arXiv subjects

Kaishen Wang

Publications and source records attributed to Kaishen Wang.

14 recordsLinked to original sources

arXiv2026

VideoResearcher: Self-Improving Tool Design for Long-Video Understanding

Video agents have made substantial progress in long-video understanding. Yet effective video-agent systems require costly, time-consuming manual design and trial and error. Current self-improvement methods either refine low-impact prompts, recombine predefined micro-tools, or struggle with convergence in harness optimization. To bridge this gap, we target high-impact video-tool with VideoResearcher, a training-free multi-agent framework that autonomously designs, tests, and refines tools for video understanding, like a human researcher. VideoResearcher operates through dual Solving and Evolving loops: it analyzes tool-use trajectories to identify capability gaps, coordinates specialized agents to develop and validate executable tools, and reuses evolved tools to strengthen evidence acquisition in subsequent video reasoning. Through iterative tool refinement and validation, it progressively strengthens evidence acquisition without updating model parameters. VideoResearcher achieves state-of-the-art performance among self-improving agents and approaches the human-designed upper bound, demonstrating a training-free paradigm for long-video understanding that expands agent capabilities through autonomous tool development while reducing costly manual engineering.

cs.CV↗

arXiv2026

Unsafe by Reciprocity: How Generation-Understanding Coupling Undermines Safety in Unified Multimodal Models

Recent advances in Large Language Models (LLMs) and Text-to-Image (T2I) models have led to the emergence of Unified Multimodal Models (UMMs), where multimodal understanding and image generation are tightly integrated within a shared architecture. Prior studies suggest that such reciprocity enhances cross-functionality performance through shared representations and joint optimization. However, the safety implications of this tight coupling remain largely unexplored, as existing safety research predominantly analyzes understanding and generation functionalities in isolation. In this work, we investigate whether cross-functionality reciprocity itself constitutes a structural source of vulnerability in UMMs. We propose RICE: Reciprocal Interaction-based Cross-functionality Exploitation, a novel attack paradigm that explicitly exploits bidirectional interactions between understanding and generation. Using this framework, we systematically evaluate Generation-to-Understanding (G-U) and Understanding-to-Generation (U-G) attack pathways, demonstrating that unsafe intermediate signals can propagate across modalities and amplify safety risks. Extensive experiments show high Attack Success Rates (ASR) in both directions, revealing previously overlooked safety weaknesses inherent to UMMs.

cs.CV↗

arXiv2026

Where to Look Matters: On-Policy Self-Distillation for Long-Video Understanding

Vision-language models (VLMs) have made substantial progress in long-video understanding, with standard backbone models typically answering questions from frames sampled across the full video. However, as videos become longer, the full-video context inevitably contains more question-irrelevant temporal content, which can distract the model from the evidence needed to answer a specific question. We empirically find that focusing the visual input on short annotated clue intervals containing question-relevant evidence consistently improves prediction accuracy across model scales compared with using the corresponding full videos, while requiring fewer input frames. Based on this finding, we introduce Clue-OPSD, a clue-privileged on-policy self-distillation framework for long-video understanding. During training, a full-video student learns from a self-teacher conditioned on the corresponding clue interval by aligning their next-token distributions along student-generated trajectories. Clue-OPSD thus uses clue intervals as privileged supervision without relying on ground-truth answer labels, while requiring no clue annotations or additional modules at inference time. Extensive experiments across multiple long-video understanding benchmarks and Qwen3.5 model scales demonstrate consistent improvements over the corresponding backbone models and strong performance against supervised post-training baselines.

cs.CV↗

arXiv2026

Mitigating Factual Hallucination in Large Reasoning Models via Mixed-Mode Advantage Regularization

Large reasoning models (LRMs) improve language model capabilities by generating explicit thinking traces before final answers. In factuality-oriented question answering (QA), such thinking often improves overall performance by helping the model recover relevant knowledge and refine its answers. However, we find that this benefit is not uniform at the instance level: explicit thinking can also overturn correct non-thinking answers and lead to factual drift. We refer to this failure mode as \emph{thinking-induced hallucination}. To explain this phenomenon, we formulate explicit thinking in factuality QA as a thinking residual over the model's direct-answer tendency, which can either recover missing knowledge or introduce unsupported associations. Based on this formulation, we propose MARGO, \underline{\textit{M}}ixed-Mode \underline{\textit{A}}dvantage \underline{\textit{R}}egularization for \underline{\textit{G}}rounded \underline{\textit{O}}ptimization, a reinforcement learning framework that uses non-thinking rollouts as same-model references in advantage estimation. By constructing mixed-mode rollout groups with both thinking and non-thinking trajectories, MARGO evaluates whether explicit thinking adds factual value beyond direct answering, thereby suppressing hallucination-prone thinking while preserving beneficial thinking behaviors. Experiments across multiple factuality-oriented QA benchmarks demonstrate that MARGO improves factual reliability over strong baselines, while evaluations on mathematical benchmarks show that it preserves general reasoning ability.

cs.CL↗

arXiv2026

Enhancing Oracle Bone Inscription Recognition via Multi-Scale Layer Attention

Oracle Bone Inscriptions (OBIs) recognition plays a crucial role in understanding ancient Chinese culture. However, accurately recognizing OBIs remains highly challenging due to their complex, irregular, and often degraded shapes. Traditional methods rely on expert knowledge and manual analysis, which are time-consuming and error-prone. Although deep learning has greatly advanced general image recognition, existing methods struggle to capture the fine-grained details and subtle variations inherent in OBIs, resulting in limited performance. Even most recent and effective layer attention techniques are designed to capture fine-grained dependencies through enhanced inter-layer interactions, yet they still exhibit only marginal improvements in OBIs recognition. To address these limitations, we propose Multi-Scale Layer Attention (MSLA), a novel paradigm that explicitly models both multi-scale and cross-layer feature interactions. By enriching the representation with fine-grained details across multiple spatial scales, MSLA enables more accurate and robust OBIs recognition. Extensive experiments on large-scale OBIs datasets demonstrate that MSLA consistently outperforms existing attention mechanisms while maintaining computational efficiency.

cs.CV↗

arXiv2026

ImAgent: A Unified Multimodal Agent Framework for Test-Time Scalable Image Generation

Recent text-to-image (T2I) models have made remarkable progress in generating visually realistic and semantically coherent images. However, they still suffer from randomness and inconsistency with the given prompts, particularly when textual descriptions are vague or underspecified. Existing approaches, such as prompt rewriting, best-of-N sampling, and self-refinement, can mitigate these issues but usually require additional modules and operate independently, hindering test-time scaling efficiency and increasing computational overhead. In this paper, we introduce ImAgent, a training-free unified multimodal agent that integrates reasoning, generation, and self-evaluation within a single framework for efficient test-time scaling. Guided by a policy controller, multiple generation actions dynamically interact and self-organize to enhance image fidelity and semantic alignment without relying on external models. Extensive experiments on image generation and editing tasks demonstrate that ImAgent consistently improves over the backbone and even surpasses other strong baselines where the backbone model fails, highlighting the potential of unified multimodal agents for adaptive and efficient image generation under test-time scaling.

cs.CV↗

arXiv2026

Multi-Crit: Benchmarking Multimodal Judges on Pluralistic Criteria-Following

Large multimodal models (LMMs) are increasingly adopted as judges in multimodal evaluation systems due to their strong instruction following and consistency with human preferences. However, their ability to follow diverse, fine-grained evaluation criteria remains underexplored. We develop Multi-Crit, a benchmark for evaluating multimodal judges on their capacity to follow pluralistic criteria and produce reliable criterion-level judgments. Covering both open-ended generation and verifiable reasoning tasks, Multi-Crit is built through a rigorous data curation pipeline that gathers challenging response pairs with multi-criterion human annotations. It further introduces three novel metrics for systematically assessing pluralistic adherence, criterion-switching flexibility, and the ability to recognize criterion-level preference conflicts. Comprehensive analysis of 25 LMMs reveals that 1) proprietary models still struggle to maintain consistent adherence to pluralistic criteria--especially in open-ended evaluation; 2) open-source models lag further behind in flexibly following diverse criteria; and 3) critic fine-tuning with holistic judgment signals enhances visual grounding but fails to generalize to pluralistic criterion-level judgment. Additional analyses on reasoning fine-tuning, test-time scaling, and boundary consistency between open-source and proprietary models further probe the limits of current multimodal judges. As a pioneering study, Multi-Crit lays the foundation for building reliable and steerable multimodal AI evaluation.

cs.CV↗

arXiv2026

Enhancing Feature Fusion of U-like Networks with Dynamic Skip Connections

U-like networks have become fundamental frameworks in medical image segmentation through skip connections that bridge high-level semantics and low-level spatial details. Despite their success, conventional skip connections exhibit two key limitations: inter-feature constraints and intra-feature constraints. The inter-feature constraint refers to the static nature of feature fusion in traditional skip connections, where information is transmitted along fixed pathways regardless of feature content. The intra-feature constraint arises from the insufficient modeling of multi-scale feature interactions, thereby hindering the effective aggregation of global contextual information. To overcome these limitations, we propose a novel Dynamic Skip Connection (DSC) block that fundamentally enhances cross-layer connectivity through adaptive mechanisms. The DSC block integrates two complementary components. (1) Test-Time Training (TTT) module. This module addresses the inter-feature constraint by enabling dynamic adaptation of hidden representations during inference, facilitating content-aware feature refinement. (2) Dynamic Multi-Scale Kernel (DMSK) module. To mitigate the intra-feature constraint, this module adaptively selects kernel sizes based on global contextual cues, enhancing the network capacity for multi-scale feature integration. The DSC block is architecture-agnostic and can be seamlessly incorporated into existing U-like network structures. Extensive experiments demonstrate the plug-and-play effectiveness of the proposed DSC block across CNN-based, Transformer-based, hybrid CNN-Transformer, and Mamba-based U-like networks.

cs.CV↗

arXiv2026

Parallel-Probe: Towards Efficient Parallel Thinking via 2D Probing

Parallel thinking has emerged as a promising paradigm for reasoning, yet it imposes significant computational burdens. Existing efficiency methods primarily rely on local, per-trajectory signals and lack principled mechanisms to exploit global dynamics across parallel branches. We introduce 2D probing, an interface that exposes the width-depth dynamics of parallel thinking by periodically eliciting intermediate answers from all branches. Our analysis reveals three key insights: non-monotonic scaling across width-depth allocations, heterogeneous reasoning branch lengths, and early stabilization of global consensus. Guided by these insights, we introduce $\textbf{Parallel-Probe}$, a training-free controller designed to optimize online parallel thinking. Parallel-Probe employs consensus-based early stopping to regulate reasoning depth and deviation-based branch pruning to dynamically adjust width. Extensive experiments across three benchmarks and multiple models demonstrate that Parallel-Probe establishes a superior Pareto frontier for test-time scaling. Compared to standard majority voting, it reduces sequential tokens by up to $\textbf{35.8}$% and total token cost by over $\textbf{25.8}$% while maintaining competitive accuracy.

cs.CL↗

arXiv2026

HalluRNN: Mitigating Hallucinations via Recurrent Cross-Layer Reasoning in Large Vision-Language Models

Though Large Vision-Language Models (LVLMs) have achieved remarkable performance across various tasks, they are still prone to hallucinations-generating outputs that are textually plausible but visually ungrounded. While prior approaches generally address this issue through data-centric fine-tuning or innovative decoding strategies, these methods often require substantial resources or task-specific configurations. In this work, we introduce an architecture-level solution, HalluRNN, which enhances model stability through recurrent cross-layer reasoning. Specifically, we propose a novel Dual-Gated Depth Propagation Unit (DG-DPU) module, which is shared across layers and recurrently refines hidden states. This allows for the adaptive propagation of information throughout the model, enforces consistency across layers, and mitigates hallucinations caused by representational drift. By fine-tuning only the DG-DPU module, HalluRNN achieves strong and robust performance across multiple benchmarks.

cs.CV↗

arXiv2026

Zer0-Jack: A Memory-efficient Gradient-based Jailbreaking Method for Black-box Multi-modal Large Language Models

Jailbreaking methods, which induce Multi-modal Large Language Models (MLLMs) to output harmful responses, raise significant safety concerns. Among these methods, gradient-based approaches, which use gradients to generate malicious prompts, have been widely studied due to their high success rates in white-box settings, where full access to the model is available. However, these methods have notable limitations: they require white-box access, which is not always feasible, and involve high memory usage. To address scenarios where white-box access is unavailable, attackers often resort to transfer attacks. In transfer attacks, malicious inputs generated using white-box models are applied to black-box models, but this typically results in reduced attack performance. To overcome these challenges, we propose Zer0-Jack, a method that bypasses the need for white-box access by leveraging zeroth-order optimization. We propose patch coordinate descent to efficiently generate malicious image inputs to directly attack black-box MLLMs, which significantly reduces memory usage further. Through extensive experiments, Zer0-Jack achieves a high attack success rate across various models, surpassing previous transfer-based methods and performing comparably with existing white-box jailbreak techniques. Notably, Zer0-Jack achieves a 95\% attack success rate on MiniGPT-4 with the Harmful Behaviors Multi-modal Dataset on a black-box setting, demonstrating its effectiveness. Additionally, we show that Zer0-Jack can directly attack commercial MLLMs such as GPT-4o. Codes are provided in the supplement.

cs.LG↗

arXiv2025

FuseUNet: A Multi-Scale Feature Fusion Method for U-like Networks

Medical image segmentation is a critical task in computer vision, with UNet serving as a milestone architecture. The typical component of UNet family is the skip connection, however, their skip connections face two significant limitations: (1) they lack effective interaction between features at different scales, and (2) they rely on simple concatenation or addition operations, which constrain efficient information integration. While recent improvements to UNet have focused on enhancing encoder and decoder capabilities, these limitations remain overlooked. To overcome these challenges, we propose a novel multi-scale feature fusion method that reimagines the UNet decoding process as solving an initial value problem (IVP), treating skip connections as discrete nodes. By leveraging principles from the linear multistep method, we propose an adaptive ordinary differential equation method to enable effective multi-scale feature fusion. Our approach is independent of the encoder and decoder architectures, making it adaptable to various U-Net-like networks. Experiments on ACDC, KiTS2023, MSD brain tumor, and ISIC2017/2018 skin lesion segmentation datasets demonstrate improved feature utilization, reduced network parameters, and maintained high performance. The code is available at https://github.com/nayutayuki/FuseUNet.

cs.CV↗

arXiv2024

Strengthening Layer Interaction via Dynamic Layer Attention

In recent years, employing layer attention to enhance interaction among hierarchical layers has proven to be a significant advancement in building network structures. In this paper, we delve into the distinction between layer attention and the general attention mechanism, noting that existing layer attention methods achieve layer interaction on fixed feature maps in a static manner. These static layer attention methods limit the ability for context feature extraction among layers. To restore the dynamic context representation capability of the attention mechanism, we propose a Dynamic Layer Attention (DLA) architecture. The DLA comprises dual paths, where the forward path utilizes an improved recurrent neural network block, named Dynamic Sharing Unit (DSU), for context feature extraction. The backward path updates features using these shared context representations. Finally, the attention mechanism is applied to these dynamically refreshed feature maps among layers. Experimental results demonstrate the effectiveness of the proposed DLA architecture, outperforming other state-of-the-art methods in image recognition and object detection tasks. Additionally, the DSU block has been evaluated as an efficient plugin in the proposed DLA architecture.The code is available at https://github.com/tunantu/Dynamic-Layer-Attention.

cs.CV↗

arXiv2024

Learning Top-k Subtask Planning Tree based on Discriminative Representation Pre-training for Decision Making

Many complicated real-world tasks can be broken down into smaller, more manageable parts, and planning with prior knowledge extracted from these simplified pieces is crucial for humans to make accurate decisions. However, replicating this process remains a challenge for AI agents and naturally raises two questions: How to extract discriminative knowledge representation from priors? How to develop a rational plan to decompose complex problems? Most existing representation learning methods employing a single encoder structure are fragile and sensitive to complex and diverse dynamics. To address this issue, we introduce a multiple-encoder and individual-predictor regime to learn task-essential representations from sufficient data for simple subtasks. Multiple encoders can extract adequate task-relevant dynamics without confusion, and the shared predictor can discriminate the task characteristics. We also use the attention mechanism to generate a top-k subtask planning tree, which customizes subtask execution plans in guiding complex decisions on unseen tasks. This process enables forward-looking and globality by flexibly adjusting the depth and width of the planning tree. Empirical results on a challenging platform composed of some basic simple tasks and combinatorially rich synthetic tasks consistently outperform some competitive baselines and demonstrate the benefits of our design.

cs.AI↗