论文检索

Deep Learning · Large Language Models

Zhendong Mi, Yixiao Chen, Pu Zhao, Xiaodong Yu, Hao Wang, Yanzhi Wang, Shaoyi Huang

Mixture-of-Experts (MoE) based Large Language Models (LLMs) have achieved superior performance, yet the massive memory overhead caused by storing multiple expert networks severely hinders their practical deployment. Singular Value Decomposition (SVD)-based compression has emerged as a promising post-training technique; however, most existing methods apply uniform rank allocation or rely solely on static weight properties. This overlooks the substantial heterogeneity in expert utilization observed in MoE models, where frequent routing patterns and intrinsic information density vary significantly across experts. In this work, we propose RFID-MoE, an effective framework for MoE compression by exploiting heterogeneous Routing Frequency and Information Density. We first introduce a fused metric that combines expert activation frequency with effective rank to measure expert importance, adaptively allocating higher ranks to critical expert groups under a fixed budget. Moreover, instead of discarding compression residuals, we reconstruct them via a parameter-efficient sparse projection mechanism to recover lost information with minimal parameter overhead. Extensive experiments on representative MoE LLMs (e.g., Qwen3, DeepSeekMoE) across multiple compression ratios demonstrate that RFID-MoE consistently outperforms state-of-the-art methods like MoBE and D2-MoE. Notably, RFID-MoE achieves a perplexity of 16.92 on PTB with the Qwen3-30B model at a 60% compression ratio, reducing perplexity by over 8.0 compared to baselines, and improves zero-shot accuracy on HellaSwag by approximately 8%.

Deep Learning · Large Language Models

Xiaozhe Li, Yang Li, Xinyu Fang, Shengyuan Ding, Peiji Li, Yongkang Chen, Yichuan Ma, TianYi Lyu, Linyang Li, Dahua Lin 等

On-policy reinforcement learning methods like GRPO suffer from \emph{mode collapse}: they exhibit reduced solution diversity, concentrating probability mass on a single solution once discovered and ceasing exploration of alternative strategies. We show this stems from reverse KL minimization's mode-seeking behavior, which reinforces the first high-reward trajectory found rather than maintaining a distribution over multiple diverse solutions. We propose DMPO (\textbf{D}istribution-\textbf{M}atching \textbf{P}olicy \textbf{O}ptimization), which prevents mode collapse through principled approximation of forward KL minimization. DMPO constructs a group-level target distribution over sampled trajectories proportional to their rewards, then aligns the policy distribution to this target. This provides mode-covering behavior without requiring sampling from the intractable global target distribution, enabling sustained exploration throughout training. We validate DMPO on NP-hard combinatorial optimization, where exponentially many feasible solutions exist but only a few approach optimality—an ideal testbed for evaluating exploration. DMPO achieves {43.9\% Quality Ratio on text-based NP-Bench (vs. GRPO's 40.1\%)} and {43.1\% on vision-based NP-Bench (vs. 38.4\%)}—demonstrating 9\% and 12\% relative improvements respectively. These gains generalize to mathematical reasoning (+2.0\%) and out-of-domain tasks (+2.3\%), showing that diversity-preserving training enhances general reasoning capabilities across modalities. Our work establishes distribution matching as a practical, principled approach to preventing mode collapse in on-policy RL, with consistent quality improvements demonstrating sustained exploration across diverse reasoning tasks.

Social Aspects · Privacy

Shihao Wang, Xueru Zhang

Applying differential privacy (DP) via DP-SGD to Low-Rank Adaptation (LoRA) is a natural approach for privacy-preserving fine-tuning. However, applying DP-SGD to LoRA poses a fundamental challenge due to its low-rank parameterization. In LoRA, each trainable update is represented as a low-rank matrix $Z=AB^\top$, but this factorization is non-identifiable. As a result, applying DP-SGD directly to factors $(A,B)$ induces gauge-dependent perturbations on $Z$, leading to uncontrolled noise amplification. We propose **PRISM**, an intrinsic DP mechanism for LoRA that is gauge invariant by construction, avoids bilinear noise amplification, and admits an efficient low-dimensional noise sampler. Moreover, PRISM yields a closed-form characterization for the effective intrinsic noise on $Z$, and enables stable privacy–utility trade-offs by being gauge invariant and keeping noise amplification bounded. We further characterize the noise amplification incurred by naive DP-LoRA and show that it can be unbounded, establish standard $(\varepsilon,\delta)$-DP guarantees for PRISM, and introduce a DP-aware, gauge-invariant adaptive update that avoids amplifying injected privacy noise under adaptive optimization, improving numerical stability in practice.

Deep Learning · Large Language Models

Haodong Zhu, Ren Yangyang, Yanjing Li, Mingbao Lin, Linlin Yang, Xuhui Liu, Xiantong Zhen, haiguang liu, Baochang Zhang

Group Relative Policy Optimization (GRPO) effectively scales LLM reasoning but incurs prohibitive computational costs due to its extensive group-based sampling requirement. While recent selective data utilization methods can mitigate this overhead, they could induce estimation bias by altering the underlying sampling distribution, compromising theoretical rigor and convergence behavior. To address this limitation, we propose Dynamic Pruning Policy Optimization (DPPO), a framework that enables dynamic pruning while preserving unbiased gradient estimation through importance sampling-based correction. By incorporating mathematically derived rescaling factors, DPPO significantly accelerates GRPO training without altering the optimization objective of the full-batch baseline. Furthermore, to mitigate the data sparsity induced by pruning, we introduce Dense Prompt Packing, a window-based greedy strategy that maximizes valid token density and hardware utilization. Extensive experiments demonstrate that DPPO consistently accelerates training across diverse models and benchmarks. For instance, on Qwen3-4B trained on MATH, DPPO achieves 2.37$\times$ training speedup and outperforms GRPO by 3.36\% in average accuracy across six mathematical reasoning benchmarks.

Applications · Time Series

Hua Wang, Xianhao Jiao, Fan Zhang

Deep forecasting models often suffer from attenuated periodic perception and entangled trend–noise representations as network depth increases. Moreover, the widely adopted channel-independent paradigm, while improving training stability, disrupts intrinsic dynamic coordination among variables, hindering the modeling of cross-variable consistency in multivariate time series. To address these issues, we propose PESD-TSF, a physics-inspired structured decomposition framework for long-term time series forecasting that jointly emphasizes interpretability and predictive accuracy. PESD-TSF introduces three key designs. First, a Multiplicative Periodic Gating mechanism incorporates continuous-time priors to dynamically modulate signal amplitudes, preserving periodic structures across deep layers. Second, a multi-scale structured encoder integrates detrended attention with hierarchical sampling to explicitly decouple long-term trends from high-frequency variations while retaining fine-grained temporal semantics. Third, to recover disrupted inter-variable dependencies, we propose Cross-Scale Collaborative Attention (CSCA) together with an RLC regularization scheme, which reconstructs global inter-variable topology in deep feature spaces and enforces physically consistent collaboration through orthogonality and consistency constraints. Extensive experiments on benchmark datasets from multiple domains demonstrate that PESD-TSF consistently achieves state-of-the-art performance, with particularly strong gains on multivariate forecasting tasks involving complex inter-variable coupling, highlighting its superior structural modeling capability and generalization.

Social Aspects · Everything Else

Xinlei Wang, Ruibo Ming, Jing Qiu, Junhua Zhao, Jinjin Gu

The scaling-law era has transformed artificial intelligence from research into a global industry, but its rapid growth raises concerns over energy usage, carbon emissions, and environmental sustainability. Unlike traditional sectors, the AI industry still lacks systematic carbon accounting methods that support large-scale estimates without reproducing the original model. This leaves open questions about how large the problem is today and how large it might be in the near future. Given that the Hugging Face (HF) platform well represents the broader open-source community, we treat it as a large-scale, publicly accessible, and audit-ready corpus for carbon accounting. We propose a FLOPs-based framework to estimate aggregate training emissions of HF open-source models. Considering their uneven disclosure quality, we introduce a tiered approach to handle incomplete metadata, supported by empirical regressions that verify the statistical significance. Compute is also converted to AI training carbon intensity (ATCI, emissions per compute), a metric to assess the sustainability efficiency of model training. Our results show that training the most popular models (with over 5,000 downloads) has resulted in approximately 5.8×10^4 tons of carbon emissions. This paper provides a framework for large-scale emission estimations and a practical methodology to guide future standards and sustainability strategies in the AI industry.

Deep Learning · Large Language Models

Kailin Jiang, Hongbo Jiang, Ning Jiang, Zhi Gao, Jinhe Bi, Yuchen Ren, Bin Li, Yuntao Du, Lei Liu, Qing Li

Large Multimodal Models encode extensive factual knowledge in their pre-trained weights. However, its knowledge remains static and limited, unable to keep pace with real-world developments, which hinders continuous knowledge acquisition. Effective knowledge injection thus becomes critical, involving two goals: knowledge adaptation (injecting new knowledge) and knowledge retention (preserving old knowledge). Existing methods often struggle to learn new knowledge and suffer from catastrophic forgetting. To address these challenges, we propose KORE, a synergistic method centered around KnOwledge-oRientEd controls. These controls are implemented through a two-stage optimization process: (1) KORE automatically converts individual knowledge items into structured and comprehensive knowledge to ensure that the model accurately learns new knowledge, enabling accurate adaptation. (2) KORE stores previous knowledge in the covariance matrix of LMM's linear layer activations and initializes the adapter by projecting the original weights into the matrix's null space, defining a fine-tuning direction that minimizes interference with previous knowledge, enabling powerful retention. Extensive experiments on various LMMs, including LLaVA-v1.5 (7B), LLaVA-v1.5 (13B), and Qwen2.5-VL (7B), show that KORE achieves superior new knowledge injection performance and effectively mitigates catastrophic forgetting.

Deep Learning · Large Language Models

Haocheng Yang, Xiang Cheng, ZONGDA HAN, Pengjie Wang, Changkang Chi, Pengfei Zhang, Sen Su

Fine-tuning Large Language Models (LLMs) enables data holders to construct proprietary, task-specific models by leveraging external high-performance computing infrastructure. However, existing paradigms typically address data privacy and model intellectual property (IP) in isolation, failing to simultaneously uphold both constraints. Privacy-prioritized methods compromise model IP by hosting parameters remotely, while IP-oriented collaborative schemes relying on end-to-end gradient flows inherently violate strict data privacy standards. To address these challenges, we present **PISA** (**P**rivacy-preserving and **I**P-protected **S**plit **A**daptation), a split fine-tuning framework designed to preserve both data privacy and model IP while maintaining high utility. In PISA, we propose three methods: a Manifold Rectification Pre-training (MRP) method to equip the server-side model with intrinsic robustness against privacy-induced distribution shifts; a Dual-Stream Semantic Compensation (DSC) method to recover feature utility using local clean data as priors; and a Utility-Aware Gradient Rectification (UGR) method to adaptively maximize the performance of the parameter-constrained local model. Experiments on GLUE show that PISA ensures dual protection and delivers a substantial 23.0\% performance gain over the privacy-prioritized baseline under strict privacy budgets.

Yuxuan Li, Yuming Chen, Yunheng Li, Ming-Ming Cheng, Xiang Li, jian Yang

Heterogeneous multi-modal remote sensing object detection aims to accurately detect objects from diverse sensors (e.g., RGB, SAR, Infrared). Existing approaches largely adopt a late alignment paradigm, in which modality alignment and task-specific optimization are entangled during downstream fine-tuning. This tight coupling complicates optimization and often results in unstable training and suboptimal generalization. To address these limitations, we propose BabelRS, a unified language-pivoted pretraining framework that explicitly decouples modality alignment from downstream task learning. BabelRS comprises two key components: Concept-Shared Instruction Aligning (CSIA) and Layerwise Visual-Semantic Annealing (LVSA). CSIA aligns each sensor modality to a shared set of linguistic concepts, using language as a semantic pivot to bridge heterogeneous visual representations. To further mitigate the granularity mismatch between high-level language representations and dense detection objectives, LVSA progressively aggregates multi-scale visual features to provide fine-grained semantic guidance. Extensive experiments demonstrate that BabelRS stabilizes training and consistently outperforms state-of-the-art methods without bells and whistles. Code will be released soon.

Deep Learning · Large Language Models

Zekai Yu, Qi Meng, Qizhi Chu, Yu Hao, Chuan Shi, Cheng Yang

Tool calling extends large language models (LLMs) by enabling grounded interaction with external executable interfaces, thereby supporting environment-coupled problem solving. However, mainstream in-context learning (ICL) approaches typically incorporate detailed tool documentation and usage examples directly into the context. This results in substantial inference overhead and heightened risks of hallucination as the context length grows. Conversely, while tuning-based methods improve general tool-calling capabilities, they often fail to effectively internalize the specific details of previously seen tools, thereby retaining a dependency on in-context documentation. To address these limitations, we propose ParaTool, a framework that projects each tool into a dedicated, loadable set of parameters. By equipping a dynamic integration of these parameterized tools, the LLM can perform tool calling without relying on in-context documents or examples. Specifically, our approach consists of three stages: (1) parametric tool pre-training encapsulates the knowledge of different tools into independent parameter modules; (2) soft tool selection employs a gating network to dynamically weigh and aggregate relevant tool parameters; and (3) parametric tool fine-tuning jointly updates tool parameters to align the training and inference processes. Experiments on Stable ToolBench and BFCL demonstrate that ParaTool significantly outperforms strong ICL-based baselines, achieving superior performance while reducing computational complexity.

Applications · Time Series

Zongjiang Shang, Chengxi Jin, Binqing Wu, Dongliang Cui, Yue Yu, Haobang Sun, Chuanlin Xu, Ling Chen

Time series forecasting plays a pivotal role in data-driven decision-making across various time series domains. Recently, leveraging their ability to extract semantically rich representations, Large Language Models (LLMs) have achieved promising results in time series forecasting. However, existing LLM-based methods struggle to obtain multi-scale retrieval-augmented representations due to entangled multi-scale representations and redundant multi-scale interference. To address this, we propose TimeMRA, an LLM-empowered Time series forecasting framework via Multi-Scale Retrieval-Augmented representations. Specifically, a scale-aware prompt generation (SAPG) module is designed to decompose time series into multiple scales and generate augmented multi-scale representations. Then, a cross-scale disentanglement constraint (CSDC) mechanism with a router network is designed to obtain the disentangled multi-scale semantic representations while mitigating interference from irrelevant scales. Finally, a cross-modality retrieval module is designed to obtain multi-scale retrieval-augmented representations for time series forecasting. Experiments on 10 real-world datasets demonstrate that TimeMRA achieves state-of-the-art (SOTA) performance.

Deep Learning · Large Language Models

Haoran Ou, Kangjie Chen, Xingshuo Han, Gelei Deng, Jie Zhang, Han Qiu, Tianwei Zhang, Kwok Yan Lam

Large Language Models (LLMs) have been augmented with web search to overcome the limitations of the static knowledge boundary by accessing up-to-date information from the open Internet. While this integration enhances model capability, it also introduces a distinct safety threat surface: the retrieval and citation process has the potential risk of exposing users to harmful or low-credibility web content. Existing red-teaming methods are largely designed for standalone LLMs as they primarily focus on unsafe generation, ignoring risks emerging from the complex search workflow. To address this gap, we propose CREST-Search, a pioneering red-teaming framework for LLMs with web search. The cornerstone of CREST-Search is three novel attack strategies that generate seemingly benign search queries yet induce unsafe citations. It also employs an iterative in-context refinement mechanism to strengthen adversarial effectiveness under black-box constraints. In addition, we construct a search-specific harmful dataset, WebSearch-Harm, which enables fine-tuning a specialized red-teaming model to improve query quality. Our experiments demonstrate that CREST-Search can effectively bypass safety filters and systematically expose vulnerabilities in web search-based LLM systems, underscoring the necessity of the development of robust search models.

Deep Learning · Large Language Models

Chenxiao Yang, Nati Srebro, Zhiyuan Li

Modern language models reason within bounded attention size, a physical constraint that poses a fundamental barrier to long-horizon reasoning. We identify recursion as a core principle for overcoming this barrier, and propose recursive models as a minimal realization, where the model can recursively invoke itself to solve subtasks in sequences that are contextually isolated. We prove that any computable problem admits a recursive decomposition where subtasks require only exponentially smaller active context than standard autoregressive models, and this approach strictly surpasses any single-context management approaches such as summarization. We further show that modern agentic systems are naturally suited for realizing recursion in a generalized way where arbitrary processing of contexts and workflows is allowed, and prove they can achieve the same power as recursive models, yet none can surpass it. Experimentally, we train a 3B model to learn recursive reasoning and evaluate on SAT, finding that it significantly outperforms frontier LLMs.

Deep Learning · Large Language Models

Yunfan Zhang, Kathleen McKeown, Smaranda Muresan

Large Language Models (LLMs) with agentic web search capabilities show strong potential for tasks requiring real-time information access and complex fact retrieval, yet evaluating such systems remains challenging. We introduce LiveNewsBench, a rigorous and regularly updated benchmark designed to assess the agentic web search abilities of LLMs. LiveNewsBench automatically generates fresh question-answer pairs from recent news articles, ensuring that questions require information beyond an LLM's training data and enabling clear separation between internal knowledge and search capability. The benchmark features intentionally difficult questions requiring multi-step search queries, page visits, and reasoning, making it well-suited for evaluating agentic search behavior. Our automated data curation and question generation pipeline enables frequent benchmark updates and supports construction of a large-scale training dataset for agentic web search models, addressing the scarcity of such data in the research community. To ensure reliable evaluation, we include a subset of human-verified samples in the test set. We evaluate a broad range of systems using LiveNewsBench, including commercial and open-weight LLMs as well as LLM-based web search APIs. The leaderboard, datasets, and code are publicly available at \url{livenewsbench.com}.

Deep Learning · Large Language Models

Udari Sehwag, Elaine Lau, Haniyeh Oskouie, Shayan Shabihi, Erich Liang, Andrea Toledo, Guillermo Mangialardi, Sergio Fonrouge, Ed-Yeremai Hernandez-Cardona, Paula Vergara 等

Accelerating scientific discovery requires the identification of which experiments would yield the best outcomes before committing resources to costly physical validation. While existing benchmarks evaluate LLMs on scientific knowledge and reasoning, their ability to predict experimental outcomes---a task where AI could significantly exceed human capabilities---remains largely underexplored. We introduce SciPredict, a benchmark comprising 405 tasks derived from recent empirical studies in 33 specialized sub-fields of physics, biology, and chemistry. SciPredict addresses two critical questions: (a) *can LLMs predict the outcome of scientific experiments with sufficient accuracy?* and (b) *can such predictions be reliably used in the scientific research process?* Evaluations reveal fundamental limitations on both fronts. Model accuracies are 14-26\% and human expert performance is $\approx$20\%. Although some frontier models exceed human performance model accuracy is still far below what would enable reliable experimental guidance. Even within the limited performance, models fail to distinguish reliable predictions from unreliable ones, achieving only $\approx$20\% accuracy regardless of their confidence or whether they judge outcomes as predictable without physical experimentation. Human experts, in contrast, demonstrate strong calibration: their accuracy increases from $\approx$5\% to $\approx$80\% as they deem outcomes more predictable without conducting the experiment. SciPredict establishes a rigorous framework demonstrating that superhuman performance in experimental science requires not just better predictions, but better awareness of prediction reliability. For reproducibility all our data and code are provided at https://anonymous.4open.science/r/SciPredict-AI01.

Applications · Time Series

Jiegang Xu, Fuyuan CAO, Jiye Liang

Federated causal discovery for time series is becoming increasingly important in many application domains. In practice, intervention policies on each client often change over time, causing the local underlying causal mechanisms to drift rather than remain fixed. Moreover, different sampling frequencies across clients yield incompatible time scales in the observed data, making the resulting local causal graphs naturally heterogeneous and difficult to aggregate consistently. Accordingly, we propose Fed-CAW, a Credibility-Aware Weighting Federated causal discovery framework for time series. Specifically, we define edge-level credibility scores that quantify per-edge reliability by summarizing (i) within-client temporal stability across windows and (ii) cross-client temporal consistency after mapping heterogeneous sampling frequencies onto a unified time scale. We then aggregate privatized edge statistics under differential privacy, treating credibility scores as weights to recover a global causal graph while preserving personalized local structures without sharing raw data. Theoretically, we demonstrate the rationale for the unified time scale mapping and establish rigorous differential privacy guarantees. Experimental results on synthetic and real-world datasets demonstrate the effectiveness of our proposed method.

Reinforcement Learning · Batch/Offline

Zhongjian Qiao, Jiafei Lyu, Chenjia Bai, Peisong Wang, Siyang Gao, Shuang Qiu

Cross-domain offline reinforcement learning (RL) aims to train an agent that performs well in the target domain using a limited target domain dataset and a source domain dataset that exhibits a dynamics shift. Training directly on the original source dataset typically leads to performance collapse. Recent studies perform data filtering from the perspective of dynamics alignment or value alignment to enable efficient policy transfer. However, these studies are typically validated on single-domain or single-behavior-policy source datasets. In this work, we explore a more general setting where the source datasets may be collected from multiple source domains by diverse behavior policies, which we name heterogeneous cross-domain offline RL. We first uncover a critical yet overlooked issue in this setting: \textit{value misassignment}. Empirically and theoretically, we demonstrate that value misassignment can undermine value alignment, mislead data filtering toward selecting suboptimal samples, and loosen the suboptimality gap, thereby degrading the agent’s performance. To address this issue, we propose V2A, a simple yet effective framework that integrates dynamics alignment, value alignment, and value assignment. V2A first employs temporally-consistent modality representation learning to extract dynamics modalities from the source dataset, followed by modality-aware advantage learning to rectify value alignment. Finally, it adopts a data filtering paradigm to selectively share source data for policy learning. Empirical results show that under general heterogeneous cross-domain offline RL settings, V2A significantly outperforms strong baseline methods and exhibits excellent performance across multiple tasks and datasets.

Deep Learning · Large Language Models

Yufei Chen, Yao Wang, Haibin Zhang, Hualin zhou, Tao Gu

Retrieval-augmented generation (RAG) systems suffer from a fundamental functional misalignment where retrievers optimize for semantic relevance, often recalling documents with high background utility but factually erroneous answer spans that generators blindly adopt as cognitive shortcuts. To resolve this, we propose the collaborative Critic-Reasoner framework that shifts robustness control from coarse-grained filtering to fine-grained cognitive decoupling. We disentangle the generation process into two serialized roles by deploying a Critic to perform surgical evidence purification through identifying and masking misleading entities while preserving supportive background context, followed by a Reasoner that switches from rote extraction to deductive reasoning based on the residual evidence. We operationalize this framework via a two-stage alignment strategy combining supervised fine-tuning (SFT) with path-aware direct preference optimization (DPO) to enforce strict behavioral synergy. Experimental results on adversarial benchmarks such as ConFiQA demonstrate that our method significantly outperforms baselines, achieving a 25.99\% accuracy gain in conflicting scenarios and effectively resolving the trust bias dilemma in real-world RAG.

Xin Zhang, Tianjian Ouyang, Yu Shang, Qingmin Liao, Yong Li

Comprehensive urban understanding requires integrating macroscopic spatial structure with fine-grained street-level semantics. However, existing urban Multimodal Large Language Models (MLLMs) primarily rely on satellite imagery, limiting their ability to capture detailed urban appearance and cross-view relationships. We propose \textbf{UrbanMLLM}, a unified MLLM that jointly learns from satellite and street-view imagery for cross-view urban perception and reasoning. To support this, we construct a large-scale dataset with paired cross-view urban images, geospatial alignment, and textual annotations. UrbanMLLM introduces a cross-view perceiver to explicitly model interactions between satellite and street-view representations, and adopts a structural interleaved pre-training paradigm that organizes cross-view image–text content as coherent urban documents to enhance cross-view knowledge fusion. We evaluate UrbanMLLM on 13 diverse urban understanding tasks spanning satellite, street-view, and cross-view settings. Experimental results demonstrate consistent improvements over strong open-source and proprietary MLLMs, highlighting effectiveness and scalability of UrbanMLLM for urban environment understanding.

Deep Learning · Large Language Models

Wenzheng Song, Pei Chen, Yichen Tan, Zejian Li, Lingyun Sun

Multi-modal Diffusion Language Models (MDLMs) have emerged as a powerful alternative to autoregressive models in vision-language understanding, offering advantages in bidirectional context modeling and parallel decoding. However, existing MDLMs suffer from severe visual hallucinations due to the static nature of visual perception. Unlike autoregressive models, MDLMs lack the sequential dependency required to dynamically interact with visual content. Therefore, MDLMs rely on fixed visual features encoded at initialization, causing the denoising process to drift toward language priors and lose its anchor to visual evidence. In this paper, we propose VGR (Visual-Guided Refinement), a framework that enables MDLMs to revisit visual details by exploiting diffusion dynamics. Our key insight is that the temporal trajectory of confidence during denoising reveals intrinsic uncertainty: while grounded tokens converge smoothly, hallucinated ones exhibit pronounced confidence fluctuation. VGR utilizes this fluctuation signal to detect uncertain spans and corrects them through targeted visual evidence extraction and in-place remasking. Extensive experiments on image captioning and hallucination evaluation benchmarks demonstrate that our method significantly reduces hallucinations and recalls more details, achieving state-of-the-art performance among MDLMs.