AAAI 2025 Papers — Page 30
AAAI Conference on Artificial Intelligence · 3028 papers
Unsupervised Domain Adaptive Person Search via Dual Self-Calibration
Linfeng Qi (Dalian Maritime University), Yang Wang (Hefei University of Technology)
RecognitionObject DetectionDomain AdaptationConvolutional Neural NetworkImage
🎯 What it does: A dual self-calibration framework is proposed to eliminate the interference of pseudo-label noise in unsupervised domain adaptation for person search.
Unsupervised Kernel-based Multi-view Feature Selection with Robust Self-representation and Binary Hashing
Rongyao Hu (University of Electronic Science and Technology of China), Mengling Wei (University of Electronic Science and Technology of China)
Multimodality
🎯 What it does: A novel unsupervised multi-view feature selection framework UKMFS based on kernel mapping, robust self-representation, and binary hashing is proposed.
Unsupervised Photometric-Consistent Depth Estimation from Endoscopic Monocular Video
Shijie Li (Sichuan University), Zheng Li (University of Chinese Academy of Sciences)
Depth EstimationConvolutional Neural NetworkVideoComputed Tomography
🎯 What it does: This paper proposes an unsupervised endoscopic monocular video depth estimation method called PC-Depth, which improves depth prediction accuracy by addressing the issue of inconsistent lighting.
Unsupervised Region-Based Image Editing of Denoising Diffusion Models
Zixiang Li (Beijing Jiaotong University), Wei Wang (Beijing Jiaotong University)
GenerationDiffusion modelImage
🎯 What it does: An unsupervised Region-Based Editing (RBE) method is proposed, which utilizes Jacobian projection to identify and control local semantics in pre-trained diffusion models without additional training or supervision.
Unsupervised Self-Prior Embedding Neural Representation for Iterative Sparse-View CT Reconstruction
Xuanyu Tian (ShanghaiTech University), Yuyao Zhang (Shanghai Jiao Tong University)
RestorationImageBiomedical DataComputed Tomography
🎯 What it does: An iterative self-supervised sparse view CT reconstruction method called Spener is proposed, which does not require external training data. It utilizes the prior features of the image from the previous iteration embedded in an implicit neural representation network to achieve robust reconstruction in extremely under-sampled and noisy scenarios.
Unsupervised Translation of Emergent Communication
Ido Levy (Technion Israel Institute of Technology), Yonatan Belinkov (Technion Israel Institute of Technology)
GenerationData SynthesisTransformerReinforcement LearningImageTextMultimodality
🎯 What it does: This paper utilizes unsupervised neural machine translation technology to translate communication protocols that spontaneously emerge in multi-agent games into natural language, without relying on parallel corpora.
Unveiling Multi-View Anomaly Detection: Intra-view Decoupling and Inter-view Fusion
Kai Mao (Xi'an Jiaotong University), Ping Wei (Xi'an Jiaotong University)
Anomaly DetectionKnowledge DistillationConvolutional Neural NetworkImage
🎯 What it does: A multi-view anomaly detection framework IDIF is proposed, which achieves anomaly recognition for multi-view data through two steps: view decoupling and view fusion.
Unveiling the Impact of Coding Data Instruction Fine-Tuning on Large Language Models Reasoning
Xinlu Zhang (University of California), Linda Ruth Petzold (University of California)
AI Code AssistantTransformerLarge Language ModelSupervised Fine-TuningText
🎯 What it does: Conducted instruction fine-tuning (IFT) on six major LLMs (Llama-1/2/3, Mistral, Qwen-1.5, Gemma), using ShareGPT data and classifying conversations into 'code' or 'non-code' through ChatGPT, constructing IFT datasets with 0%, 50%, and 100% code ratios; subsequently performed zero-shot evaluation on 12 reasoning tasks (symbolic, logical, arithmetic), systematically analyzing the impact of different code ratios, model families, and scales on overall, domain-specific, and task-specific reasoning performance.
Unveiling the Knowledge of CLIP for Training-Free Open-Vocabulary Semantic Segmentation
Yajie Liu (Beihang University), Di Huang (Beihang University)
SegmentationTransformerVision Language ModelContrastive LearningImage
🎯 What it does: Achieve open vocabulary semantic segmentation using a frozen CLIP vision-language model without any fine-tuning;
Unveiling the Threat of Fraud Gangs to Graph Neural Networks: Multi-Target Graph Injection Attacks Against GNN-Based Fraud Detectors
Jinhyeok Choi (KAIST), Joyce Jiyoung Whang (KAIST)
Anomaly DetectionAdversarial AttackGraph Neural NetworkTransformerGraph
🎯 What it does: This paper studies multi-target graph injection attacks on fraud detectors based on graph neural networks and proposes a new attack model, MonTi, to inject attack nodes at once and induce fraudulent nodes to be misclassified as normal.
UP-Restorer: When Unrolling Meets Prompts for Unified Image Restoration
Minghao Liu (Peking University), Jiaying Liu (Peking University)
RestorationTransformerPrompt EngineeringDiffusion modelScore-based ModelImage
🎯 What it does: A unified image restoration framework named Up-Restorer is proposed, which utilizes prompt generation combined with ADMM solving to achieve single-model restoration for various degradations (rain, fog, noise, etc.).
UrBench: A Comprehensive Benchmark for Evaluating Large Multimodal Models in Multi-View Urban Scenarios
Baichuan Zhou (Shanghai AI Laboratory), Weijia Li (Sun Yat-Sen University)
Object DetectionRetrievalTransformerLarge Language ModelImageMultimodalityBenchmark
🎯 What it does: This paper presents UrBench, a multi-view, multi-task urban scene evaluation benchmark designed to test the performance of large-scale multimodal models (LMM) in urban environments.
USDRL: Unified Skeleton-Based Dense Representation Learning with Multi-Grained Feature Decorrelation
Wanjiang Weng (Southeast University), Guo-Sen Xie (Nanjing University of Science and Technology)
RecognitionRetrievalRepresentation LearningConvolutional Neural NetworkContrastive LearningVideo
🎯 What it does: A USDRL framework is proposed, utilizing multi-layer feature decorrelation to achieve dense representation learning of skeleton sequences.
User Preference Meets Pareto-Optimality in Multi-Objective Bayesian Optimization
Joshua Hang Sai Ip (University of California), Diego Romeres (Mitsubishi Electric Research Laboratories)
Optimization
🎯 What it does: A multi-objective Bayesian optimization method called PUB-MOBO is proposed, which combines user preferences with local multi-gradient descent.
Utilize the Flow Before Stepping into the Same River Twice: Certainty Represented Knowledge Flow for Refusal-Aware Instruction Tuning
Runchuan Zhu (Shanghai Artificial Intelligence Laboratory), Conghui He (Shanghai Artificial Intelligence Laboratory)
TransformerLarge Language ModelSupervised Fine-TuningText
🎯 What it does: To address the phenomenon of over-refusal in large language models during the RAIT process, the CRaFT method is proposed, which reduces static and dynamic conflicts by incorporating answer confidence screening and rehearsal training, thereby enhancing the reliability of the model's refusals.
Utterance-level Emotion Recognition in Conversation with Conversation-level Supervision
Ximing Li (Jilin University), Lin Yuanbo Wu (Swansea University)
RecognitionGraph Neural NetworkTransformerText
🎯 What it does: This paper proposes a weakly supervised learning framework for dialogue emotion recognition (ERC) called DERC-PL, which utilizes only conversation-level emotion sets and employs pseudo-label self-training and a gradual noise-injection learning strategy.
V2C-CBM: Building Concept Bottlenecks with Vision-to-Concept Tokenizer
Hangzhou He (Peking University), Yanye Lu (Peking University)
ClassificationExplainability and InterpretabilityComputational EfficiencyImage
🎯 What it does: This study proposes V2C-CBM, which utilizes a Vision-to-Concept (V2C) tokenizer to directly generate visual concepts from unlabeled images, constructing a concept bottleneck model to achieve interpretable image classification without the need for LLMs.
V2Xum-LLM: Cross-Modal Video Summarization with Temporal Prompt Instruction Tuning
Hang Hua (University of Rochester), Jiebo Luo (University of Rochester)
GenerationData SynthesisTransformerLarge Language ModelPrompt EngineeringVideoTextMultimodality
🎯 What it does: A cross-modal video summarization framework V2Xum-LLM is proposed, which can achieve video-to-video, video-to-text, and video+text summarization through a unified language model decoder. Additionally, an Instruct-V2Xum cross-modal summarization dataset covering 30,000 YouTube videos across multiple domains has been constructed. Evaluation metrics based on CLIP, namely F-CLIP and Cross-F-CLIP, are also introduced.
VA-AR: Learning Velocity-Aware Action Representations with Mixture of Window Attention
Jiangning Wei (Beijing University of Posts and Telecommunications), Jun Liu (Beijing University of Posts and Telecommunications)
RecognitionPose EstimationGraph Neural NetworkTransformerMixture of ExpertsVideoMultimodality
🎯 What it does: This study investigates the impact of speed on skeleton action recognition and proposes the Velocity-Aware Action Recognition (VA-AR) framework to achieve robust recognition of actions at different speeds.
VarCMP: Adapting Cross-Modal Pre-Training Models for Video Anomaly Retrieval
Peng Wu (Northwestern Polytechnical University), Yanning Zhang (Northwestern Polytechnical University)
RetrievalAnomaly DetectionContrastive LearningVideoTextMultimodalityAudio
🎯 What it does: A VarCMP method based on cross-modal pre-trained models (CLIP/CLAP) is proposed for video anomaly retrieval, achieving video-text and video-audio retrieval through a unified hierarchical alignment strategy and anomaly bias weighting mechanism.
VarDrop: Enhancing Training Efficiency by Reducing Variate Redundancy in Periodic Time Series Forecasting
Junhyeok Kang (LG AI Research), Jae-Gil Lee (Korea Advanced Institute of Science and Technology)
TransformerTime Series
🎯 What it does: A strategy called VarDrop is proposed to dynamically eliminate redundant variables during the training of the mutation-worded Transformer, allowing the model to use only the necessary variables to make attention calculations more efficient.
VCR: A “Cone of Experience” Driven Synthetic Data Generation Framework for Mathematical Reasoning
Sannyuya Liu (Central China Normal University), Jianwen Sun (Central China Normal University)
GenerationData SynthesisTransformerLarge Language ModelSupervised Fine-TuningText
🎯 What it does: A multi-agent virtual classroom (VCR) framework driven by LLM has been established, utilizing the three types of experiences—direct, symbolic, and iconic—from the human learning theory 'Cone of Experience' to simulate roles such as teacher, student, teaching assistant, and recorder. It generates high-quality and diverse mathematical reasoning synthesis data through teaching activities like lecturing, discussion, and reflection, and enhances data quality through adaptive weighting and iteration.
VE-Bench: Subjective-Aligned Benchmark Suite for Text-Driven Video Editing Quality Assessment
Shangkun Sun (Peking University), Wei Gao (Peking University)
GenerationData SynthesisOptimizationConvolutional Neural NetworkVision Language ModelVideoTextMultimodalityBenchmark
🎯 What it does: A VE-Bench benchmark has been constructed, which includes a subjective alignment database (VE-Bench DB) specifically designed for text-driven video editing and a quantitative evaluation network (VE-Bench QA) based on this database, aimed at comprehensive assessment of the quality of edited videos.
VEGAS: Towards Visually Explainable and Grounded Artificial Social Intelligence
Hao Li (Wuhan University), Zheng Wang (Wuhan University)
Explainability and InterpretabilityTransformerLarge Language ModelSupervised Fine-TuningVision Language ModelVideoTextMultimodalityAudio
🎯 What it does: Proposes the VEGAS framework, which combines visual interpretability with language-guided frame sampling to generate interpretable multimodal answers.
Verifying Proportionality in Temporal Voting
Edith Elkind (Northwestern University), Nicholas Teh (University of Oxford)
🎯 What it does: This paper studies the verification problem of proportional representation under the time voting model, proving that it is coNP-hard under various representational axioms, and provides several solvable special cases and parameterized algorithms; at the same time, it proposes a two-stage greedy consistency rule that can produce results satisfying EJR, along with a corresponding ILP scheme.
VerilogCoder: Autonomous Verilog Coding Agents with Graph-based Planning and Abstract Syntax Tree (AST)-based Waveform Tracing Tool
Chia-Tung Ho (NVIDIA Research), Brucek Khailany (NVIDIA Research)
AI Code AssistantLarge Language ModelAgentic AITextBenchmark
🎯 What it does: This paper presents VerilogCoder, a multi-agent system capable of automatically generating Verilog code and fixing functional errors through syntax checking, simulation, and a novel AST waveform tracing tool.
VERO: Verification and Zero-Shot Feedback Acquisition for Few-Shot Multimodal Aspect-Level Sentiment Classification
Kai Sun (Xi'an Jiaotong University), Bo Dong (Xi'an Jiaotong University)
ClassificationTransformerLarge Language ModelPrompt EngineeringVision Language ModelTextMultimodality
🎯 What it does: A self-verification-based sample collection method called VERO was designed and implemented to select challenging samples from unlabeled multimodal data, followed by few-shot fine-tuning of large visual-language models (LLaVA-7b/13b) to complete the multimodal aspect-level sentiment classification task.
VersaFusion: A Versatile Diffusion-Based Framework for Fine-Grained Image Editing and Enhancement
Haocun Ye (Institute of Computing Technology, Chinese Academy of Sciences), Yiqiang Chen
Image TranslationGenerationDiffusion modelScore-based ModelImage
🎯 What it does: A dual-branch framework called VersaFusion based on a pre-trained diffusion model is proposed for fine-grained image editing and enhancement, supporting various operations such as drag-and-drop editing, object transfer, size adjustment, and style transfer.
VersaGen: Unleashing Versatile Visual Control for Text-to-Image Synthesis
Zhipeng Chen (Beijing University of Posts and Telecommunications), Yi-Zhe Song (University of Surrey)
GenerationData SynthesisDiffusion modelImageText
🎯 What it does: Proposes VersaGen, a control framework for text-to-image synthesis that can be guided by user-drawn inputs at multiple levels.
VERSE: Verification-based Self-Play for Code Instructions
Hao Jiang (University of Science and Technology of China), Yu Su (Hefei Normal University)
OptimizationAI Code AssistantTransformerLarge Language ModelSupervised Fine-TuningPrompt EngineeringText
🎯 What it does: The VERSE method is proposed, which utilizes code LLMs to perform code verification during the self-generation of instructions, responses, and validation scripts, and selects high-quality data for self-fine-tuning based on execution results and self-consistency scores.
VFM-Adapter: Adapting Visual Foundation Models for Dense Prediction with Dynamic Hybrid Operation Mapping
Zheng Chen (University of Science and Technology of China), Feng Zhao (University of Science and Technology of China)
Object DetectionSegmentationTransformerSupervised Fine-TuningImage
🎯 What it does: To address the adaptation problem of Vision Foundation Models (VFM) in dense prediction tasks (such as object detection, instance segmentation, and semantic segmentation), a novel parameter-efficient fine-tuning method called VFM-Adapter is proposed.
VG-TVP: Multimodal Procedural Planning via Visually Grounded Text-Video Prompting
Muhammet Furkan Ilaslan (National University of Singapore), Qianli Xu (Institute for Infocomm Research, Agency for Science, Technology, and Research)
GenerationData SynthesisTransformerLarge Language ModelPrompt EngineeringDiffusion modelVideoTextMultimodality
🎯 What it does: This paper proposes a multimodal program planning framework called VG‑TVP based on large language models, which can generate coherent text and video program steps according to user-defined task objectives, addressing the limitations of traditional unimodal planning.
VHM: Versatile and Honest Vision Language Model for Remote Sensing Image Analysis
Chao Pang (Wuhan University), Conghui He (Shanghai Artificial Intelligence Laboratory)
ClassificationRecognitionTransformerLarge Language ModelSupervised Fine-TuningVision Language ModelImageTextMultimodality
🎯 What it does: A multi-task visual language model VHM aimed at remote sensing image analysis has been developed, and it achieves multifunctional understanding and honest responses by combining a newly constructed multi-content description dataset VersaD and a truthfulness instruction set HnstD containing both factual and deceptive questions.
VidChain: Chain-of-Tasks with Metric-based Direct Preference Optimization for Dense Video Captioning
Ji Soo Lee (Korea University), Hyunwoo J. Kim (Korea University)
GenerationOptimizationTransformerLarge Language ModelSupervised Fine-TuningVideoTextMultimodality
🎯 What it does: In the task of dense video captioning, improvements are made to existing large video models by proposing to decompose the task into multiple sub-tasks (Chain-of-Tasks) and utilizing Metric-based Direct Preference Optimization to enhance the model's fine-grained temporal understanding capabilities.
Video Anomaly Detection with Motion and Appearance Guided Patch Diffusion Model
Hang Zhou (Huazhong University of Science and Technology), Wei Yang (Huazhong University of Science and Technology)
Anomaly DetectionDiffusion modelVideo
🎯 What it does: This paper studies a patch-based diffusion model for video anomaly detection using motion and appearance conditions.
Video Diffusion Models Are Strong Video Inpainter
Minhyeok Lee (Yonsei University), Sangyoun Lee (Yonsei University)
RestorationGenerationDiffusion modelOptical FlowVideo
🎯 What it does: A first-frame filling-based image-to-video diffusion model FFF-VDI is proposed for video inpainting and object removal tasks.
Video Repurposing from User Generated Content: A Large-scale Dataset and Benchmark
Yongliang Wu (Southeast University), Xu Yang (Opus AI Research)
GenerationRetrievalTransformerLarge Language ModelVideoMultimodalityBenchmark
🎯 What it does: A research framework for video repurposing tasks is proposed, and a large-scale user-generated content (UGC) video repurposing dataset, Repurpose-10K, is constructed. An end-to-end baseline model based on a multi-modal Transformer is developed, capable of automatically generating short video clips of about 60 seconds from long videos.
Video Summarization Using Denoising Diffusion Probabilistic Model
Zirui Shang (Beijing Institute of Technology), Xinxiao Wu (Shenzhen MSU-BIT University)
GenerationData SynthesisTransformerDiffusion modelVideo
🎯 What it does: A generative video summarization method based on Denoising Diffusion Probabilistic Model (DDPM) is proposed, which generates video summaries by iteratively denoising frame importance scores through noise prediction.
VideoElevator: Elevating Video Generation Quality with Versatile Text-to-Image Diffusion Models
Yabo Zhang (Harbin Institute of Technology), Wangmeng Zuo (Harbin Institute of Technology)
GenerationData SynthesisDiffusion modelVideoTextBenchmarkStochastic Differential Equation
🎯 What it does: A training-agnostic, plug-and-play method called VideoElevator is proposed, which utilizes existing text-to-image diffusion models to enhance the quality of text-to-video diffusion models. This method explicitly splits each sampling step into two sub-steps: temporal motion refinement and spatial quality enhancement.
VidEvent: A Large Dataset for Understanding Dynamic Evolution of Events in Videos
Baoyu Liang (Beihang University), Chao Tong (Beihang University)
ClassificationRecognitionRetrievalTransformerContrastive LearningVideoTextMultimodality
🎯 What it does: This paper proposes a video event understanding task and constructs a large-scale VidEvent dataset containing over 23,000 events and 17,000 relationships, providing a four-step baseline framework (event localization, extraction, relationship classification, and script reasoning) for future research.
Vietnamese Words Are Not Constructed from Syllables: Rethinking the Role of Word Segmentation in Natural Language Processing for Vietnamese Texts
Nghia Hieu Nguyen (Vietnam National University), Ngan Luu-Thuy Nguyen (Vietnam National University)
TransformerText
🎯 What it does: This paper studies the theory of word formation in Vietnamese, proposing the ViWordFormer model to capture the vocabulary and phrase structure of Vietnamese at the word level without the need for word segmentation.
View Transformation Robustness for Multi-View 3D Object Reconstruction with Reconstruction Error-Guided View Selection
Qi Zhang (Shenzhen University), Hui Huang (Shenzhen University)
Object DetectionSegmentationGenerationData SynthesisOptimizationDiffusion modelPoint Cloud
🎯 What it does: A perspective selection method based on reconstruction error guidance is proposed, and new perspective images are synthesized using the Stable Diffusion model to enhance the robustness of multi-view 3D reconstruction models under perspective transformations.
ViFactCheck: A New Benchmark Dataset and Methods for Multi-Domain News Fact-Checking In Vietnamese
Tran Thai Hoa (University of Information Technology), Kiet Van Nguyen (University of Information Technology)
TransformerLarge Language ModelSupervised Fine-TuningPrompt EngineeringTextBenchmark
🎯 What it does: This study first constructed the ViFactCheck dataset, a Vietnamese multi-domain fact-checking benchmark translated from Chinese, and conducted model training and evaluation on this dataset.
ViG: Linear-complexity Visual Sequence Learning with Gated Linear Attention
Bencheng Liao (Huazhong University of Science and Technology), Chang Huang (Horizon Robotics)
ClassificationObject DetectionSegmentationTransformerVision Language ModelImage
🎯 What it does: ViG proposes a visual sequence learning framework with linear complexity.
VIoTGPT: Learning to Schedule Vision Tools Towards Intelligent Video Internet of Things
Yaoyao Zhong (Beijing University of Posts and Telecommunications), Huadong Ma (Beijing University of Posts and Telecommunications)
RecognitionObject DetectionPose EstimationTransformerLarge Language ModelSupervised Fine-TuningPrompt EngineeringVideoText
🎯 What it does: A framework called VIoTGPT based on large language models is proposed, which can automatically invoke video tools according to human queries, enabling unified and intelligent analysis of large-scale videos in the Video Internet of Things (VIoT) and providing responses.
ViPCap: Retrieval Text-Based Visual Prompts for Lightweight Image Captioning
Taewhan Kim (Hanyang University), Dong-Jin Kim (Hanyang University)
GenerationRetrievalTransformerPrompt EngineeringVision Language ModelImageText
🎯 What it does: A lightweight image description method called ViPCap is proposed, which utilizes retrieved text to generate visual prompts to enhance image description performance.
ViPOcc: Leveraging Visual Priors from Vision Foundation Models for Single-View 3D Occupancy Prediction
Yi Feng (Tongji University), Rui Fan (Tongji University)
Depth EstimationAutonomous DrivingTransformerNeural Radiance FieldPoint Cloud
🎯 What it does: Proposes the ViPOcc framework, which combines visual priors to achieve single-view 3D occupancy prediction and depth estimation.
Virtual Nodes Can Help: Tackling Distribution Shifts in Federated Graph Learning
Xingbo Fu (University of Virginia), Jundong Li (University of Virginia)
Federated LearningGraph Neural NetworkContrastive LearningGraph
🎯 What it does: The FedVN framework is proposed, which eliminates the shift caused by inconsistent data distribution in federated graph learning by learning adaptive graph enhancement strategies for each client (using shared multiple virtual nodes and personalized edge generators), thereby achieving efficient training of the global GNN model.
Vision Transformers Beat WideResNets on Small Scale Datasets Adversarial Robustness
Juntao Wu (Jinan University), Ke Wang (Jinan University)
ClassificationAdversarial AttackTransformerDiffusion modelImage
🎯 What it does: This paper explores and implements adversarial training using data generated by Vision Transformer (ViT) combined with diffusion models, achieving higher robustness and accuracy than WideResNet on small-scale datasets.
Vision-aware Multimodal Prompt Tuning for Uploadable Multi-source Few-shot Domain Adaptation
Kuanghong Liu (Yunnan University), Xuejie Zhang (Yunnan University)
Domain AdaptationPrompt EngineeringMultimodality
🎯 What it does: This paper proposes an Uploadable Multi-source Few-shot Domain Adaptation (UMFDA) framework and designs a Visual Perception Multi-modal Prompt Tuning (VAMP) scheme to achieve collaborative transfer across multiple source domains in low-computation, low-annotation environments on edge devices.
Vision-Based Generic Potential Function for Policy Alignment in Multi-Agent Reinforcement Learning
Hao Ma (University of Chinese Academy of Sciences), Xiaolin Ai (Chinese Academy of Sciences)
TransformerReinforcement LearningVision Language ModelVideo
🎯 What it does: A hierarchical visual-based potential function reward shaping method V-GEPF is proposed, which utilizes visual language models (CLIP) and visual large language models (MiniCPM) to align policies with human common sense in multi-agent reinforcement learning.
Vision-guided Text Mining for Unsupervised Cross-modal Hashing with Community Similarity Quantization
Haozhi Fan (University of Pennsylvania), Yuan Cao (Ocean University of China)
Object DetectionRetrievalOptimizationVision Language ModelContrastive LearningImageTextMultimodality
🎯 What it does: This paper proposes an unsupervised cross-modal hashing method VTM-UCH based on visually guided text mining, which enhances text semantics using CLIP and object detection, and optimizes hash distribution through community detection.
VisRec: A Semi-Supervised Approach to Visibility Data Reconstruction in Radio Astronomy
Ruoqi Wang (Hong Kong University of Science and Technology), Hejun Wu (Guangzhou University)
RestorationConvolutional Neural NetworkSupervised Fine-TuningNeural Radiance FieldImagePhysics Related
🎯 What it does: We propose VisRec, a model-agnostic semi-supervised learning framework for the reconstruction of visibility data from sparse to dense in radio interferometry.
Visual Perturbation for Text-Based Person Search
Pengcheng Zhang (Beihang University), Jin Zheng (Beihang University)
RecognitionRetrievalTransformerVision Language ModelContrastive LearningImageText
🎯 What it does: This paper proposes a Visual Perturbation Network (ViPer) that improves the alignment of visual and linguistic features for the Text-Based Person Search (TBPS) task.
Visual Prompting Upgrades Neural Network Sparsification: A Data-Model Perspective
Can Jin (Rutgers University), Tianlong Chen (University of North Carolina at Chapel Hill)
OptimizationConvolutional Neural NetworkPrompt EngineeringImage
🎯 What it does: Jointly learning visual prompts and weight sparsification on pre-trained visual models to construct the VPNs framework.
Visual Reinforcement Learning with Residual Action
Zhenxian Liu (Peking University), Yonghong Tian (Peking University)
Autonomous DrivingConvolutional Neural NetworkReinforcement LearningImage
🎯 What it does: A framework called ResAct for residual action learning and observation difference learning in visual RL is proposed to simplify action learning.
VLScene: Vision-Language Guidance Distillation for Camera-Based 3D Semantic Scene Completion
Meng Wang (Hunan University), Kenli Li (Hunan University)
SegmentationKnowledge DistillationTransformerVision Language ModelImagePoint Cloud
🎯 What it does: Guided by a visual language model, distillation enhances 3D semantic scene completion using a monocular camera.
VOILA: Complexity-Aware Universal Segmentation of CT Images by Voxel Interacting with Language
Zishuo Wan (University of Science and Technology Beijing), Dawei Ding (University of Science and Technology Beijing)
SegmentationConvolutional Neural NetworkContrastive LearningImageBiomedical DataComputed Tomography
🎯 What it does: This paper presents VOILA, a general CT image segmentation method based on voxel-language alignment, utilizing variable voxel sampling and contrastive learning to achieve multi-class segmentation.
Voter Priming Campaigns: Strategies, Equilibria, and Algorithms
Jonathan Shaki (Bar Ilan University), Sarit Kraus (Bar Ilan University)
Optimization
🎯 What it does: This paper constructs a model of voter priming and analyzes the strategies, Nash equilibrium, and optimal investment allocation of candidates in multi-party, multi-issue elections.
Vox-UDA: Voxel-wise Unsupervised Domain Adaptation for Cryo-Electron Subtomogram Segmentation with Denoised Pseudo-Labeling
Haoran Li (University of Wollongong), Min Xu (Carnegie Mellon University)
SegmentationDomain AdaptationKnowledge DistillationConvolutional Neural NetworkImage
🎯 What it does: Proposes the Vox-UDA voxel-level unsupervised domain adaptation framework for segmentation tasks on unlabeled cryo-ET subtomograms.
VProChart: Answering Chart Question Through Visual Perception Alignment Agent and Programmatic Solution Reasoning
Muye Huang (Xi'an Jiaotong University), Jun Liu (Xi'an Jiaotong University)
TransformerLarge Language ModelAgentic AIImageText
🎯 What it does: The VProChart framework is proposed, which combines a lightweight visual alignment agent (VPAgent) with LLM-based programmatic solution reasoning for chart question-answering tasks.
VQ4DiT: Efficient Post-Training Vector Quantization for Diffusion Transformers
Juncan Deng (Zhejiang University), Kejie Huang (Vivo Mobile Communication Co., Ltd)
GenerationCompressionTransformerDiffusion modelImage
🎯 What it does: A post-training vector quantization method named VQ4DiT is proposed to compress Diffusion Transformers (DiT) models to extremely low bit widths (2-bit), achieving significant reductions in memory and inference time.
VQA4CIR: Boosting Composed Image Retrieval with Visual Question Answering
Chun-Mei Feng (Institute of High Performance Computing Agency for Science Technology and Research), Yong Liu (Harbin Institute of Technology)
RetrievalTransformerLarge Language ModelSupervised Fine-TuningVision Language ModelImageTextMultimodality
🎯 What it does: This paper proposes a post-processing framework called VQA4CIR, which first generates QA pairs related to relative descriptions using LLaMA, and then performs visual question answering (VQA) with Fine-tuned LLaVA to self-validate the consistency of the retrieval results with the descriptions, and accordingly re-ranks the retrieval list to improve the performance of synthesized image retrieval.
VQTalker: Towards Multilingual Talking Avatars Through Facial Motion Tokenization
Tao Liu (Shanghai Jiao Tong University), Kai Yu (AISpeech Ltd)
GenerationData SynthesisLarge Language ModelSupervised Fine-TuningVideoAudio
🎯 What it does: A multilingual facial animation generation framework called VQTalker is proposed, which converts audio tokens into discrete facial motion tokens to achieve realistic lip synchronization and expressions.
VRVVC: Variable-Rate NeRF-Based Volumetric Video Compression
Qiang Hu (Shanghai Jiao Tong University), Yanfeng Wang (Shanghai Jiao Tong University)
CompressionNeural Radiance FieldVideo
🎯 What it does: A variable bitrate NeRF-based volumetric video compression framework VRVVC is proposed, which jointly optimizes three-plane residual representation and variable rate entropy coding, and employs an end-to-end progressive (coarse-to-fine) training strategy.
VTG-LLM: Integrating Timestamp Knowledge into Video LLMs for Enhanced Video Temporal Grounding
Yongxin Guo (Chinese University of Hong Kong), Kevin Zhao (Tencent)
RecognitionRetrievalOptimizationTransformerLarge Language ModelVision Language ModelVideoText
🎯 What it does: This paper proposes VTG-LLM, which significantly improves the zero-shot performance of video large language models in video temporal localization tasks by integrating timestamp knowledge.
VVRec: Reconstruction Attacks on DL-based Volumetric Video Upstreaming via Latent Diffusion Model with Gamma Distribution
Rui Lu (Hong Kong Polytechnic University), Dan Wang (Hong Kong Polytechnic University)
RestorationCompressionAdversarial AttackDiffusion modelAuto EncoderVideoPoint Cloud
🎯 What it does: This paper proposes a reconstruction attack framework named VVRec, which can intercept intermediate results from deep learning-compressed volumetric videos to recover the original point cloud.
Walk Wisely on Graph: Knowledge Graph Reasoning with Dual Agents via Efficient Guidance-Exploration
Zijian Wang (China University of Petroleum), Hongbo Dou (China University of Petroleum)
OptimizationRecurrent Neural NetworkGraph Neural NetworkReinforcement LearningAgentic AIGraph
🎯 What it does: This paper proposes a multi-hop reasoning model for knowledge graphs based on dual-agent hierarchical reinforcement learning—FULORA. It utilizes a high-level agent, GIANT, to quickly provide phase guidance on a clustering-level graph, while a low-level agent, DWARF, autonomously explores on an entity-level graph and balances guidance and exploration through supervised learning. Additionally, dynamic path feedback is introduced to enhance the learning efficiency of GIANT, and a graph attention mechanism helps DWARF focus on relevant neighbors.
Walking the Web of Concept-Class Relationships in Incrementally Trained Interpretable Models
Susmit Agrawal (Indian Institute of Technology Hyderabad), Vineeth N. Balasubramanian (Indian Institute of Technology Hyderabad)
ClassificationExplainability and InterpretabilityTransformerImageMultimodality
🎯 What it does: This paper studies the maintenance and expansion of concept and category relationships in the context of incremental learning, and proposes a MuCIL model that achieves multimodal concept embedding without increasing parameters.
Wasserstein Distance Constraint and Parameter Sparsification for Batched and Iterative Knowledge Editing
Shanbao Qiao (Jeonbuk National University), Seung-Hoon Na (Jeonbuk National University)
TransformerLarge Language ModelText
🎯 What it does: This study investigates the issue of parameter distribution drift leading to model performance collapse during batched iterative editing on large language models, and proposes two improvement strategies: Wasserstein distance constraint and parameter sparsification, to maintain the stability of model parameter distribution and enhance editing effectiveness.
Watch Out for Your Guidance on Generation! Exploring Conditional Backdoor Attacks against Large Language Models
Jiaming He (University of Electronic Science and Technology of China), Hongwei Li (University of Electronic Science and Technology of China)
GenerationAdversarial AttackTransformerLarge Language ModelSupervised Fine-TuningText
🎯 What it does: A backdoor attack framework for LLMs called BrieFool is proposed, which can activate the attack by specifying generation conditions (such as token limits, language, number of sentences, etc.) without changing the input text.
Watch Video, Catch Keyword: Context-aware Keyword Attention for Moment Retrieval and Highlight Detection
Sung Jin Um (Kyung Hee University), Jung Uk Kim (Kyung Hee University)
RetrievalOptimizationTransformerContrastive LearningVideoTextMultimodalityAudio
🎯 What it does: A video context-aware keyword attention module is proposed, which jointly optimizes video moment retrieval and highlight detection through keyword weighted contrastive learning.
WatE: A Wasserstein t-distributed Embedding Method for Information-enriched Graph Visualization
Minjie Cheng (Renmin University of China), Hongteng Xu (Renmin University of China)
OptimizationRepresentation LearningGraph Neural NetworkGraph
🎯 What it does: This paper proposes a t-distribution embedding method based on Wasserstein distance, called WatE, which uses GNN to learn the mean and covariance of the node embedding distribution for each graph, visualizing the graph as an ellipse, thus balancing graph-level clustering and node-level structural information.
WaterDiffusion: Learning a Prior-involved Unrolling Diffusion for Joint Underwater Saliency Detection and Visual Restoration
Laibin Chang (Wuhan University), Chang Xu (University of Sydney)
RestorationObject DetectionDiffusion modelImage
🎯 What it does: A diffusion model called WaterDiffusion is proposed, which combines seawater attenuation priors to simultaneously achieve underwater salient object detection and image restoration.
Wavelet-Assisted Multi-Frequency Attention Network for Pansharpening
Jie Huang (University of Electronic Science and Technology of China), Liang-Jian Deng
RestorationImageStochastic Differential Equation
🎯 What it does: A multi-frequency attention network based on wavelet transform, WFANet, is proposed for the fusion of high-resolution multispectral images;
Wavelet-Driven Masked Image Modeling: A Path to Efficient Visual Representation
Wenzhao Xiang (Chinese Academy of Sciences), Xilin Chen (Chinese Academy of Sciences)
Object DetectionSegmentationComputational EfficiencyRepresentation LearningTransformerImage
🎯 What it does: This paper studies a multi-scale mask image modeling framework based on wavelet transform, WaMIM, which uses wavelet coefficients as the reconstruction target to enhance the efficiency and effectiveness of MIM pre-training.
WaveletMixer: A Multi-Resolution Wavelets Based MLP-Mixer for Multivariate Long-Term Time Series Forecasting
Zichi Zhang (Queen's University Belfast), Son T. Mai (Queen's University Belfast)
TransformerTime Series
🎯 What it does: This paper proposes WaveletMixer, which combines multi-resolution wavelet decomposition with multi-stage training, specifically targeting the problem of multivariate long-term time series forecasting.
WaveLoss: An Adaptive Dynamic Loss for Deep Gait Recognition
Zicheng Wang (South China University of Technology), Qiuxia Wu (South China University of Technology)
RecognitionVideo
🎯 What it does: A dynamic adaptive loss function called WaveLoss is designed and proposed for deep gait recognition, which can adaptively focus on samples of varying difficulty during the training process.
Weak Strategyproofness in Randomized Social Choice
Felix Brandt (Technical University of Munich), Patrick Lederer (UNSW Sydney)
Score-based Model
🎯 What it does: This paper studies weak strategyproofness in random social choice and proposes the construction and analysis of corresponding random voting rules.
Weakly Supervised Gland Segmentation with Class Semantic Consistency and Purified Labels Filtration
Siyang Feng (Guilin University of Electronic Technology), Xipeng Pan (Guilin University of Electronic Technology)
SegmentationConvolutional Neural NetworkImage
🎯 What it does: A weakly supervised gland segmentation method is proposed, achieving high-precision segmentation through improved CAM generation and pseudo-label filtering.
WebPilot: A Versatile and Autonomous Multi-Agent System for Web Task Execution with Strategic Exploration
Yao Zhang (Ludwig Maximilian University Munich), Volker Tresp (Technical University of Munich)
Large Language ModelReinforcement LearningAgentic AIText
🎯 What it does: Proposes WebPilot, a multi-agent system based on dual MCTS optimization to solve complex web tasks.
Weighted Embeddings for Low-Dimensional Graph Representation
Thomas Bläsius (Karlsruhe Institute of Technology), Nikolai Maas (Karlsruhe Institute of Technology)
OptimizationRepresentation LearningGraph Neural NetworkGraph
🎯 What it does: A new weighted embedding method (WEMBED) is proposed, which assigns weights to each node and uses weighted Euclidean distance to approximate hyperbolic geometry, generating low-dimensional graph embeddings.
Weighted Poisson-disk Resampling on Large-Scale Point Clouds
Xianhe Jiao (Qingdao University), Yong-Jin Liu (Tsinghua University)
Point Cloud
🎯 What it does: This paper proposes a Weighted Poisson-Disk Resampling method (WPD) that can accurately control the number of points and achieve equidistant distribution in large-scale point clouds.
Welfare-Optimal Serial Dictatorships Have Polynomial Query Complexity
Ioannis Caragiannis (Aarhus University), Nidhi Rathi (Max Planck Institute for Informatics)
Optimization
🎯 What it does: A polynomial query complexity algorithm is proposed to find a serial dictatorship action sequence that maximizes social welfare for one-sided matching problems (OSM) under the constraint of obtaining preference values from agents only through sequential inquiries.
WEPO: Web Element Preference Optimization for LLM-based Web Navigation
Jiarun Liu (Beijing University of Posts and Telecommunications), Zheng Hu (Beijing University of Posts and Telecommunications)
Recommendation SystemOptimizationTransformerLarge Language ModelContrastive LearningTextBenchmark
🎯 What it does: This paper proposes a web navigation method WEPO based on LLM, optimizing the selection of web elements through unsupervised preference learning.
WHALE-FL: Wireless and Heterogeneity Aware Latency Efficient Federated Learning over Mobile Devices via Adaptive Subnetwork Scheduling
Huai-An Su, Miao Pan (Stevens Institute of Technology)
Federated LearningComputational EfficiencyConvolutional Neural NetworkTransformerImageText
🎯 What it does: This paper proposes a wireless and heterogeneous perception delay-efficient federated learning framework called WHALE-FL, which allows mobile devices to dynamically select an appropriately sized subnetwork for local training in each training round based on their real-time computing/communication capabilities and training progress.
What Are Step-Level Reward Models Rewarding? Counterintuitive Findings from MCTS-Boosted Mathematical Reasoning
Yiran Ma (Zhejiang University), Weiqi Luo (Jinan University)
Reinforcement Learning from Human FeedbackReinforcement LearningTextChain-of-Thought
🎯 What it does: This study investigates the role of the Step Reward Model (SRM) in mathematical reasoning, utilizing MCTS to automatically generate preference data and train the SRM, exploring the necessity of natural language descriptions in the SRM.
What Is a Good Question? Assessing Question Quality via Meta-Fact Checking
Bo Zhang (Nanjing Normal University), Junsheng Zhou (Beihang University)
TransformerLarge Language ModelPrompt EngineeringText
🎯 What it does: This paper proposes the Meta-Fact Checking (MFC) method, which interacts with large language models (LLMs) and knowledge graphs (KGs) to obtain complete knowledge, enabling automatic quality assessment of knowledge-based questions and improving LLM performance in multi-hop reasoning tasks.
What Kind of Visual Tokens Do We Need? Training-Free Visual Token Pruning for Multi-Modal Large Language Models from the Perspective of Graph
Yutao Jiang (Xiamen University), Yiyi Zhou (Xiamen University)
Computational EfficiencyGraph Neural NetworkLarge Language ModelVision Language ModelMultimodality
🎯 What it does: This study investigates the issue of visual token redundancy in multimodal large language models (MLLMs) and proposes a training-independent graph-structured visual token pruning method called G-Prune, which can retain important tokens for both foreground and background while significantly reducing computational load.
What to Preserve and What to Transfer: Faithful, Identity-Preserving Diffusion-based Hairstyle Transfer
Chaeyeon Chung (Korea Advanced Institute of Science and Technology), Jaegul Choo (Korea Advanced Institute of Science and Technology)
Image TranslationGenerationPose EstimationDiffusion modelImageVideo
🎯 What it does: This paper proposes HairFusion, a single-stage hairstyle transfer framework based on diffusion models, which can transfer reference hairstyles to target facial images while preserving original features such as identity, clothing, and background.
When Can We Approximate Wide Contrastive Models with Neural Tangent Kernels and Principal Component Analysis?
Gautham Govind Anil (Indian Institute of Technology Madras), Debarghya Ghoshdastidar (Technical University of Munich)
OptimizationRepresentation LearningContrastive LearningImage
🎯 What it does: Analyzes the training dynamics of a two-layer nonlinear contrastive learning model, proving that when using cosine similarity loss, the neural tangent kernel (NTK) of the network remains almost unchanged in wide networks, and reveals that contrastive learning under orthogonal constraints is equivalent to PCA of a random feature matrix.
When Hypergraph Meets Heterophily: New Benchmark Datasets and Baseline
Ming Li (Zhejiang Normal University), Pietro Liò (Cambridge University)
Graph Neural NetworkGraphBenchmark
🎯 What it does: Proposed a heterogeneous hypergraph learning framework, defined heterogeneity metrics, constructed a diverse benchmark dataset, and introduced the framelet-based hypergraph neural network HyperUFG;
When Open-Vocabulary Visual Question Answering Meets Causal Adapter: Benchmark and Approach
Feifei Zhang (Tianjin University of Technology), Changsheng Xu (National Laboratory of Pattern Recognition)
TransformerVision Language ModelImageBenchmark
🎯 What it does: Proposes an Open Vocabulary Visual Question Answering (OVVQA) benchmark and designs a causal adapter based on front-door adjustment to enhance the model's generalization ability on unseen answers.
When Shadow Removal Meets Intrinsic Image Decomposition: A Joint Learning Framework Using Unpaired Data
Rongjia Zheng (Sun Yat-sen University), Wei-Shi Zheng (South China University of Technology)
Image TranslationRestorationConvolutional Neural NetworkGenerative Adversarial NetworkImage
🎯 What it does: This paper proposes a joint learning framework that simultaneously performs shadow removal and intrinsic image decomposition (reflectance and illumination) using unpaired shadow and non-shadow images.
When Should We Prefer State-to-Visual DAgger over Visual Reinforcement Learning?
Tongzhou Mu (University of California San Diego), Hao Su (University of California San Diego)
Robotic IntelligenceConvolutional Neural NetworkReinforcement LearningImage
🎯 What it does: Compared the learning efficiency, progressive performance, and computational cost of State-to-Visual DAgger and Visual RL across 16 different tasks (from ManiSkill, DMControl, Adroit), systematically evaluating the strengths and weaknesses of the two paradigms and providing practical recommendations.
When Witnesses Defend: A Witness Graph Topological Layer for Adversarial Graph Learning
Naheed Anjum Arafat (Nanyang Technological University), Yuzhou Chen (University of California)
Computational EfficiencyAdversarial AttackGraph Neural NetworkGraph
🎯 What it does: This paper proposes the Witness Graph Topological Layer (WGTL), which introduces persistent homology and witness complexes to provide a defense mechanism against adversarial attacks for Graph Neural Networks (GNNs); WGTL enhances robustness against interference in graph node classification tasks.
Where Precision Meets Efficiency: Transformation Diffusion Model for Point Cloud Registration
Yongzhe Yuan (Xidian University), Wenping Ma (Xidian University)
Autonomous DrivingOptimizationComputational EfficiencyDiffusion modelPoint Cloud
🎯 What it does: A point cloud registration method based on the Transform Diffusion Model (TDM) is proposed, treating the registration task as a denoising diffusion process from noise transformation to true transformation.
Who’s the (Multi-)Fairest of Them All: Rethinking Interpolation-Based Data Augmentation Through the Lens of Multicalibration
Karina Halevy (Carnegie Mellon University), Charumathi Badrinath (Harvard University)
ClassificationData-Centric LearningTabular
🎯 What it does: This study investigates the impact of interpolation-based data augmentation (Mixup and Fair Mixup) on the multicalibration fairness and accuracy of binary classification models in multi-group small sample scenarios, comparing it with traditional post-processing multicalibration methods.
Whole Genome Transformer for Gene Interaction Effects in Microbiome Habitat Specificity
Zhufeng Li (Technical University of Munich), Niki Kilbertus (Max Planck Institute for Biology)
ClassificationExplainability and InterpretabilityTransformerLarge Language ModelBiomedical Data
🎯 What it does: A Transformer framework based on whole-genome sequences is proposed for predicting habitat specificity in microorganisms and explaining gene interactions.
Why Does Dropping Edges Usually Outperform Adding Edges in Graph Contrastive Learning?
Yanchen Xu (Northwestern Polytechnical University), Xuelong Li (China Telecom)
Representation LearningGraph Neural NetworkContrastive LearningGraph
🎯 What it does: A self-supervised graph contrastive learning framework EPAGCL based on Error Propagation Rate (EPR) is proposed, which can selectively add and delete edges during graph view generation, thereby maintaining a low error propagation rate of the graph.