AAAI 2026 Papers — Page 41
AAAI Conference on Artificial Intelligence · 4149 papers
VCapsBench: A Large-scale Fine-grained Benchmark for Video Caption Quality Evaluation
Shi-Xue Zhang (University of Science and Technology Beijing), Xu-Cheng Yin (University of Science and Technology Beijing)
GenerationTransformerLarge Language ModelVision Language ModelVideoTextBenchmark
🎯 What it does: Propose VCapsBench, a fine-grained video caption quality assessment benchmark containing 5,677 videos and 109,796 QA pairs, aiming to systematically evaluate the accuracy, coverage, and consistency of captions across 21-dimensional fine-grained attributes.
VCGD: Visual Clue Guided Decoding with Caption Model for Mitigating Hallucination in Multimodal Large Language Models
Guoqing Chen (Northeastern University), Jingwei Cheng (Northeastern University)
Reinforcement Learning from Human FeedbackTransformerReinforcement LearningVision Language ModelContrastive LearningMultimodalityBenchmark
🎯 What it does: Proposed the Visual Cue Guided Decoding (VCGD) method, which utilizes an external Caption model to generate precise visual cues during the decoding process to guide multimodal large language models (MLLM), thereby significantly reducing hallucination generation.
VEDA: Generation of 3D Molecules via Variance-Exploding Diffusion with Annealing
Peining Zhang (University of Connecticut), Minghu Song (Institute of Health and Medicine Hefei Comprehensive National Science Center)
Drug DiscoveryGraph Neural NetworkDiffusion modelGraphBiomedical Data
🎯 What it does: This paper proposes a unified SE(3)-equivariant scattering generation framework, VEDA, which combines variance explosion (VE) diffusion with heating denoising and preprocessing to efficiently generate geometrically accurate and chemically valid 3D molecular structures within a small number of sampling steps.
Veli: Unsupervised Method and Unified Benchmark for Low-Cost Air Quality Sensor Correction
Yahia Dalbah (University of Amsterdam), Yen-Chia Hsu (University of Amsterdam)
RestorationAuto EncoderTime SeriesBenchmark
🎯 What it does: Propose a low-cost, unsupervised, no-reference site air quality sensor reading calibration method called Veli, and create the largest air quality sensor data warehouse, AQ-SDR.
Venom: Liquid Diffusion-Guided Gradient Inversion for Breaking Differential Privacy in Federated Learning
Bin Hu (Wuhan University of Technology), Chuang Hu (Wuhan University)
Federated LearningSafty and PrivacyAdversarial AttackRecurrent Neural NetworkDiffusion modelScore-based ModelImage
🎯 What it does: Propose Venom, a gradient inversion attack based on liquid diffusion, which can recover private data from differentially private gradients without requiring a noise prior.
Verb Mirage: Unveiling and Assessing Verb Concept Hallucinations in Multimodal Large Language Models
Zehao Wang (Shanghai Jiao Tong University), Yong-Lu Li (Shanghai Jiao Tong University)
Explainability and InterpretabilityLarge Language ModelSupervised Fine-TuningVision Language ModelContrastive LearningMultimodalityBenchmark
🎯 What it does: This paper first constructs a benchmark for evaluating false positives in multimodal large language models (MLLM) focused on verb concepts, systematically assesses the performance of various mainstream MLLMs on verb hallucination, and verifies the inadequacy of existing hallucination mitigation methods without training interventions for this task; subsequently, it proposes a parameter-efficient fine-tuning scheme based on enriched verb knowledge (utilizing the Pangea semantic structure), significantly reducing the verb hallucination rate in MLLMs.
VeriFlow: Modeling Distributions for Neural Network Verification
Faried Abu Zaid (Independent Researcher), Mustafa Yalçıner (TU Dortmund University)
OptimizationExplainability and InterpretabilityFlow-based ModelImage
🎯 What it does: Proposed a flow model called VeriFlow, which restricts the verification scope of neural networks to the high-density regions of the probability distribution learned by the model, thereby enabling formal verification of 'typical' inputs;
VerifyBench: A Systematic Benchmark for Evaluating Reasoning Verifiers Across Domains
Xuzhao Li (Ant Group), Wentao Zhang (Zhongguancun Academy)
Large Language ModelTextBenchmarkChain-of-Thought
🎯 What it does: Proposed a cross-domain VerifyBench benchmark, systematically evaluating specialized verifiers and general-purpose LLM verifiers across four disciplines (mathematics, physics, chemistry, biology).
Versatile Vision-Language Model for 3D Computed Tomography
Jiayu Lei (University of Science and Technology of China), Yanfeng Wang (Shanghai Artificial Intelligence Laboratory)
ClassificationSegmentationGenerationConvolutional Neural NetworkTransformerLarge Language ModelVision Language ModelContrastive LearningMultimodalityBiomedical DataComputed Tomography
🎯 What it does: Developed a multi-modal model CTInstruct for 3D CT, supporting multiple tasks such as diagnosis, segmentation, and generation;
VFCionX: Bridging Large and Small Models for Robust Vulnerability-Fixing Commit Identification
Xing Cui (Institute of Software, Chinese Academy of Sciences), Xiang Ling (Institute of Software, Chinese Academy of Sciences)
ClassificationAI Code AssistantConvolutional Neural NetworkTransformerLarge Language ModelSupervised Fine-TuningTextSequential
🎯 What it does: Propose the VFCionX framework, which collaborates between large language models and small models to automatically identify vulnerability fixes in software submissions.
VGD: Value-Guided Diffusion Toward High-Utility Medical Image Segmentation
Hongyu Zhang (Jilin University), Yingda Lyu (Jilin University)
SegmentationConvolutional Neural NetworkDiffusion modelImageBiomedical Data
🎯 What it does: Proposes a Value-Guided Diffusion (VGD) sampling framework that leverages the loss and uncertainty of downstream segmentation models as value signals to dynamically guide the generation of high-utility medical image-segmentation pairs during the reverse process of diffusion models.
VGGS: VGGT-guided Gaussian Splatting for Efficient and Faithful Sparse-View Surface Reconstruction
Peng Xiang (Tsinghua University), Zhizhong Han (Wayne State University)
GenerationDepth EstimationComputational EfficiencyGaussian SplattingImage
🎯 What it does: Proposed a VGGS method based on 3D Gaussian Splatting, which uses the multi-view depth prior VGGT to guide surface reconstruction under sparse views.
VGGTFace: Topologically Consistent Facial Geometry Reconstruction in the Wild
Xin Ming (Tsinghua University), Feng Xu (Tsinghua University)
GenerationTransformerImagePoint Cloud
🎯 What it does: Automated, topology-consistent 3D geometry reconstruction of multi-view, in-the-wild face images using the VGGT base model.
ViCToR: Improving Visual Comprehension via Token Reconstruction for Pretraining LMMs
Yin Xie (DeepGlint), Jiankang Deng (DeepGlint)
Representation LearningTransformerLarge Language ModelVision Language ModelAuto EncoderImageTextMultimodality
🎯 What it does: Propose the ViCToR framework, which introduces a visual understanding stage in multi-modal pre-training. It replaces original visual tokens with a learnable visual token pool and Hungarian matching, helping LLMs better understand images;
Video Camera Trajectory Editing with Generative Rendering from Estimated Geometry
Junyoung Seo (KAIST AI), Yuki Mitsufuji (Sony AI and Sony Group Corporation)
GenerationData SynthesisDiffusion modelNeural Radiance FieldOptical FlowVideoPoint Cloud
🎯 What it does: Propose a framework named Vid-CamEdit that can resample monocular videos according to user-defined camera trajectories, achieving high-quality video reconstruction from arbitrary viewpoints.
Video Echoed in Music: Semantic, Temporal, and Rhythmic Alignment for Video-to-Music Generation
Xinyi Tong (Central Conservatory of Music), Song-Chun Zhu (Alibaba Group)
GenerationConvolutional Neural NetworkRecurrent Neural NetworkVision Language ModelDiffusion modelAuto EncoderGenerative Adversarial NetworkContrastive LearningVideoMultimodalityAudio
🎯 What it does: Designed and implemented a video-to-music generation framework VeM based on latent diffusion models, capable of generating background music highly consistent with video semantics, timing, and rhythm.
Video Mirror Detection with the Motion-in-Depth Cue
Alex Warren (Swansea University), Rynson W. H. Lau
Object DetectionConvolutional Neural NetworkTransformerOptical FlowVideo
🎯 What it does: Proposes MiD-VMD, a new framework for video mirror detection that utilizes Motion-in-Depth (MiD) hints combining 3D motion, depth, and image features.
Video SimpleQA: Towards Factuality Evaluation in Large Video Language Models
Meng Cao (MBZUAI), Xiaodan Liang (MBZUAI)
Large Language ModelVision Language ModelVideoTextMultimodalityBenchmarkRetrieval-Augmented Generation
🎯 What it does: Constructed a factual consistency benchmark named Video SimpleQA for evaluating factual adherence of large video-language models in video context;
Video Spatial Reasoning with Object-Centric 3D Rollout
Haoran Tang, Xiaodan Liang (Sun Yat Sen University)
Large Language ModelSupervised Fine-TuningReinforcement LearningVision Language ModelVideoTextChain-of-Thought
🎯 What it does: Proposed a 3D Rollout (OCR) strategy based on objects, using structured noise to perturb 3D objects in videos, enabling multimodal large language models to learn global context reasoning in video space reasoning tasks.
VideoChat-A1: Thinking with Long Videos by Chain-of-Shot Reasoning
Zikang Wang (Shanghai Jiao Tong University), Yali Wang (Chinese Academy Of Sciences)
RetrievalExplainability and InterpretabilityComputational EfficiencyTransformerLarge Language ModelVision Language ModelVideoMultimodalityRetrieval-Augmented GenerationChain-of-Thought
🎯 What it does: Proposed VideoChat-A1, a multimodal large model agent based on chain-of-shot reasoning for long video segments, which progressively retrieves, segments, and reflects on video shots to answer long video questions.
VideoSeg-R1:Reasoning Video Object Segmentation via Reinforcement Learning
Zishan Xu (South China Normal University), Lihua Cai (South China Normal University)
SegmentationExplainability and InterpretabilityTransformerLarge Language ModelReinforcement LearningVision Language ModelVideoTextMultimodalityChain-of-Thought
🎯 What it does: Propose VideoSeg-R1, a framework that introduces reinforcement learning into video reasoning segmentation, enabling precise segmentation and tracking of video objects given natural language descriptions.
ViDia2Std: A Parallel Corpus and Methods for Low-Resource Vietnamese Dialect-to-Standard Translation
Khoa Anh Ta (University of Information Technology), Kiet Van Nguyen (University of Information Technology)
TransformerLarge Language ModelSupervised Fine-TuningTextBenchmark
🎯 What it does: This paper constructs a nationwide parallel corpus from Vietnamese dialects to standard Vietnamese called ViDia2Std, trains and benchmarks various seq2seq models on this corpus, and evaluates the practical value of dialect normalization through downstream tasks such as machine translation and sentiment analysis.
VietCheckMed: Explainable Regulatory Compliance Checking for Medical Advertisements on Vietnamese Social Media
Nguyen Thanh Tam (University of Science), Binh T. Nguyen (University of Science)
Explainability and InterpretabilityTransformerLarge Language ModelSupervised Fine-TuningTextBenchmarkRetrieval-Augmented GenerationChain-of-Thought
🎯 What it does: Constructed the VietCheckMed framework to enable explainable regulatory compliance checks on Vietnamese medical advertisements, and released the first expert-verified benchmark, VietAestheticAds.
View-on-Graph: Zero-Shot 3D Visual Grounding via Vision-Language Reasoning on Scene Graphs
Yuanyuan Liu (Dalian University of Technology), Xin Yang (Dalian University of Technology)
RecognitionGraph Neural NetworkVision Language ModelMultimodalityGraph
🎯 What it does: Proposes the View-on-Graph (VoG) method to achieve zero-shot 3D visual localization by leveraging multimodal multi-layer scene graphs for interactive exploration with VLM;
Views Attention Fusion of Granular-ball Fuzzy Representations Split for Improved Multi-view Clustering
Shuaiyu Liu (University of Electronic Science and Technology of China), Guoying Wang
Representation LearningAuto EncoderContrastive LearningImageText
🎯 What it does: Designed a two-stage multi-view clustering framework named GFSAF, first using granular sphere fuzzy contrastive learning to extract mutual information and obtaining complementary information through noise stripping loss, then employing cross-view attention fusion to achieve robust clustering features.
ViG-RAG: Video-aware Graph Retrieval-Augmented Generation via Temporal and Semantic Hybrid Reasoning
Zongsheng Cao (Shanghai Artificial Intelligence Laboratory), Zigan Wang (Tsinghua University)
GenerationRetrievalTransformerVision Language ModelVideoMultimodalityRetrieval-Augmented Generation
🎯 What it does: This paper proposes the ViG-RAG framework, which for long video understanding constructs a probabilistic temporal knowledge graph and combines semantic and temporal dual retrieval to achieve retrieval-augmented generation.
VIL2C: Value-of-Information Aware Low-Latency Communication for Multi-Agent Reinforcement Learning
Qian Zhang (Northwestern Polytechnical University), Jun Zhang (Northwestern Polytechnical University)
OptimizationReinforcement LearningSequential
🎯 What it does: Proposes the VIL2C scheme, which proactively adjusts the delay distribution of multi-agent communication to enhance the performance of collaborative reinforcement learning.
VILTA: A VLM-in-the-Loop Adversary for Enhancing Driving Policy Robustness
Qimao Chen (Tsinghua University), Zhi-Xin Yang (University of Macau)
Autonomous DrivingLarge Language ModelReinforcement LearningVision Language ModelImageMultimodality
🎯 What it does: Propose the VILTA framework, which utilizes Vision-Language Models (VLM) to directly edit the trajectories of surrounding vehicles in closed-loop training, generating diverse and challenging long-tail driving scenarios;
VipAct: Visual-Perception Enhancement via Specialized VLM Agent Collaboration and Tool-use
Zhehao Zhang (Ohio State University), Nedim Lipka (Adobe Inc)
Object DetectionSegmentationDepth EstimationLarge Language ModelAgentic AIVision Language ModelImageTextMultimodality
🎯 What it does: Propose the VIPACT framework, integrating visual language models with multi-agent collaboration and visual expert models to enhance reasoning and execution capabilities for fine-grained visual perception tasks.
VIR-Bench: Evaluating Geospatial and Temporal Understanding of MLLMs via Travel Video Itinerary Reconstruction
Hao Wang (Waseda University), Daisuke Kawahara (Waseda University)
TransformerLarge Language ModelVision Language ModelVideoTextMultimodalityBenchmarkChain-of-ThoughtAudio
🎯 What it does: Propose the VIR-Bench benchmark, construct 200 Japanese travel vlogs and generate itinerary maps, split into node prediction (province, city, POI) and edge prediction (relationship and time transition), and develop a travel planning agent based on this benchmark.
Virtual Multiplex Staining for Histological Images Using a Marker-Wise Conditioned Diffusion Model
Hyun-Jic Oh (Korea University), Won-Ki Jeong (Harvard University)
GenerationData SynthesisDiffusion modelBiomedical Data
🎯 What it does: Propose a virtual multiplex staining framework based on a marker-wise conditional diffusion model, capable of directly generating multiple markers (up to 18) from single-channel H&E images
VirtualEnv: A Platform for Embodied AI Research
Kabir Swain (Massachusetts Institute of Technology), Antonio Torralba (Massachusetts Institute of Technology)
Large Language ModelAgentic AIVision Language ModelWorld ModelTextMultimodalityMeshBenchmarkChain-of-Thought
🎯 What it does: Built a high-fidelity, procedurally generated, and multi-agent collaborative virtual experiment platform called VirtualEnv based on Unreal Engine 5, implementing features such as LLM-driven natural language task generation, escape room benchmark, and real-time environment editing within it.
VisAssist: A Visually Impaired-Captured Video Question Answering Benchmark for Assistive Systems
Qi Gao, Xinyu Chai (Shanghai Jiao Tong University)
Large Language ModelVision Language ModelVideoTextMultimodalityBenchmark
🎯 What it does: Built VisAssist, the first large-scale first-person video question-answering dataset recorded by visually impaired individuals, and conducted benchmark evaluations on multiple state-of-the-art video question-answering models;
Vision Transformers Are Circulant Attention Learners
Dongchen Han (Tsinghua University), Gao Huang (Tsinghua University)
ClassificationObject DetectionSegmentationComputational EfficiencyTransformerImage
🎯 What it does: Designed and implemented a Circulant Attention mechanism, projecting the self-attention matrix in Vision Transformers into a Block Circulant with Circulant Blocks (BCCB) structure, and achieving efficient O(N log N) computation using 2D discrete Fourier transform.
Vision-G1: Towards General Reasoning Vision-Language Models via Reinforcement Learning
Yuheng Zha (UC San Diego), Zhiting Hu (UC San Diego)
Reinforcement LearningVision Language ModelMultimodality
🎯 What it does: Constructed a visual reasoning dataset covering 5 domains and 13 dimensions with 46 tasks, and trained the Vision-G1 VLM using reinforcement learning from verifiable rewards to enhance its general visual reasoning capabilities.
Vision-language Incremental Learning with Dual Class-individual Memory
Fuhai Chen (Fuzhou University), Xuri Ge (Shandong University)
Representation LearningMeta LearningConvolutional Neural NetworkTransformerVision Language ModelAuto EncoderMultimodalityBenchmark
🎯 What it does: Proposes the Dual Class-Individual Memory (DCIM) framework to address category-level and scene-level forgetting issues in visual-lingual incremental learning.
Vision-Language Models Guided Graph Concept Reasoning for Interpretable Diabetic Retinopathy Diagnosis
Qihao Xu (Shenzhen University), Yong Xu (Harbin Institute of Technology)
ClassificationExplainability and InterpretabilityGraph Neural NetworkTransformerMixture of ExpertsVision Language ModelContrastive LearningMultimodalityBiomedical Data
🎯 What it does: Proposed the VLM-GCR framework, achieving interpretable diabetic retinopathy (DR) diagnosis through vision-language models and hierarchical perception concept graphs.
Vision-Language Reasoning for Geolocalization: A Reinforcement Learning Approach
Biao Wu (University of Technology Sydney), Jun Wang (University College London)
TransformerSupervised Fine-TuningReinforcement LearningVision Language ModelImageMultimodalityChain-of-Thought
🎯 What it does: Proposed a retrieval-free, reasoning-based global image geolocation framework Geo-R, which directly predicts coordinates using structured geographic reasoning chains and reinforcement learning.
Vision-MoR: Scaling Vision Transformer via Patch-Level Mixture-of-Recursions
Yunhong He (Independent Researcher), Lichao Sun (University Of Notre Dame)
ClassificationObject DetectionTransformerMixture of ExpertsImage
🎯 What it does: Propose Vision-MoR, a unified Vision Transformer architecture that integrates parameter sharing, spatially adaptive computation, and memory efficiency.
Vision-Only Gaussian Splatting for Collaborative Semantic Occupancy Prediction
Cheng Chen, Saurabh Bagchi (Purdue University)
SegmentationAutonomous DrivingTransformerGaussian SplattingImage
🎯 What it does: Propose a visual collaborative semantic occupancy prediction framework based on sparse 3D Gaussian splitting.
VisionReward: Fine-Grained Multi-Dimensional Human Preference Learning for Image and Video Generation
Jiazheng Xu (Tsinghua University), Yuxiao Dong (Z.AI)
GenerationReinforcement Learning from Human FeedbackSupervised Fine-TuningReinforcement LearningVision Language ModelImageVideo
🎯 What it does: This paper proposes VisionReward, a multidimensional fine-grained human preference learning framework for image and video generation tasks, and uses it as a reward model for reinforcement learning optimization.
Vista: Scene-Aware Optimization for Streaming Video Question Answering Under Post-Hoc Queries
Haocheng Lu (Huazhong University of Science and Technology), Jianzong Wang (Ping An Technology (Shenzhen) Co., Ltd.)
RetrievalCompressionOptimizationVision Language ModelVideoMultimodalityBenchmark
🎯 What it does: Proposed the Vista framework for real-time streaming video question answering, achieving efficient and scalable long-context reasoning through scene-aware segmentation, compression, and retrieval of videos.
Visual Bridge: Universal Visual Perception Representations Generating
Yilin Gao (Shanghai University), Shugong Xu (Xi'an Jiaotong-Liverpool University)
ClassificationObject DetectionSegmentationDepth EstimationRetrievalTransformerDiffusion modelFlow-based ModelImageOrdinary Differential Equation
🎯 What it does: Proposes Vision Bridge, a general-purpose visual perception framework based on flow matching, which maps visual foundation model (e.g., DINOv2) tokens to task-specific representations for multiple tasks including classification, detection, segmentation, depth estimation, and image-text retrieval through a single velocity field.
ViTCoP: Accelerating Large Vision-Language Models via Visual and Textual Semantic Collaborative Pruning
Wen Luo (Huazhong University of Science and Technology), LiQun Huang (Huazhong University of Science and Technology)
Computational EfficiencyTransformerVision Language ModelImageVideoTextMultimodality
🎯 What it does: Proposes a vision-text collaborative pruning framework named ViTCoP, which can significantly reduce the number of visual tokens in vision-language models and lower inference costs.
ViTE: Virtual Graph Trajectory Expert Router for Pedestrian Trajectory Prediction
Ruochen Li (Durham University), Hubert P. H. Shum (Durham University)
Autonomous DrivingComputational EfficiencyGraph Neural NetworkMixture of ExpertsVideo
🎯 What it does: Propose the ViTE framework, which predicts pedestrian trajectories by utilizing virtual graphs constructed with virtual nodes and an expert router based on Mixture-of-Experts.
ViType: High-Fidelity Visual Text Rendering via Glyph-Aware Multimodal Diffusion
Lishuai Gao (Tianjin University of Technology), Xiaoming Wei (Meituan)
GenerationTransformerLarge Language ModelSupervised Fine-TuningDiffusion modelImageTextMultimodality
🎯 What it does: Proposes the ViType framework, enhancing text-to-image generation models to achieve high-fidelity visual text rendering.
VividListener: Expressive and Controllable Listener Dynamics Modeling for Multi-Modal Responsive Interaction
Shiying Li (Beijing University of Posts and Telecommunications), Zhenan Sun (Hong Kong University of Science and Technology)
GenerationData SynthesisTransformerVision Language ModelDiffusion modelTextMultimodalityMeshAudio
🎯 What it does: Propose a multimodal controllable listener head dynamic generation framework called VividListener, which can generate fine-grained, emotionally adjustable head movements based on speaker audio, head motion, text descriptions, and emotional intensity labels.
VK-Det: Visual Knowledge Guided Prototype Learning for Open-Vocabulary Aerial Object Detection
Jianhang Yao (National University of Defense Technology), Peng Sun (National University of Defense Technology)
Object DetectionKnowledge DistillationRepresentation LearningVision Language ModelImage
🎯 What it does: Proposes the VK-Det framework, achieving open-vocabulary aerial object detection without additional supervision through prototype learning guided by visual knowledge.
VLA-Adapter: An Effective Paradigm for Tiny-Scale Vision-Language-Action Model
Yihao Wang (Beijing University of Posts and Telecommunications), Donglin Wang (Beijing University of Posts and Telecommunications)
Computational EfficiencyRobotic IntelligenceTransformerVision-Language-Action ModelMultimodality
🎯 What it does: Proposed VLA-Adapter, a lightweight bridging paradigm that can directly map visual-language representations to the action space without requiring large-scale VLM pre-training;
VMChill: A Dataset for Fine-Grained Visual-Musical Synergy
Xiaowei Chi (Hong Kong University of Science and Technology), Wei Xue (Hong Kong University of Science and Technology)
GenerationData SynthesisTransformerLarge Language ModelSupervised Fine-TuningVision Language ModelDiffusion modelVideoTextMultimodalityBenchmarkAudio
🎯 What it does: Constructed the VMChill large-scale multimodal dataset based on movie trailers, containing over 20M short video clips and 2M high-quality multimodal annotations, along with a music detail subset and a human evaluation set;
vMFCoOp: Towards Equilibrium on a Unified Hyperspherical Manifold for Prompting Biomedical VLMs
Minye Shao (Durham University), Yang Long (Tsinghua University)
ClassificationExplainability and InterpretabilityRepresentation LearningLarge Language ModelPrompt EngineeringVision Language ModelContrastive LearningBiomedical DataMagnetic Resonance ImagingComputed TomographyUltrasound
🎯 What it does: This paper proposes a unified spherical prompt learning framework called vMFCoOp based on the von Mises-Fisher (vMF) distribution, aiming to achieve efficient prompt optimization for medical vision-language models in few-shot scenarios through semantically inverse-estimated anchors on a sphere.
VoiceCloak: A Multi-Dimensional Defense Framework Against Unauthorized Diffusion-Based Voice Cloning
Qianyue Hu (Sun Yat-sen University), Xiangyang Luo (State Key Laboratory of Mathematical Engineering and Advanced Computing)
Safty and PrivacyAdversarial AttackTransformerDiffusion modelScore-based ModelAudio
🎯 What it does: This study proposes VoiceCloak, a defense framework that utilizes adversarial perturbations to multidimensionally interfere with the speaker recognition and synthesis processes of diffusion models, achieving both speaker identity obfuscation and degradation of synthesized speech quality in unauthorized voice cloning.
Voices, Faces, and Feelings: Multi-modal Emotion-Cognition Captioning for Mental Health Understanding
Zhiyuan Zhou (Hefei University of Technology), Shijie Hao (Hefei University of Technology)
GenerationExplainability and InterpretabilityTransformerLarge Language ModelVision Language ModelContrastive LearningVideoTextMultimodalityAudio
🎯 What it does: Propose a multimodal emotion-cognition collaborative generation framework (ECMC) that can generate interpretable emotion-cognition descriptions from video, audio, and text multimodal data for mental health assessment.
VORTEX: Aligning Task Utility and Human Preferences Through LLM-Guided Reward Shaping
Guojun Xiong (Harvard University), Milind Tambe (Harvard University)
OptimizationReinforcement Learning from Human FeedbackTransformerLarge Language ModelReinforcement LearningPrompt EngineeringText
🎯 What it does: Through the VORTEX framework, which uses LLM-guided reward shaping, human preferences are integrated into existing decision optimization systems via natural language, forming a closed-loop of multi-objective optimization and reward shaping.
VP-Bench: A Comprehensive Benchmark for Visual Prompting in Multimodal Large Language Models
Mingjie Xu (Hong Kong University of Science and Technology), Wenqiang Lei (Huazhong University of Science and Technology)
TransformerPrompt EngineeringVision Language ModelImageMultimodalityBenchmark
🎯 What it does: This paper proposes VP-Bench, a two-phase evaluation framework designed to measure the performance of multimodal large language models in visual prompt perception and practical tasks.
VPHO: Joint Visual-Physical Cue Learning and Aggregation for Hand-Object Pose Estimation
Jun Zhou (China University of Geosciences), Li Cheng (University of Alberta)
Pose EstimationConvolutional Neural NetworkDiffusion modelImageBenchmark
🎯 What it does: Propose a hand-object pose estimation framework named VPHO that simultaneously utilizes visual and physical cues
VPN: Visual Prompt Navigation
Shuo Feng (Nanjing University of Aeronautics and Astronautics), Shuqiang Jiang (University of Chinese Academy of Sciences)
Graph Neural NetworkTransformerPrompt EngineeringVision Language ModelImageGraph
🎯 What it does: Proposed the Visual Prompt Navigation (VPN) paradigm, which utilizes visual prompts drawn by users on a 2D panoramic map to guide agents in completing navigation tasks.
VPSentry: Semi-supervised Video Polyp Segmentation via Sentry-guided Long-term Prototype Fusion with Correlation Dynamic Propagation
Guilian Chen (Shenzhen University), Jing Qin (Hong Kong Polytechnic University)
SegmentationConvolutional Neural NetworkTransformerContrastive LearningVideoBiomedical DataBenchmark
🎯 What it does: This paper proposes a semi-supervised video polyp segmentation model called VPSentry, which achieves long-term feature aggregation and detail enhancement through adaptive prototypes, prototype memory, and a related dynamic propagation module. It introduces a sentry mechanism to evaluate inter-frame continuity, avoiding noise accumulation caused by abrupt scene changes and significantly improving segmentation accuracy on unannotated frames.
VQ-Insight: Teaching VLMs for AI-Generated Video Quality Understanding via Progressive Visual Reinforcement Learning
Xuanyu Zhang (Peking University), Jian Zhang (ByteDance Inc)
OptimizationReinforcement Learning from Human FeedbackTransformerSupervised Fine-TuningReinforcement LearningVision Language ModelVideoTextMultimodality
🎯 What it does: Propose VQ-Insight, a video quality assessment framework for AIGC (AI-generated content) based on reasoning-driven vision-language models, capable of video preference comparison, multi-dimensional quality scoring, and natural video scoring, with support for joint fine-tuning with generative models;
VQAThinker: Exploring Generalizable and Explainable Video Quality Assessment via Reinforcement Learning
Linhan Cao (Shanghai Jiao Tong University), Xiongkuo Min (Shanghai Jiao Tong University)
Explainability and InterpretabilityConvolutional Neural NetworkTransformerLarge Language ModelReinforcement LearningVision Language ModelVideoMultimodalityChain-of-Thought
🎯 What it does: Proposed a no-reference video quality assessment framework called VQAThinker, which leverages a large multimodal model combined with reinforcement learning (GRPO) to achieve joint modeling of video quality understanding and scoring, and generates explainable results through reasoning trajectories;
VRAgent-R1: Boosting Video Recommendation with MLLM-based Agents via Reinforcement Learning
Siran Chen (Shenzhen Institutes of Advanced Technology, Chinese Academy of Sciences), Yali Wang (Shenzhen Institutes of Advanced Technology, Chinese Academy of Sciences)
Recommendation SystemTransformerLarge Language ModelReinforcement LearningVision Language ModelVideoTextMultimodalityRetrieval-Augmented GenerationChain-of-Thought
🎯 What it does: Proposed the VRAgent-R1 framework, leveraging MLLM-driven Item Perception Agent (IP) for multimodal video understanding, and User Simulation Agent (US) through reinforcement learning to simulate user behavior, thereby enhancing video recommendation performance.
VSPO: Validating Semantic Pitfalls in Ontology via LLM-Based CQ Generation
Hyojun Choi (Yonsei University), Kyong-Ho Lee (Yonsei University)
TransformerLarge Language ModelSupervised Fine-TuningTextGraph
🎯 What it does: Constructed an alignment error-injected dataset and trained LLaMA-3.1-8B-Instruct to generate CQs with verifiable semantic flaws, addressing the limitations of traditional CQs that rely solely on similarity-based evaluation.
VTD-CLIP: Video-to-Text Discretization via Prompting CLIP
Wencheng Zhu (Tianjin University), Pengfei Zhu (Tianjin University)
ClassificationTransformerPrompt EngineeringVision Language ModelVideo
🎯 What it does: A frozen CLIP model is used to map video frames to text prototypes, generating discrete text representations. The codebook is updated through adaptive text prompts, and confidence-aware fusion is applied to enhance video classification performance.
VTinker: Guided Flow Upsampling and Texture Mapping for High-Resolution Video Frame Interpolation
Chenyang Wu (Nankai University), Chongyi Li (Nankai University)
RestorationSuper ResolutionConvolutional Neural NetworkFlow-based ModelAuto EncoderOptical FlowVideo
🎯 What it does: This paper proposes the VTinker framework, combining guided flow upscaling (GFU) with texture mapping to achieve high-resolution video frame interpolation;
Vulnerability-Aware Robust Multimodal Adversarial Training
Junrui Zhang (University of Science & Technology of China), Tianlong Chen (University of North Carolina at Chapel Hill)
Adversarial AttackMultimodalityBenchmark
🎯 What it does: This paper proposes a multi-modal adversarial training method called VARMAT based on modal vulnerability identification.
W2S-AlignTree: Weak-to-Strong Inference-Time Alignment for Large Language Models via Monte Carlo Tree Search
Zhenyu Ding (Xi'an Jiaotong University), Ning Ding (Xi'an Jiaotong University)
ClassificationGenerationTransformerLarge Language ModelText
🎯 What it does: Proposed a plug-and-play framework called W2S-AlignTree that aligns large language models (LLMs) during the reasoning phase by combining Monte Carlo Tree Search (MCTS) with the weak-to-strong (W2S) strategy;
Walk Before You Dance: High-fidelity and Editable Dance Synthesis via Generative Masked Motion Prior
Foram N Shah (University of North Carolina at Charlotte), Ahmed Helmy (University of North Carolina at Charlotte)
Data SynthesisTransformerAuto EncoderTextMultimodalityAudio
🎯 What it does: Proposed the DanceMosaic framework, achieving high-fidelity 3D dance generation that is editable based on multimodal inputs (music, text, pose).
Walking Further: Semantic-Aware Multimodal Gait Recognition Under Long-Range Conditions
Zhiyang Lu (Fujian Key Laboratory of Sensing and Computing for Smart Cities, Xiamen University), Ming Cheng (OPPO Research Institute)
RecognitionConvolutional Neural NetworkGraph Neural NetworkTransformerSupervised Fine-TuningVision Language ModelContrastive LearningVideoTextMultimodalityPoint CloudBenchmark
🎯 What it does: Proposes the LRGait multimodal long-range gait recognition benchmark and the EMGaitNet framework, directly using raw RGB videos and LiDAR point clouds for end-to-end semantic-guided multimodal fusion.
WALKSAFE: Risk-aware Graph Random Walk with Bi-GRPO for LLM Safety
Shilong Pan (National University of Defense Technology), Dongsheng Li (National University of Defense Technology)
OptimizationSafty and PrivacyGraph Neural NetworkSupervised Fine-TuningReinforcement LearningGraph
🎯 What it does: Propose the WALKSAFE framework, which constructs entity-relationship graphs and performs risk-aware random walks, filters out dangerous relationships, and applies dual-group relative policy optimization (Bi-GRPO) to enhance safety and helpfulness of LLMs.
Wasserstein-Aligned Hyperbolic Multi-View Clustering
Rui Wang (Jiangnan University), Ziheng Chen (University of Trento)
Representation LearningContrastive LearningMultimodalityBenchmark
🎯 What it does: This paper proposes WAH-MVC, a multi-view clustering method that employs Wasserstein-aligned hyperbolic encoding on the Lorentz manifold.
Wasserstein-Aware Transfer: Class-Level Alignment for Robust Diffusion Model Adaptation
Zixian Huang, Chuan-Xian Ren (Sun Yat-Sen University)
GenerationDomain AdaptationTransformerDiffusion modelImage
🎯 What it does: Propose the Wasserstein-Aware Transfer (WAT) method, achieving robust generation of diffusion models under distribution shift through class-level Wasserstein distance matching and linear combination of pre-trained and fine-tuned models.
WaterMod: Modular Token-Rank Partitioning for Probability-Balanced LLM Watermarking
Shinwoo Park (Yonsei University), Yo-Sub Han (Yonsei University)
GenerationTransformerLarge Language ModelText
🎯 What it does: Propose a block-weighted watermarking method called WaterMod based on probabilistic ranking, which uses modulo operations to divide the vocabulary into different color groups according to increasing probability order and applies a slight bias to the logits of the corresponding groups, thereby achieving zero-bit or multi-bit watermark embedding;
WaveC2R: Wavelet-Driven Coarse-to-Refined Hierarchical Learning for Radar Retrieval
Chunlei Shi (Southeast University), Dan Niu (Tsinghua University)
Data SynthesisDiffusion modelMultimodalityPhysics Related
🎯 What it does: Proposed the WaveC2R framework, combining multiscale wavelet decomposition from multisource satellite observations with a conditional diffusion model to achieve precise reconstruction from satellite to radar reflectivity.
WaveDiST: A Wavelet Diffusion Transformer for Spatio-Temporal Estimation on Unobserved Locations
Huiling Qin (Beijing Normal University), Weijia Jia (Beijing Normal University)
TransformerDiffusion modelTime Series
🎯 What it does: This paper proposes a Wavelet Diffusion Transformer (WaveDiST) framework for spatiotemporal state estimation of unobserved spatial locations in urban sensing networks without historical reference data.
WaveEx: Accelerating Flow Matching-based Speech Generation via Wavelet-guided Extrapolation
Xiaoqian Liu (Northeastern University), Linfeng Zhang (Shanghai Jiao Tong University)
GenerationComputational EfficiencyFlow-based ModelOrdinary Differential EquationAudio
🎯 What it does: Propose the WaveEx framework, which replaces part of ODE solving in flow matching speech generation with wavelet decomposition and Taylor extrapolation, significantly accelerating inference.
WaveFormer: Frequency-Time Decoupled Vision Modeling with Wave Equation
Zishan Shu (Peking University), Jie Chen (Peking University)
ClassificationObject DetectionSegmentationTransformerImage
🎯 What it does: Proposed WaveFormer, a frequency-time decoupled visual model based on the damped wave equation, using the Wave Propagation Operator (WPO) instead of traditional attention mechanisms to achieve global semantic propagation.
Wavefront-Constrained Passive Obscured Object Detection
Zhiwen Zheng (Hangzhou Dianzi University), Xingru Huang (Hangzhou Dianzi University)
Object DetectionSegmentationConvolutional Neural NetworkImagePhysics Related
🎯 What it does: A physics-driven WavePCNet network is studied for detecting and segmenting occluded objects in non-line-of-sight (NLOS) scenes using sparse light spot images without active illumination.
Wavelet Enhanced Adaptive Frequency Filter for Sequential Recommendation
Huayang Xu (Soochow University), Pengpeng Zhao (Soochow University)
Recommendation SystemTransformerSequential
🎯 What it does: Proposes a sequence recommendation model called WEARec that integrates dynamic frequency domain filtering with wavelet enhancement
WDT-MD: Wavelet Diffusion Transformers for Microaneurysm Detection in Fundus Images
Yifei Sun (Zhejiang University), Hongxia Xu (Zhejiang University)
ClassificationAnomaly DetectionTransformerDiffusion modelImageBiomedical Data
🎯 What it does: Proposed a diffusion transformer framework based on wavelet decomposition for detecting retinal microaneurysms.
Weakest Bidder Types and New Core-Selecting Combinatorial Auctions
Siddharth Prasad (Toyota Technological Institute at Chicago), Tuomas Sandholm (Carnegie Mellon University)
OptimizationBenchmarkFinance Related
🎯 What it does: This paper proposes a class of core-selecting combinatorial auction designs that leverage information from the bidder type space, and proves that traditional core-selecting auctions can overcome feasibility limitations when the type space is sufficiently constrained.
Weakly-Supervised Image Forgery Localization via Vision-Language Collaborative Reasoning Framework
Ziqi Sheng (Sun Yat-sen University), Jiantao Zhou (University of Macau)
SegmentationAnomaly DetectionGraph Neural NetworkTransformerPrompt EngineeringVision Language ModelContrastive LearningImageMultimodality
🎯 What it does: Propose the ViLaCo framework, which achieves weakly supervised image forgery localization using only image-level labels through semantic supervision from pre-trained vision-language models.
Weather-Robust LiDAR Perception: Point Cloud Restoration from Adverse Weather
Chenghao Sun, Xiangmo Zhao (Chang'an University)
RestorationObject DetectionConvolutional Neural NetworkPoint CloudBenchmark
🎯 What it does: Propose LTDNet, an end-to-end network that learns to recover clean geometry from point clouds degraded under adverse weather conditions, and introduce the IQA3D benchmark to evaluate restoration quality and detection robustness
WeatherEdit: Controllable Weather Editing with 4D Gaussian Field
Chenghao Qian (University of Leeds), Gustav Markkula (University of Leeds)
Image TranslationGenerationData SynthesisAutonomous DrivingDiffusion modelNeural Radiance FieldImageVideo
🎯 What it does: Achieve 3D editing of controllable intensity multi-weather (rain, snow, fog) from ordinary scenes by fusing multi-weather style diffusion models and combining with 4D Gaussian fields;
WeightFlow: Learning Stochastic Dynamics via Evolving Weight of Neural Network
Ruikun Li (Tsinghua University), Yong Li (Tsinghua University)
Graph Neural NetworkTransformerAuto EncoderBiomedical DataStochastic Differential EquationOrdinary Differential Equation
🎯 What it does: Proposed the WeightFlow framework, which directly captures the probability density evolution of stochastic dynamics by modeling probability distributions in the neural network weight space and learning the continuous evolution of weight graphs;
Well Begun, Half Done: Reinforcement Learning with Prefix Optimization for LLM Reasoning
Yiliu Sun (Nanjing University of Science and Technology), Chen Gong (Shanghai Jiao Tong University)
OptimizationTransformerLarge Language ModelReinforcement LearningText
🎯 What it does: Enhancing the reasoning capability of large language models by proposing the PPPO method, which applies reinforcement learning exclusively to prefix tokens.
WenetSpeech-Yue: A Large-Scale Cantonese Speech Corpus with Multi-dimensional Annotation
Longhao Li (Northwestern Polytechnical University), Lei Xie (Hong Kong University of Science and Technology)
Data SynthesisTransformerLarge Language ModelSupervised Fine-TuningMultimodalityBenchmarkAudio
🎯 What it does: Built the WenetSpeech-Pipe data processing pipeline, and used this pipeline to collect and annotate 21,800 hours of Cantonese speech data, generating the WenetSpeech-Yue large-scale corpus, while releasing the WSYue-eval benchmark set covering ASR and TTS evaluations.
What Makes a Good Generated Image? Investigating Human and Multimodal LLM Image Preference Alignment
Rishab Parthasarathy (MIT), Cory Stephenson (Databricks Mosaic AI Research)
GenerationLarge Language ModelVision Language ModelDiffusion modelImageMultimodality
🎯 What it does: Investigate differences in preferences and relationships between humans and multimodal large language models (LLMs) when evaluating generated images for various quality attributes, constructing a paired image assessment dataset that includes aesthetics, distortion, anatomy, composition, object adherence, style, and overall scores, and analyzing attribute correlations through human-LLM comparative analysis;
What Makes a Good Speech Tokenizer for LLM-Centric Speech Generation? A Systematic Study
Xiaoran Fan (Fudan University), Tao Gui (Fudan University)
GenerationTransformerLarge Language ModelSupervised Fine-TuningTextAudio
🎯 What it does: Systematically investigates the impact of speech tokenizer design in LLM-centric speech generation on cross-modal alignment and speech quality, and proposes a multi-word prediction (MTP) and speaker-aware generation scheme.
What to Ask Next? Probing the Imaginative Reasoning of LLMs with TurtleSoup Puzzles
Mengtao Zhou (Huazhong University of Science and Technology), Bang Liu (Huazhong University of Science and Technology)
Large Language ModelTextBenchmarkChain-of-Thought
🎯 What it does: This study proposes an interactive benchmark called TurtleSoup-Bench centered on turtle soup riddles, and constructs a multi-stage Mosaic-Agent model to evaluate the imaginative reasoning capabilities of large language models in information-scarce environments.
What to Trust? A Trust-aware Knowledge-guided Method for Zero-shot Object State Understanding in Videos
Yayun Qi (Beijing Institute of Technology), Xinxiao Wu (Beijing Institute of Technology)
RecognitionTransformerLarge Language ModelVideoTextRetrieval-Augmented Generation
🎯 What it does: Achieve zero-shot multi-state object state recognition in videos using knowledge-guided voting and credibility modeling.
What Voting Rules Actually Do: A Data-Driven Analysis of Multi-Winner Voting
Joshua Caiata (University of Waterloo), Kate Larson (University of Waterloo)
Tabular
🎯 What it does: This paper proposes a data-driven framework that evaluates the frequency with which multi-winner voting rules violate axioms under various preference distributions using the average violation rate (AVR), and learns new rules via neural networks to minimize violations.
What You See Is What You Reach: Towards Spatial Navigation with High-Level Human Instructions
Lingfeng Zhang (Tsinghua University), Wenbo Ding (Institute of Automation, Chinese Academy of Sciences)
Autonomous DrivingRobotic IntelligenceReinforcement Learning from Human FeedbackTransformerVision Language ModelVision-Language-Action ModelSimultaneous Localization and MappingMultimodality
🎯 What it does: This paper proposes spatial navigation tasks (SpON+SpAN) and constructs a 10K trajectory dataset, introducing the SpNav hierarchical framework: utilizing VLM for instruction understanding, NaviPoint for target point localization, and Map-to-Action for map construction and path planning, achieving precise navigation based on high-level human instructions.
What-Meets-Where: Unified Learning of Action and Contact Localization in Images
Yuxiao Wang (South China University of Technology), Qi Liu (South China University of Technology)
RecognitionSegmentationTransformerImageBenchmark
🎯 What it does: Jointly recognize action categories and body part contact regions in images, achieving simultaneous action classification and contact segmentation.
What, Whether and How? Unveiling Process Reward Models for Thinking with Images Reasoning
Yujin Zhou (Hong Kong University of Science and Technology), Sirui Han (Hong Kong University of Science and Technology)
Large Language ModelVision Language ModelMultimodalityBenchmark
🎯 What it does: This paper proposes an evaluation of the Process Reward Model (PRM) under the 'Thinking with Images' paradigm, constructing the first dedicated benchmark, ThinkWithImages-PRMBENCH, with systematic experiments.
When Eyes and Ears Disagree: Can MLLMs Discern Audio-Visual Confusion?
Qilang Ye (Nankai University), Yu Zhou (Nankai University)
TransformerLarge Language ModelReinforcement LearningVision Language ModelVideoMultimodalityBenchmarkAudio
🎯 What it does: Proposed the AVConfuseBench audio-visual confusion benchmark and designed the RL-CoMM method to enhance the reasoning and answer accuracy of multi-modal large language models in scenarios where audio is missing or tampered with.
When Genes Speak: A Semantic-Guided Framework for Spatially Resolved Transcriptomics Data Clustering
Jiangkai Long (China University of Geosciences), Xuesong Yan (China University of Geosciences)
ClassificationGraph Neural NetworkLarge Language ModelBiomedical Data
🎯 What it does: Propose the SemST framework, integrating the semantic embeddings of gene symbols with spatial graph neural networks to achieve clustering of spatial transcriptomic data.
When Instinct Guides and Insight Grounds: Staged RL Training for LLM Agents
Zijing Zhang (Peking University), Boning Zhang (Institute of Automation, Chinese Academy of Sciences)
TransformerLarge Language ModelReinforcement LearningAgentic AITextBenchmark
🎯 What it does: Designed and implemented a two-phase RL training framework called ActRe, where the LLM first takes actions then reasons to promote exploration, and then reverts to the traditional reasoning-action order to maintain interpretability, performing end-to-end reinforcement learning for LLM agents on ALFWorld and WebShop environments.
When Natural Strategies Meet Fuzziness and Resource-Bounded Actions
Marco Aruta (University of Naples Federico II), Aniello Murano (University of Naples Federico II)
Explainability and InterpretabilityReinforcement LearningTabular
🎯 What it does: This paper proposes the HumanATL[F] logic, combining natural strategies with fuzzy semantics and consumable resource constraints to construct interpretable and budget-constrained multi-agent strategies.
When Person Re-Identification Meets Event Camera: A Benchmark Dataset and an Attribute-Guided Re-Identification Framework
Xiao Wang (Anhui University), Shiliang Zhang (Peng Cheng Laboratory)
RecognitionRetrievalTransformerVision Language ModelContrastive LearningImageMultimodalityBenchmark
🎯 What it does: This paper proposes a new large-scale RGB-event camera person re-identification dataset called EvReID, and designs an attribute-guided tri-modal contrastive learning framework named TriProReID based on this dataset.
When Privacy Meets Recovery: The Overlooked Half of Surrogate-Driven Privacy Preservation for MLLM Editing
Siyuan Xu, Sam Kwong (City University of Hong Kong)
RestorationSafty and PrivacyTransformerVision Language ModelDiffusion modelAuto EncoderMultimodalityBenchmark
🎯 What it does: Developed a privacy-preserving framework based on proxy images, capable of restoring editing results of multi-modal large language models without uploading original images.