近年来，多模态大推理模型（MLRMs）在视觉问答任务中的性能显著提升。然而，我们发现过渡词（如because, however, wait）与幻觉密切相关，并倾向于在高熵状态下表现出高熵状态。我们主张可以从token概率分布中直接提取足够的上下文推理信息。受叠加表示理论的启发，我们提出利用潜在叠加推理来整合多个候选语义并保持潜在推理轨迹。假设依赖离散文本输入可能导致模型倾向于顺序显式推理，在高熵推理阶段可能低估密集上下文线索。因此，我们提出从token概率分布中构建丰富的语义表示以增强上下文推理。为此，我们提出了Latent Entropy-Aware Decoding（LEAD），一种高效的即插即用解码策略，利用语义上下文实现可靠的推理。该方法的核心在于熵感知推理模式切换。模型在高熵状态下使用概率加权的连续嵌入，并在熵降低时过渡回离散token嵌入。此外，我们还提出了一种先验引导的视觉锚点注入策略，鼓励模型关注视觉信息。大量实验表明，LEAD在多个基准上有效减轻了各种MLRMs中的幻觉问题。
英文摘要：
Recent advancements in multimodal large reasoning models (MLRMs) have significantly improved performance in visual question answering. However, we observe that transition words (e.g., because, however, and wait) are closely associated with hallucinations and tend to exhibit high-entropy states. We argue that adequate contextual reasoning information can be directly extracted from the token probability distribution. Inspired by superposed representation theory, we propose leveraging latent superposed reasoning to integrate multiple candidate semantics and maintain latent reasoning trajectories. The hypothesis is that reliance on discrete textual inputs may drive the model toward sequential explicit reasoning, underutilizing dense contextual cues during high-entropy reasoning stages. Therefore, we propose constructing rich semantic representations from the token probability distributions to enhance in-context reasoning. With this goal, we present Latent Entropy-Aware Decoding (LEAD), an efficient plug-and-play decoding strategy that leverages semantic context to achieve reliable reasoning. The heart of our method lies in entropy-aware reasoning mode switching. The model employs probability-weighted continuous embeddings under high-entropy states and transitions back to discrete token embeddings as entropy decreases. Moreover, we propose a prior-guided visual anchor injection strategy that encourages the model to focus on visual information. Extensive experiments show that LEAD effectively mitigates hallucinations across various MLRMs on multiple benchmarks.
With the advent of rich visual representations and pre-trained language models, video captioning has seen continuous improvement over time. Despite the performance improvement, video captioning models are prone to hallucination. Hallucination refers to the generation of highly pathological descriptions that are detached from the source material. In video captioning, there are two kinds of hallucination: object and action hallucination. Instead of endeavoring to learn better representations of a video, in this work, we investigate the fundamental sources of the hallucination problem. We identif…
Understanding and Mitigating Hallucinations in Multimodal Chain-of-Thought Models
School of Computer Science and Ningbo Institute, Northwestern Polytechnical University, China（西北工业大学计算机学院和宁波研究院）；National Engineering Laboratory for Integrated Aero-Space-Ground-Ocean Big Data Application Technology, China（国家空天地海一体化大数据应用技术国家工程实验室）
Cognitive Pivot Points and Visual Anchoring: Unveiling and Rectifying Hallucinations in Multimodal Reasoning Models
Gaoling School of Artificial Intelligence, Renmin University of China（中国人民大学 龙城人工智能学院）；South China Agricultural University（华南农业大学）；South China Normal University（华南师范大学）；Monash University（莫纳什大学）；Jishou University（吉首大学）；Hong Kong University of Science and Technology（香港科技大学）；Tsinghua University（清华大学）
GraphThinker: Reinforcing Temporally Grounded Video Reasoning with Event Graph Thinking
Queen Mary University of London（伦敦玛丽女王大学）；Samsung AI Centre Cambridge（剑桥三星人工智能中心）；Shanghai Artificial Intelligence Laboratory（上海人工智能实验室）；Nanyang Technological University（南洋理工大学）
GraphThink: Graph-Enhanced LLM Thinking for Long-Horizon Embodied Task Planning
Gaoling School of Artificial Intelligence, Renmin University of China（中国人民大学高瓴人工智能学院）；Beijing Academy of Artificial Intelligence（北京人工智能研究院）；Tsinghua University（清华大学）；The Hong Kong Polytechnic University（香港理工大学）
A Survey on Large Language Model Hallucination via a Creativity Perspective
Institute of Computing Technology, CAS（中国科学院计算技术研究所）；University of Chinese Academy of Science（中国科学院大学）；International Digital Economy Academy, IDEA Research（国际数字经济学院IDEA研究院）
GeoReason: Aligning Thinking And Answering In Remote Sensing Vision-Language Models Via Logical Consistency Reinforcement Learning
Aerospace Information Research Institute, Chinese Academy of Sciences（中国科学院航空信息研究所）；Key Laboratory of Target Cognition and Application Technology, Chinese Academy of Sciences（中国科学院目标认知与应用技术重点实验室）；University of Chinese Academy of Sciences（中国科学院大学）
DeepRAG: Thinking to Retrieve Step by Step for Large Language Models
Chinese Information Processing Laboratory, Institute of Software, Chinese Academy of Sciences（中国科学院软件研究所信息处理实验室）；University of Chinese Academy of Sciences（中国科学院大学）；Pattern Recognition Center, WeChat AI, Tencent Inc, China（腾讯人工智能部模式识别中心）
Thinking Like a Botanist: Challenging Multimodal Language Models with Intent-Driven Chain-of-Inquiry
Department of Robotics and Mechatronics Engineering, University of Dhaka（达卡大学机器人与机电工程系）；Department of Computer Science and Engineering, University of Dhaka（达卡大学计算机科学与工程系）；Department of Agronomy, Gazipur Agricultural University（加兹ipur农业大学作物学系）；Department of Botany, University of Dhaka（达卡大学植物学系）
Diagnosing Hallucination Risk in AI Surgical Decision-Support: A Sequential Framework for Sequential Validation
Orthopaedic Centre, The University of Hong Kong - Shenzhen Hospital（香港大学深圳医院骨科中心）；Translational Medicine Centre, The University of Hong Kong - Shenzhen Hospital（香港大学深圳医院转化医学中心）；Department of Orthopaedics & Traumatology, Li Ka Shing Faculty of Medicine, The University of Hong Kong（香港大学李嘉诚医学院骨科与创伤外科部）
Exploring Hallucination of Large Multimodal Models in Video Understanding: Benchmark, Analysis and Mitigation
University of Chinese Academy of Sciences（中国科学院大学）；University of Cincinnati（辛辛那提大学）；National University of Singapore（新加坡国立大学）；Beijing Jiaotong University（北京交通大学）
GuardTrace-VL: Detecting Unsafe Multimodel Reasoning via Iterative Safety Supervision
School of Cyber Science and Technology, University of Science and Technology of China（中国科学技术大学网络安全学院）；Anhui Province Key Laboratory of Digital Security（安徽省数字安全重点实验室）；The University of Hong Kong（香港大学）
SaFeR-VLM: Toward Safety-aware Fine-grained Reasoning in Multimodal Models
West China Biomedical Big Data Center, West China Hospital, SCU（西昌生物医学大数据中心、西昌医院、SCU）；NTU；USTC；TeleAI, China Telecom（TeleAI、中国电信）；BUPT；Tsinghua University（清华大学）；HKUST(Guangzhou)（HKUST（广州））
Evaluating Frontier LLMs on PhD-Level Mathematical Reasoning: A Benchmark on a Textbook in Theoretical Computer Science about Randomized Algorithms
Wyoming Seminary（怀俄明神学院）；San Jose State University（圣何塞州立大学）；Guilin University of Electronic Technology（桂林电子科技大学）；University of California, Berkeley（加州大学伯克利分校）；Northeastern University（东北大学）；Guangdong University of Technology（广东工业大学）
Bridging Modal Isolation in Interleaved Thinking: Supervising Modality Transitions via Stepwise Reinforcement
Shanghai Artificial Intelligence Laboratory（上海人工智能实验室）；Shanghai Jiaotong University（上海交通大学）；Zhejiang University（浙江大学）；University of Chinese Academy of Sciences（中国科学院大学）
