natural language into logical formulae and then checking whether the proposed claims are logically so… ▽ More As large language models (LLMs) are increasing integrated into fact-checking pipelines, formal logic is often proposed as a rigorous means by which to mitigate bias, errors and hallucinations in these models' outputs. For example, some neurosymbolic systems verify claims by using LLMs to translate natural language into logical formulae and then checking whether the proposed claims are logically sound, i.e. whether they can be validly derived from premises that are verified to be true. We argue that such approaches structurally fail to detect misleading claims due to systematic divergences between conclusions that are logically sound and inferences that humans typically make and accept. Drawing on studies in cognitive science and pragmatics, we present a typology of cases in which logically sound conclusions systematically elicit human inferences that are unsupported by the underlying premises. Consequently, we advocate for a complementary approach: leveraging the human-like reasoning tendencies of LLMs as a feature rather than a bug, and using these models to validate the outputs of formal components in neurosymbolic systems against potentially misleading conclusions. △ Less

Submitted 5 April, 2026; originally announced April 2026.

Comments: Preprint

NTIRE 2026 3D Restoration and Reconstruction in Real-world Adverse Conditions: RealX3D Challenge Results

Authors: Shuhong Liu, Chenyu Bao, Ziteng Cui, Xuangeng Chu, Bin Ren, Lin Gu, Xiang Chen, Mingrui Li, Long Ma, Marcos V. Conde, Radu Timofte, Yun Liu, Ryo Umagami, Tomohiro Hashimoto, Zijian Hu, Yuan Gan, Tianhan Xu, Yusuke Kurose, Tatsuya Harada, Junwei Yuan, Gengjia Chang, Xining Ge, Mache You, Qida Cao, Zeliang Li , et al. (81 additional authors not shown)

Abstract: This paper presents a comprehensive review of the NTIRE 2026 3D Restoration and Reconstruction (3DRR) Challenge, detailing the proposed methods and results. The challenge seeks to identify robust reconstruction pipelines that are robust under real-world adverse conditions, specifically extreme low-light and smoke-degraded environments, as captured by our RealX3D benchmark. A total of 279 participa… ▽ More This paper presents a comprehensive review of the NTIRE 2026 3D Restoration and Reconstruction (3DRR) Challenge, detailing the proposed methods and results. The challenge seeks to identify robust reconstruction pipelines that are robust under real-world adverse conditions, specifically extreme low-light and smoke-degraded environments, as captured by our RealX3D benchmark. A total of 279 participants registered for the competition, of whom 33 teams submitted valid results. We thoroughly evaluate the submitted approaches against state-of-the-art baselines, revealing significant progress in 3D reconstruction under adverse conditions. Our analysis highlights shared design principles among top-performing methods and provides insights into effective strategies for handling 3D scene degradation. △ Less

Submitted 5 April, 2026; originally announced April 2026.

MultiPress: A Multi-Agent Framework for Interpretable Multimodal News Classification

Authors: Tailong Luo, Hao Li, Rong Fu, Xinyue Jiang, Huaxuan Ding, Yiduo Zhang, Zilin Zhao, Simon Fong, Guangyin Jin, Jianyuan Ni

Abstract: With the growing prevalence of multimodal news content, effective news topic classification demands models capable of jointly understanding and reasoning over heterogeneous data such as text and images. Existing methods often process modalities independently or employ simplistic fusion strategies, limiting their ability to capture complex cross-modal interactions and leverage external knowledge. T… ▽ More With the growing prevalence of multimodal news content, effective news topic classification demands models capable of jointly understanding and reasoning over heterogeneous data such as text and images. Existing methods often process modalities independently or employ simplistic fusion strategies, limiting their ability to capture complex cross-modal interactions and leverage external knowledge. To overcome these limitations, we propose MultiPress, a novel three-stage multi-agent framework for multimodal news classification. MultiPress integrates specialized agents for multimodal perception, retrieval-augmented reasoning, and gated fusion scoring, followed by a reward-driven iterative optimization mechanism. We validate MultiPress on a newly constructed large-scale multimodal news dataset, demonstrating significant improvements over strong baselines and highlighting the effectiveness of modular multi-agent collaboration and retrieval-augmented reasoning in enhancing classification accuracy and interpretability. △ Less

Submitted 4 April, 2026; originally announced April 2026.

Comments: Accepted in International Joint Conference on Neural Networks (IJCNN) 2026

CharTool: Tool-Integrated Visual Reasoning for Chart Understanding

Authors: Situo Zhang, Yifan Zhang, Zichen Zhu, Da Ma, Lei Pan, Danyang Zhang, Zihan Zhao, Lu Chen, Kai Yu

Abstract: Charts are ubiquitous in scientific and financial literature for presenting structured data. However, chart reasoning remains challenging for multimodal large language models (MLLMs) due to the lack of high-quality training data, as well as the need for fine-grained visual grounding and precise numerical computation. To address these challenges, we first propose DuoChart, a scalable dual-source da… ▽ More Charts are ubiquitous in scientific and financial literature for presenting structured data. However, chart reasoning remains challenging for multimodal large language models (MLLMs) due to the lack of high-quality training data, as well as the need for fine-grained visual grounding and precise numerical computation. To address these challenges, we first propose DuoChart, a scalable dual-source data pipeline that combines synthesized charts with real-world charts to construct diverse, high-quality chart training data. We then introduce CharTool, which equips MLLMs with external tools, including image cropping for localized visual perception and code-based computation for accurate numerical reasoning. Through agentic reinforcement learning on DuoChart, CharTool learns tool-integrated reasoning grounded in chart content. Extensive experiments on six chart benchmarks show that our method consistently improves over strong MLLM baselines across model scales. Notably, CharTool-7B outperforms the base model by **+8.0%** on CharXiv (Reasoning) and **+9.78%** on ChartQAPro, while achieving competitive performance with substantially larger or proprietary models. Moreover, CharTool demonstrates positive generalization to out-of-domain visual math reasoning benchmarks. △ Less

Submitted 3 April, 2026; originally announced April 2026.

Differentiable Stroke Planning with Dual Parameterization for Efficient and High-Fidelity Painting Creation

Authors: Jinfan Liu, Wuze Zhang, Zhangli Hu, Zhehan Zhao, Ye Chen, Bingbing Ni

Abstract: In stroke-based rendering, search methods often get trapped in local minima due to discrete stroke placement, while differentiable optimizers lack structural awareness and produce unstructured layouts. To bridge this gap, we propose a dual representation that couples discrete polylines with continuous Bézier control points via a bidirectional mapping mechanism. This enables collaborative optimizat… ▽ More In stroke-based rendering, search methods often get trapped in local minima due to discrete stroke placement, while differentiable optimizers lack structural awareness and produce unstructured layouts. To bridge this gap, we propose a dual representation that couples discrete polylines with continuous Bézier control points via a bidirectional mapping mechanism. This enables collaborative optimization: local gradients refine global stroke structures, while content-aware stroke proposals help escape poor local optima. Our representation further supports Gaussian-splatting-inspired initialization, enabling highly parallel stroke optimization across the image. Experiments show that our approach reduces the number of strokes by 30-50%, achieves more structurally coherent layouts, and improves reconstruction quality, while cutting optimization time by 30-40% compared to existing differentiable vectorization methods. △ Less

Submitted 3 April, 2026; originally announced April 2026.

Agentic AI-Empowered Wireless Agent Networks With Semantic-Aware Collaboration via ILAC

Authors: Zhouxiang Zhao, Jiaxiang Wang, Zhaohui Yang, Kun Yang, Zhaoyang Zhang, Mingzhe Chen, Kaibin Huang

Abstract: The rapid development of agentic artificial intelligence (AI) is driving future wireless networks to evolve from passive data pipes into intelligent collaborative ecosystems under the emerging paradigm of integrated learning and communication (ILAC). However, realizing efficient agentic collaboration faces challenges not only in handling semantic redundancy but also in the lack of an integrated me… ▽ More The rapid development of agentic artificial intelligence (AI) is driving future wireless networks to evolve from passive data pipes into intelligent collaborative ecosystems under the emerging paradigm of integrated learning and communication (ILAC). However, realizing efficient agentic collaboration faces challenges not only in handling semantic redundancy but also in the lack of an integrated mechanism for communication, computation, and control. To address this, we propose a wireless agent network (WAN) framework that orchestrates a progressive knowledge aggregation mechanism. Specifically, we formulate the aggregation process as a joint energy minimization problem where the agents perform semantic compression to eliminate redundancy, optimize transmission power to deliver semantic payloads, and adjust physical trajectories to proactively enhance channel qualities. To solve this problem, we develop a hierarchical algorithm that integrates inner-level resource optimization with outer-level topology evolution. Theoretically, we reveal that incorporating a potential field into the topology evolution effectively overcomes the short-sightedness of greedy matching, providing a mathematically rigorous heuristic for long-term energy minimization. Simulation results demonstrate that the proposed framework achieves superior energy efficiency and scalability compared to conventional benchmarks, validating the efficacy of semantic-aware collaboration in dynamic environments. △ Less

Submitted 1 April, 2026; originally announced April 2026.

A Practical Two-Stage Framework for GPU Resource and Power Prediction in Heterogeneous HPC Systems

Authors: Beste Oztop, Dhruva Kulkarni, Zhengji Zhao, Ayse Kivilcim Coskun, Kadidia Konate

Abstract: Efficient utilization of GPU resources and power has become critical with the growing demand for GPUs in high-performance computing (HPC). In this paper, we analyze GPU utilization and GPU memory utilization, as well as the power consumption of the Vienna ab initio Simulation Package (VASP), using the Slurm workload manager historical logs and GPU performance metrics collected by NVIDIA's Data Cen… ▽ More Efficient utilization of GPU resources and power has become critical with the growing demand for GPUs in high-performance computing (HPC). In this paper, we analyze GPU utilization and GPU memory utilization, as well as the power consumption of the Vienna ab initio Simulation Package (VASP), using the Slurm workload manager historical logs and GPU performance metrics collected by NVIDIA's Data Center GPU Manager (DCGM). VASP is a widely used materials science application on Perlmutter at NERSC, an HPE Cray EX system based on NVIDIA A100 GPUs. Using our insights from the resource utilization analysis of VASP applications, we propose a resource prediction framework to predict the average GPU power, maximum GPU utilization, and maximum GPU memory utilization values of heterogeneous HPC system applications to enable more efficient scheduling decisions and power-aware system operation. Our prediction framework consists of two stages: 1) using only the Slurm accounting logs as training data and 2) augmenting the training data with historical GPU profiling metrics collected with DCGM. The maximum GPU utilization predictions using only the Slurm submission features achieve up to 97% accuracy. Furthermore, features engineered from GPU-compute and memory activity metrics exhibit good correlations with average power utilization, and our runtime power usage prediction experiments result in up to 92% prediction accuracy. These findings demonstrate the effectiveness of DCGM metrics in capturing application characteristics and highlight their potential for developing predictive models to support dynamic power management in HPC systems. △ Less

Submitted 2 April, 2026; originally announced April 2026.

Comments: 9 pages, 6 figures

From Component Manipulation to System Compromise: Understanding and Detecting Malicious MCP Servers

Authors: Yiheng Huang, Zhijia Zhao, Bihuan Chen, Susheng Wu, Zhuotong Zhou, Yiheng Cao, Xin Hu, Xin Peng

Abstract: The model context protocol (MCP) standardizes how LLMs connect to external tools and data sources, enabling faster integration but introducing new attack vectors. Despite the growing adoption of MCP, existing MCP security studies classify attacks by their observable effects, obscuring how attacks behave across different MCP server components and overlooking multi-component attack chains. Meanwhile… ▽ More The model context protocol (MCP) standardizes how LLMs connect to external tools and data sources, enabling faster integration but introducing new attack vectors. Despite the growing adoption of MCP, existing MCP security studies classify attacks by their observable effects, obscuring how attacks behave across different MCP server components and overlooking multi-component attack chains. Meanwhile, existing defenses are less effective when facing multi-component attacks or previously unknown malicious behaviors. This work presents a component-centric perspective for understanding and detecting malicious MCP servers. First, we build the first component-centric PoC dataset of 114 malicious MCP servers where attacks are achieved as manipulation over MCP components and their compositions. We evaluate these attacks' effectiveness across two MCP hosts and five LLMs, and uncover that (1) component position shapes attack success rate; and (2) multi-component compositions often outperform single-component attacks by distributing malicious logic. Second, we propose and implement Connor, a two-stage behavioral deviation detector for malicious MCP servers. It first performs pre-execution analysis to detect malicious shell commands and extract each tool's function intent, and then conducts step-wise in-execution analysis to trace each tool's behavioral trajectories and detect deviations from its function intent. Evaluation on our curated dataset indicates that Connor achieves an F1-score of 94.6%, outperforming the state of the art by 8.9% to 59.6%. In real-world detection, Connor identifies two malicious servers. △ Less

Submitted 2 April, 2026; originally announced April 2026.

GPA: Learning GUI Process Automation from Demonstrations

Authors: Zirui Zhao, Jun Hao Liew, Yan Yang, Wenzhuo Yang, Ziyang Luo, Doyen Sahoo, Silvio Savarese, Junnan Li

Abstract: GUI Process Automation (GPA) is a lightweight but general vision-based Robotic Process Automation (RPA), which enables fast and stable process replay with only a single demo. Addressing the fragility of traditional RPA and the non-deterministic risks of current vision language model-based GUI agents, GPA introduces three core benefits: (1) Robustness via Sequential Monte Carlo-based localization t… ▽ More GUI Process Automation (GPA) is a lightweight but general vision-based Robotic Process Automation (RPA), which enables fast and stable process replay with only a single demo. Addressing the fragility of traditional RPA and the non-deterministic risks of current vision language model-based GUI agents, GPA introduces three core benefits: (1) Robustness via Sequential Monte Carlo-based localization to handle rescaling and detection uncertainty; (2) Deterministic and Reliability safeguarded by readiness calibration; and (3) Privacy through fast, fully local execution. This approach delivers the adaptability, robustness, and security required for enterprise workflows. It can also be used as an MCP/CLI tool by other agents with coding capabilities so that the agent only reasons and orchestrates while GPA handles the GUI execution. We conducted a pilot experiment to compare GPA with Gemini 3 Pro (with CUA tools) and found that GPA achieves higher success rate with 10 times faster execution speed in finishing long-horizon GUI tasks. △ Less

Submitted 3 April, 2026; v1 submitted 2 April, 2026; originally announced April 2026.

UniRecGen: Unifying Multi-View 3D Reconstruction and Generation

Authors: Zhisheng Huang, Jiahao Chen, Cheng Lin, Chenyu Hu, Hanzhuo Huang, Zhengming Yu, Mengfei Li, Yuheng Liu, Zekai Gu, Zibo Zhao, Yuan Liu, Xin Li, Wenping Wang

Abstract: Sparse-view 3D modeling represents a fundamental tension between reconstruction fidelity and generative plausibility. While feed-forward reconstruction excels in efficiency and input alignment, it often lacks the global priors needed for structural completeness. Conversely, diffusion-based generation provides rich geometric details but struggles with multi-view consistency. We present UniRecGen, a… ▽ More Sparse-view 3D modeling represents a fundamental tension between reconstruction fidelity and generative plausibility. While feed-forward reconstruction excels in efficiency and input alignment, it often lacks the global priors needed for structural completeness. Conversely, diffusion-based generation provides rich geometric details but struggles with multi-view consistency. We present UniRecGen, a unified framework that integrates these two paradigms into a single cooperative system. To overcome inherent conflicts in coordinate spaces, 3D representations, and training objectives, we align both models within a shared canonical space. We employ disentangled cooperative learning, which maintains stable training while enabling seamless collaboration during inference. Specifically, the reconstruction module is adapted to provide canonical geometric anchors, while the diffusion generator leverages latent-augmented conditioning to refine and complete the geometric structure. Experimental results demonstrate that UniRecGen achieves superior fidelity and robustness, outperforming existing methods in creating complete and consistent 3D models from sparse observations. Code is available at https://github.com/zsh523/UniRecGen. △ Less

Submitted 2 April, 2026; v1 submitted 1 April, 2026; originally announced April 2026.

Paper Reconstruction Evaluation: Evaluating Presentation and Hallucination in AI-written Papers

Authors: Atsuyuki Miyai, Mashiro Toyooka, Zaiying Zhao, Kenta Watanabe, Toshihiko Yamasaki, Kiyoharu Aizawa

Abstract: This paper introduces the first systematic evaluation framework for quantifying the quality and risks of papers written by modern coding agents. While AI-driven paper writing has become a growing concern, rigorous evaluation of the quality and potential risks of AI-written papers remains limited, and a unified understanding of their reliability is still lacking. We introduce Paper Reconstruction E… ▽ More This paper introduces the first systematic evaluation framework for quantifying the quality and risks of papers written by modern coding agents. While AI-driven paper writing has become a growing concern, rigorous evaluation of the quality and potential risks of AI-written papers remains limited, and a unified understanding of their reliability is still lacking. We introduce Paper Reconstruction Evaluation (PaperRecon), an evaluation framework in which an overview (overview.md) is created from an existing paper, after which an agent generates a full paper based on the overview and minimal additional resources, and the result is subsequently compared against the original paper. PaperRecon disentangles the evaluation of the AI-written papers into two orthogonal dimensions, Presentation and Hallucination, where Presentation is evaluated using a rubric and Hallucination is assessed via agentic evaluation grounded in the original paper source. For evaluation, we introduce PaperWrite-Bench, a benchmark of 51 papers from top-tier venues across diverse domains published after 2025. Our experiments reveal a clear trade-off: while both ClaudeCode and Codex improve with model advances, ClaudeCode achieves higher presentation quality at the cost of more than 10 hallucinations per paper on average, whereas Codex produces fewer hallucinations but lower presentation quality. This work takes a first step toward establishing evaluation frameworks for AI-driven paper writing and improving the understanding of its risks within the research community. △ Less

Submitted 1 April, 2026; originally announced April 2026.

Comments: Project Page: https://agent4science-utokyo.github.io/PaperRecon_HP/

From Early Encoding to Late Suppression: Interpreting LLMs on Character Counting Tasks

Authors: Ayan Datta, Mounika Marreddy, Alexander Mehler, Zhixue Zhao, Radhika Mamidi

Abstract: Large language models (LLMs) exhibit failures on elementary symbolic tasks such as character counting in a word, despite excelling on complex benchmarks. Although this limitation has been noted, the internal reasons remain unclear. We use character counting (e.g., "How many p's are in apple?") as a minimal, controlled probe that isolates token-level reasoning from higher-level confounds. Using thi… ▽ More Large language models (LLMs) exhibit failures on elementary symbolic tasks such as character counting in a word, despite excelling on complex benchmarks. Although this limitation has been noted, the internal reasons remain unclear. We use character counting (e.g., "How many p's are in apple?") as a minimal, controlled probe that isolates token-level reasoning from higher-level confounds. Using this setting, we uncover a consistent phenomenon across modern architectures, including LLaMA, Qwen, and Gemma: models often compute the correct answer internally yet fail to express it at the output layer. Through mechanistic analysis combining probing classifiers, activation patching, logit lens analysis, and attention head tracing, we show that character-level information is encoded in early and mid-layer representations. However, this information is attenuated by a small set of components in later layers, especially the penultimate and final layer MLP. We identify these components as negative circuits: subnetworks that downweight correct signals in favor of higher-probability but incorrect outputs. Our results lead to two contributions. First, we show that symbolic reasoning failures in LLMs are not due to missing representations or insufficient scale, but arise from structured interference within the model's computation graph. This explains why such errors persist and can worsen under scaling and instruction tuning. Second, we provide evidence that LLM forward passes implement a form of competitive decoding, in which correct and incorrect hypotheses coexist and are dynamically reweighted, with final outputs determined by suppression as much as by amplification. These findings carry implications for interpretability and robustness: simple symbolic reasoning exposes weaknesses in modern LLMs, underscoring need for design strategies that ensure information is encoded and reliably used. △ Less

Submitted 1 April, 2026; originally announced April 2026.

Gloria: Consistent Character Video Generation via Content Anchors

Authors: Yuhang Yang, Fan Zhang, Huaijin Pi, Shuai Guo, Guowei Xu, Wei Zhai, Yang Cao, Zheng-Jun Zha

Abstract: Digital characters are central to modern media, yet generating character videos with long-duration, consistent multi-view appearance and expressive identity remains challenging. Existing approaches either provide insufficient context to preserve identity or leverage non-character-centric information as the memory, leading to suboptimal consistency. Recognizing that character video generation inher… ▽ More Digital characters are central to modern media, yet generating character videos with long-duration, consistent multi-view appearance and expressive identity remains challenging. Existing approaches either provide insufficient context to preserve identity or leverage non-character-centric information as the memory, leading to suboptimal consistency. Recognizing that character video generation inherently resembles an outside-looking-in scenario. In this work, we propose representing the character visual attributes through a compact set of anchor frames. This design provides stable references for consistency, while reference-based video generation inherently faces challenges of copy-pasting and multi-reference conflicts. To address these, we introduce two mechanisms: Superset Content Anchoring, providing intra- and extra-training clip cues to prevent duplication, and RoPE as Weak Condition, encoding positional offsets to distinguish multiple anchors. Furthermore, we construct a scalable pipeline to extract these anchors from massive videos. Experiments show our method generates high-quality character videos exceeding 10 minutes, and achieves expressive identity and appearance consistency across views, surpassing existing methods. △ Less

Submitted 31 March, 2026; originally announced March 2026.

Comments: Accepted by CVPR2026 Main, project: https://yyvhang.github.io/Gloria_Page/

An Empirical Study of Multi-Agent Collaboration for Automated Research

Authors: Yang Shen, Zhenyi Yi, Ziyi Zhao, Lijun Sun, Dongyang Li, Chin-Teng Lin, Yuhui Shi

Abstract: As AI agents evolve, the community is rapidly shifting from single Large Language Models (LLMs) to Multi-Agent Systems (MAS) to overcome cognitive bottlenecks in automated research. However, the optimal multi-agent coordination framework for these autonomous agents remains largely unexplored. In this paper, we present a systematic empirical study investigating the comparative efficacy of distinct… ▽ More As AI agents evolve, the community is rapidly shifting from single Large Language Models (LLMs) to Multi-Agent Systems (MAS) to overcome cognitive bottlenecks in automated research. However, the optimal multi-agent coordination framework for these autonomous agents remains largely unexplored. In this paper, we present a systematic empirical study investigating the comparative efficacy of distinct multi-agent structures for automated machine learning optimization. Utilizing a rigorously controlled, execution-based testbed equipped with Git worktree isolation and explicit global memory, we benchmark a single-agent baseline against two multi-agent paradigms: a subagent architecture (parallel exploration with post-hoc consolidation) and an agent team architecture (experts with pre-execution handoffs). By evaluating these systems under strictly fixed computational time budgets, our findings reveal a fundamental trade-off between operational stability and theoretical deliberation. The subagent mode functions as a highly resilient, high-throughput search engine optimal for broad, shallow optimizations under strict time constraints. Conversely, the agent team topology exhibits higher operational fragility due to multi-author code generation but achieves the deep theoretical alignment necessary for complex architectural refactoring given extended compute budgets. These empirical insights provide actionable guidelines for designing future autoresearch systems, advocating for dynamically routed architectures that adapt their collaborative structures to real-time task complexity. △ Less

Submitted 31 March, 2026; originally announced March 2026.

Kilohertz-Safe: A Scalable Framework for Constrained Dexterous Retargeting

Authors: Yinxiao Tian, Ziyi Yang, Zinan Zhao, Zhen Kan

Abstract: Dexterous hand teleoperation requires motion re-targeting methods that simultaneously achieve high-frequency real-time performance and enforcement of heterogeneous kinematic and safety constraints. Existing nonlinear optimization-based approaches often incur prohibitive computational cost, limiting their applicability to kilohertz-level control, while learning-based methods typically lack formal s… ▽ More Dexterous hand teleoperation requires motion re-targeting methods that simultaneously achieve high-frequency real-time performance and enforcement of heterogeneous kinematic and safety constraints. Existing nonlinear optimization-based approaches often incur prohibitive computational cost, limiting their applicability to kilohertz-level control, while learning-based methods typically lack formal safety guarantees. This paper proposes a scalable motion retargeting framework that reformulates the nonlinear retargeting problem into a convex quadratic program in joint differential space. Heterogeneous constraints, including kinematic limits and collision avoidance, are incorporated through systematic linearization, resulting in improved computational efficiency and numerical stability. Control barrier functions are further integrated to provide formal safety guarantees during the retargeting process. The proposed framework is validated through simulations and hardware experiments on the Wuji Hand platform, outperforming state-of-the-art methods such as Dex-Retargeting and GeoRT. The framework achieves high-frequency operation with an average latency of 9.05 ms, while over 95% of retargeted frames satisfy the safety criteria, effectively mitigating self-collision and penetration during complex manipulation tasks. △ Less

Submitted 30 March, 2026; originally announced March 2026.

Comments: 8 pages,6 Figures,Under Reiview

Chat-Scene++: Exploiting Context-Rich Object Identification for 3D LLM

Authors: Haifeng Huang, Yilun Chen, Zehan Wang, Jiangmiao Pang, Zhou Zhao

Abstract: Recent advancements in multi-modal large language models (MLLMs) have shown strong potential for 3D scene understanding. However, existing methods struggle with fine-grained object grounding and contextual reasoning, limiting their ability to interpret and interact with complex 3D environments. In this paper, we present Chat-Scene++, an MLLM framework that represents 3D scenes as context-rich obje… ▽ More Recent advancements in multi-modal large language models (MLLMs) have shown strong potential for 3D scene understanding. However, existing methods struggle with fine-grained object grounding and contextual reasoning, limiting their ability to interpret and interact with complex 3D environments. In this paper, we present Chat-Scene++, an MLLM framework that represents 3D scenes as context-rich object sequences. By structuring scenes as sequences of objects with contextual semantics, Chat-Scene++ enables object-centric representation and interaction. It decomposes a 3D scene into object representations paired with identifier tokens, allowing LLMs to follow instructions across diverse 3D vision-language tasks. To capture inter-object relationships and global semantics, Chat-Scene++ extracts context-rich object features using large-scale pre-trained 3D scene-level and 2D image-level encoders, unlike the isolated per-object features in Chat-Scene. Its flexible object-centric design also supports grounded chain-of-thought (G-CoT) reasoning, enabling the model to distinguish objects at both category and spatial levels during multi-step inference. Without the need for additional task-specific heads or fine-tuning, Chat-Scene++ achieves state-of-the-art performance on five major 3D vision-language benchmarks: ScanRefer, Multi3DRefer, Scan2Cap, ScanQA, and SQA3D. These results highlight its effectiveness in scene comprehension, object grounding, and spatial reasoning. Additionally, without reconstructing 3D worlds through computationally expensive processes, we demonstrate its applicability to real-world scenarios using only 2D inputs. △ Less

Submitted 29 March, 2026; originally announced March 2026.