IJCAI 2025 Papers — Page 10
International Joint Conference on Artificial Intelligence · 1014 papers
Strategyproofness and Monotone Allocation of Auction in Social Networks
Yuhang Guo (University of Electronic Science and Technology of China), Bakh Khoussainov (University of Electronic Science and Technology of China)
OptimizationGraph
🎯 What it does: Study truthful auctions in social networks, propose two monotonic allocation rules ID-MON and IP-MON, and provide an optimal payment scheme.
Streaming Multi-agent Pathfinding
Mingkai Tang (Hong Kong University of Science and Technology), Kaichen Zhang (Hong Kong University of Science and Technology)
OptimizationGraphBenchmark
🎯 What it does: This paper proposes the stream-based multi-agent path finding (S-MAPF) problem for assembly lines, and presents the ASCBS algorithm based on conflict-driven search to solve this problem;
Structure-Aware Handwritten Text Recognition via Graph-Enhanced Cross-Modal Mutual Learning
Ji Gan (Chongqing University of Posts and Telecommunications), Xinbo Gao (Chongqing University of Posts and Telecommunications)
RecognitionKnowledge DistillationConvolutional Neural NetworkTransformerVision Language ModelImageGraph
🎯 What it does: Proposed a graph-enhanced cross-modal mutual learning network (GCM), achieving structure-aware handwritten text recognition by simultaneously processing handwritten text images and their corresponding geometric graph structures;
Subgraph Information Bottleneck with Causal Dependency for Stable Molecular Relational Learning
Peiliang Zhang (Wuhan University of Technology), Lin Li (Wuhan University of Technology)
Representation LearningDrug DiscoveryGraph Neural NetworkGraph
🎯 What it does: Proposed the Causal Subgraph Information Bottleneck (CausalGIB), generating stable molecular relationship learning representations by introducing causal subgraph information and the subgraph information bottleneck;
Suit the Node Pair to the Case: A Multi-Scale Node Pair Grouping Strategy for Graph-MLP Distillation
Rui Dong (Southeast University), Youyong Kong (Southeast University)
Knowledge DistillationGraph Neural NetworkContrastive LearningGraph
🎯 What it does: Propose a multi-scale node pair grouping strategy and a corresponding multi-scale knowledge distillation optimization loss to achieve efficient knowledge distillation from GNN to MLP;
SyncAnimation: A Real-Time End-to-End Framework for Audio-Driven Human Pose and Talking Head Animation
Yujian Liu (AiShiWeiLai AI Research), Xiaoli Liu (AiShiWeiLai AI Research)
GenerationPose EstimationNeural Radiance FieldAuto EncoderVideoAudio
🎯 What it does: Propose SyncAnimation, an end-to-end real-time audio-driven NeRF framework capable of generating head pose, facial expressions, and upper-body movements synchronized with audio.
SyncGaussian: Stable 3D Gaussian-Based Talking Head Generation with Enhanced Lip Sync via Discriminative Speech Features
Ke Liu (University of Electronic Science and Technology of China), Yang Yang (University of Electronic Science and Technology of China)
GenerationGaussian SplattingVideoMultimodalityAudio
🎯 What it does: This paper proposes the SyncGaussian framework, which realizes real-time and stable lip-sync speaker generation using 3D Gaussian splatting.
Synthesis of Communication Policies for Multi-Agent Systems Robust to Communication Restrictions
Saleh Soudijani (CISPA Helmholtz Center for Information Security), Rayna Dimitrova (CISPA Helmholtz Center for Information Security)
OptimizationReinforcement LearningBenchmark
🎯 What it does: This paper proposes a method to synthesize joint actions and communication strategies in multi-agent systems with communication constraints, achieving optimal goal attainment-avoidance objectives while minimizing additional communication overhead;
Synthesising Minimum Cost Dynamic Norms
Natasha Alechina (Open University Netherlands), Giuseppe Perelli (Sapienza University of Rome)
Optimization
🎯 What it does: This paper studies how to automatically synthesize dynamic specifications with minimal cost in multi-agent systems to satisfy system-level objectives expressed in ATL*sc.
T-T: Table Transformer for Tagging-based Aspect Sentiment Triplet Extraction
Kun Peng (Chinese Academy of Sciences), Philip S. Yu (University of Illinois at Chicago)
RecognitionTransformerTextBenchmark
🎯 What it does: This paper proposes Table-Transformer (T-T), which encodes sentences into 2D tables and employs an improved Transformer layer (stripe attention + loop-shift) for relation learning, achieving superior performance in the Aspect Sentiment Triplet Extraction task.
T2S: High-resolution Time Series Generation with Text-to-Series Diffusion Models
Yunfeng Ge (Xidian University), Shirui Pan (Griffith University)
GenerationTransformerDiffusion modelFlow-based ModelAuto EncoderTextTime Series
🎯 What it does: This paper proposes the T2S framework to achieve text-to-time series generation and constructs a fragment-level high-resolution text-time series pair dataset at the 600K scale.
Tackling Long-Tailed Data Challenges in Spiking Neural Networks via Heterogeneous Knowledge Distillation
Moqi Li (Xidian University), Cheng Deng (Xidian University)
ClassificationKnowledge DistillationSpiking Neural NetworkTransformerImage
🎯 What it does: Propose the LT-SpikingFormer framework, combining Spike Transformer as the SNN backbone and addressing long-tailed visual classification through heterogeneous CNN-SNN knowledge distillation (global logits distillation + local normalization distillation).
TCDM: A Temporal Correlation-Empowered Diffusion Model for Time Series Forecasting
Huibo Xu (University of Science and Technology of China), Qi Liu (University of Science and Technology of China)
Diffusion modelGenerative Adversarial NetworkTime Series
🎯 What it does: Proposes a diffusion model called TCDM based on time series decomposition and enhancement, using a baseline model to predict trends, an improved diffusion model to predict residuals, and finally fusing the results to obtain the final prediction.
Template-based Uncertainty Multimodal Fusion Network for RGBT Tracking
Zhaodong Ding (National Key Laboratory Of Opto Electronic Information Acquisition And Protection Technology), Jin Tang (National Key Laboratory Of Opto Electronic Information Acquisition And Protection Technology)
Object TrackingTransformerContrastive LearningMultimodality
🎯 What it does: A template-based uncertainty multimodal fusion network is constructed for RGBT tracking.
Template3D-AD: Point Cloud Template Matching Method Based on Center Points for 3D Anomaly Detection
Yi Liu (Northeastern University), Yufei Yang (Northeastern University)
Anomaly DetectionPoint Cloud
🎯 What it does: Proposed a 3D anomaly detection method called Template3D-AD based on template matching, achieving detection through center point matching and global-local feature fusion.
Temporal Consistency Constrained Transferable Adversarial Attacks with Background Mixup for Action Recognition
Ping Li (Hangzhou Dianzi University), Bo Pang (Hangzhou Dianzi University)
RecognitionAdversarial AttackConvolutional Neural NetworkTransformerReinforcement LearningImageVideo
🎯 What it does: Propose a transferable adversarial attack method BMTC based on background mixing and temporal consistency for video action recognition models.
Tensorial Multi-view Clustering with Deep Anchor Graph Projection
Wei Feng (Northwest A&F University), Bo Dong (Xi'an Jiaotong University)
OptimizationRepresentation LearningMultimodality
🎯 What it does: Propose a tensor multi-view clustering method TMVC-DAGP based on deep anchor graph projection, directly projecting the anchor graph into the label space, integrating deep feature extraction, tensor Schatten p-norm fusion, and sparse regularization;
Test-Time Adaptation on Recommender System with Data-Centric Graph Transformation
Yating Liu (Dalian University of Technology), Yanqing Guo (Dalian University of Technology)
Domain AdaptationRecommendation SystemData-Centric LearningGraph Neural NetworkContrastive LearningGraph
🎯 What it does: Proposes a model-free fine-tuning test-time adaptation framework called TTA-GREC, which dynamically reconstructs user-item interaction graphs and knowledge graphs during testing using graph transformation techniques in data centers, alleviating distribution shift between training and testing.
TEST-V: TEst-time Support-set Tuning for Zero-shot Video Classification
Rui Yan (Nanjing University of Science and Technology), Tieniu Tan (Nanjing University)
RecognitionLarge Language ModelVision Language ModelVideoText
🎯 What it does: Proposes the TEST-V framework, which achieves zero-shot video classification by expanding the support set with multiple prompts during testing and performing time-aware weight optimization.
TESTN: A Triad-Enhanced Spatio-Temporal Network for Multi-Temporal POI Relationship Inference
Hongyu Wang (University of Electronic Science and Technology of China), Shuo Shang (University of Electronic Science and Technology of China)
Recommendation SystemGraph Neural NetworkGraphTime Series
🎯 What it does: Proposed a triplet-enhanced spatiotemporal network called TESTN for multi-temporal POI relationship reasoning.
TextMEF: Text-guided Prompt Learning for Multi-exposure Image Fusion
Jinyuan Liu (Dalian University of Technology), Xin Fan (Dalian University of Technology)
RestorationRecurrent Neural NetworkPrompt EngineeringVision Language ModelImageMultimodality
🎯 What it does: Achieve multi-exposure image fusion and generate high dynamic range images through CLIP prompt learning and Mamba network;
The Core of Approval-Based Committee Elections with Few Seats
Dominik Peters (Université Paris Dauphine - PSL)
🎯 What it does: Proves that in approval elections, the core is always non-empty when the number of seats k ≤ 8 or the number of candidates m ≤ 15, and provides the corresponding existence proof.
The Devil is in Fine-tuning and Long-tailed Problems: A New Benchmark for Scene Text Detection
Tianjiao Cao (Chinese Academy of Sciences), Yu Zhou (Nankai University)
Object DetectionVision Language ModelAuto EncoderImageBenchmark
🎯 What it does: This paper addresses the performance gap between scene text detection in academic benchmarks and real-world applications by analyzing the fine-tuning gap and long-tailed distribution issues, proposing the Joint Dataset Learning (JDL) evaluation protocol and Long-Tailed Benchmark (LTB), and providing the MAEDet self-supervised baseline.
The First Theoretical Approximation Guarantees for the Non-Dominated Sorting Genetic Algorithm III (NSGA-III)
Renzhong Deng (Harbin Institute of Technology), Benjamin Doerr (Ecole Polytechnique)
OptimizationBenchmark
🎯 What it does: First theoretical analysis of NSGA-III's approximation performance on the ONEMINMAX and LOTZ benchmarks, providing an approximate upper bound for the maximum spacing interval (MEI);
The Proportional Veto Principle for Approval Ballots
Daniel Halpern (Harvard University), Warut Suksompong (National University of Singapore)
🎯 What it does: This paper proposes a flexible voter representation (FVR) based on a proportional veto principle, and presents the optimal scoring rule R_OPT that realizes this principle. Subsequently, the concept is extended to multi-win election scenarios, and its feasibility and limitations are analyzed.
The Role of Video Generation in Enhancing Data-Limited Action Understanding
Wei Li (Institute of Information Engineering, Chinese Academy of Sciences), Yu Zhou (Nankai University)
ClassificationRecognitionGenerationData SynthesisAnomaly DetectionTransformerVision Language ModelDiffusion modelVideoTextMultimodality
🎯 What it does: Use a text-to-video diffusion Transformer (CogVideoX-2B) to automatically generate an unlimited-scale labeled video dataset from tags, and propose an information enhancement strategy (environment and role description) and an uncertainty label smoothing strategy to improve video action understanding performance in data-scarce environments.
Theoretical Analysis of Evolutionary Algorithms with Quality Diversity for a Classical Path Planning Problem
Duc-Cuong Dang (University of Passau), Dirk Sudholt (University of Passau)
OptimizationGraph
🎯 What it does: Analyze the theoretical performance of Quality Diversity (QD) algorithms on the All-Pairs Shortest Path (APSP) problem, and provide the collaborative working mechanisms of mutation and crossover operators in the behavior space along with runtime upper bounds;
Theoretical Insights into Fine-Tuning Attention Mechanism: Generalization and Optimization
Xinhao Yao (Renmin University of China), Yong Liu (Renmin University of China)
OptimizationComputational EfficiencyTransformerLarge Language ModelSupervised Fine-TuningTextBenchmark
🎯 What it does: The paper investigates the attention mechanism in large language models during fine-tuning, proposing that fine-tuning only the query (Wq) and value (Wv) matrices can achieve results comparable to or even better than full-parameter fine-tuning, and highlights that using a higher learning rate for Wv accelerates convergence;
Think Twice Before Adaptation: Improving Adaptability of DeepFake Detection via Online Test-Time Adaptation
Hong-Hanh Nguyen-Le (University College Dublin), Nhien-An Le-Khac (University College Dublin)
Domain AdaptationAnomaly DetectionConvolutional Neural NetworkImageVideo
🎯 What it does: Proposed an online test-time adaptation method, T A 2, to enhance the robustness of deep fake detectors under unknown post-processing and distribution shifts.
Tight Runtime Guarantees From Understanding the Population Dynamics of the GSEMO Multi-Objective Evolutionary Algorithm
Benjamin Doerr (Laboratoire d'Informatique (LIX), CNRS, Ecole Polytechnique, Institut Polytechnique de Paris), Andre Opris (University of Passau)
OptimizationBenchmark
🎯 What it does: Studied the population dynamics of GSEMO on the COCZ benchmark and provided a lower bound on its runtime.
Time-Frequency Disentanglement Boosted Pre-Training: A Universal Spatio-Temporal Modeling Framework
Yudong Zhang (University of Science and Technology of China), Yang Wang (University of Science and Technology of China)
Domain AdaptationRepresentation LearningData-Centric LearningGraph Neural NetworkTransformerPrompt EngineeringAuto EncoderContrastive LearningGraphTime Series
🎯 What it does: Propose a generic spatiotemporal modeling framework USTC, achieving cross-city, cross-task few-shot transfer learning, supporting spatiotemporal prediction, missing value imputation, and extrapolation of unobserved nodes.
Token-Level Accept or Reject: A Micro Alignment Approach for Large Language Models
Yang Zhang (Hong Kong Polytechnic University), Edward Chung (Hong Kong Polytechnic University)
Reinforcement Learning from Human FeedbackLarge Language ModelReinforcement LearningText
🎯 What it does: Designed a micro alignment model MArA that uses an accept/reject mechanism to perform binary classification on each candidate token during generation, thereby achieving alignment of LLMs.
Top-Down Guidance for Learning Object-Centric Representations
Junhong Zou (Chinese Academy of Sciences), Zhen Lei (Chinese Academy of Sciences)
Representation LearningAuto EncoderImageVideo
🎯 What it does: Proposed a TDGNet with top-down guidance and conflict detection mechanisms to improve unsupervised object-centric representations;
Top-I2P: Explore Open-Domain Image-to-Point Cloud Registration Using Topology Relationship
Pei An (Huazhong University of Science and Technology), Liangliang Nan (Delft University of Technology)
Pose EstimationConvolutional Neural NetworkGraph Neural NetworkImagePoint Cloud
🎯 What it does: Propose a Top-I2P framework for open-domain image-to-point cloud (I2P) registration based on topological relationships, addressing the sparsity of cross-modal feature interactions and computational bottlenecks.
TOTF: Missing-Aware Encoders for Clustering on Multi-View Incomplete Attributed Graphs
Mengyao Li (Hunan University), Kenli Li (Hunan University)
Representation LearningGraph Neural NetworkAuto EncoderGenerative Adversarial NetworkGraph
🎯 What it does: Proposed a TOTF (Train Once Then Freeze) framework for clustering on multi-view incomplete attribute graphs.
Toward Reliable Scientific Hypothesis Generation: Evaluating Truthfulness and Hallucination in Large Language Models
Guangzhi Xiong (University of Virginia), Aidong Zhang (University of Virginia)
TransformerLarge Language ModelBiomedical DataBenchmarkRetrieval-Augmented GenerationChain-of-Thought
🎯 What it does: Propose the TruthHypo benchmark and KnowHD framework to evaluate the authenticity and hallucination issues of large language models when generating scientific hypotheses.
Towards a Unified View of Social Laws with Instantaneous Actions
Alexander Tuisov (Technion Israel Institute of Technology), Erez Karpas (Technion Israel Institute of Technology)
Benchmark
🎯 What it does: This paper proposes a new compilation method for verifying the robustness of social laws in multi-agent planning;
Towards Anytime Retrieval: A Benchmark for Anytime Person Re-Identification
Xulin Li (University of Science and Technology of China), Nenghai Yu (University of Science and Technology of China)
RetrievalTransformerMixture of ExpertsImageBenchmark
🎯 What it does: Proposed the AT-ReID task, constructed the AT-USTC large-scale dataset, and designed the Uni-AT model to achieve real-time person re-identification across day/night, seasonal, and long/short-term scenarios.
Towards Automatic Sampling of User Behaviors for Sequential Recommender Systems
Hao Zhang (University of Science and Technology of China), Junzhe Jiang (University of Science and Technology of China)
Recommendation SystemTransformerReinforcement LearningSequential
🎯 What it does: Propose the AutoSAM automatic sampling framework, which uses reinforcement learning to perform non-uniform sampling on historical behaviors in sequence recommendation, thereby enhancing the model's generalization ability.
Towards Debiased Generalized Category Discovery
Pengcheng Guo (Xi'an Jiaotong University), Boyu Wang (University of Western Ontario)
ClassificationRepresentation LearningTransformerImage
🎯 What it does: In the Generalized Category Discovery (GCD) task, the DeGCD method is proposed, addressing the competition between new and old categories by introducing a debiasing classification head;
Towards Equilibrium: An Instantaneous Probe-and-Rebalance Multimodal Learning Approach
Yang Yang (Nanjing University of Science and Technology), Qing-Yuan Jiang (Nanjing University of Science and Technology)
Convolutional Neural NetworkTransformerContrastive LearningImageVideoTextMultimodalityAudio
🎯 What it does: This paper proposes an instant detection-rebalancing multi-modal learning framework (IPRM), which first instantly evaluates modality strength through two forward passes, and then instantly recalibrates modality weights based on the evaluation results, achieving real-time balanced training for modality imbalance.
Towards Fairness with Limited Demographics via Disentangled Learning
Zichong Wang (Florida International University), Wenbin Zhang (Florida International University)
Auto EncoderImageTabular
🎯 What it does: Propose a framework (FDVAE) that infers missing data and simultaneously optimizes model fairness using limited demographic information when complete demographic information is missing, achieving fair machine learning;
Towards Generalizable Neural Simulators: Addressing Distribution Shifts Induced by Environmental and Temporal Variations
Jiaqi Liu (Jilin University), Bo Yang (Jilin University)
Domain AdaptationTime SeriesStochastic Differential EquationOrdinary Differential Equation
🎯 What it does: Propose a neural simulator named CoPoNDP that can simultaneously address the effects of environmental and temporal distribution shifts on dynamical systems;
Towards Improved Risk Bounds for Transductive Learning
Bowei Zhu (Renmin University of China), Yong Liu (Renmin University of China)
Domain Adaptation
🎯 What it does: Proposes a new slicing technique based on functional layering, derives tighter unified local convergence upper bounds in transfer learning with without-replacement sampling, and designs a time-varying penalty estimator that incorporates variance information.
Towards Micro-Action Recognition with Limited Annotations: An Asynchronous Pseudo Labeling and Training Approach
Yan Zhang (Hefei University of Technology), Meng Wang (Hefei University of Technology)
RecognitionConvolutional Neural NetworkVideo
🎯 What it does: Proposed an asynchronous pseudo-labeling and training framework APLT for semi-supervised learning in micro-action recognition, significantly reducing annotation requirements.
Towards Recognizing Spatial-temporal Collaboration of EEG Phase Brain Networks for Emotion Understanding
Jiangfeng Sun (Beijing University of Posts and Telecommunications), Meina Song (Beijing University of Posts and Telecommunications)
ClassificationRecognitionRecurrent Neural NetworkGraph Neural NetworkBiomedical Data
🎯 What it does: This paper proposes the Stella framework, which constructs phase brain networks using adaptive band selection and combines Fourier graph operators with spatial-temporal encoders to achieve emotion recognition.
Towards Region-Adaptive Feature Disentanglement and Enhancement for Small Object Detection
Yanchao Bi (Shandong Jianzhu University), Leida Li (Xidian University)
Object DetectionConvolutional Neural NetworkImage
🎯 What it does: Propose a Regional Adaptive Feature Decoupling and Enhancement (RAFDE) strategy for UAV small target detection, which includes Boundary Transitional Region-enhanced Downsampling (BTRD) and Regional Adaptive Feature Fusion (RAFF) modules;
Towards Regularized Mixture of Predictions for Class-Imbalanced Semi-Supervised Facial Expression Recognition
Hangyu Li (Hong Kong Baptist University), Bo Han (Hong Kong Baptist University)
RecognitionImage
🎯 What it does: Propose a semi-supervised facial expression recognition method (ReMoP) addressing class imbalance, generating high-quality pseudo labels by combining linear classifier predictions with feature similarity predictions, and enhancing the geometric properties of feature distribution through a class regularization term.
Towards Robust Deterministic and Probabilistic Modeling for Predictive Learning
Xuesong Nie (Zhejiang University), Xiaofeng Liu (Yale University)
GenerationDiffusion modelVideo
🎯 What it does: Proposed the DDP framework, which models low-level details and high-level motion separately in the latent space using deterministic visual paths and probabilistic motion paths;
Towards Robust Incremental Learning Under Ambiguous Supervision
Rui Wang (Zhejiang University), Chang Yao (Zhejiang University)
ClassificationKnowledge DistillationRepresentation LearningImage
🎯 What it does: This paper proposes an Incremental Partial Label Learning (IPLL) framework, and implements the Prototype-guided Disambiguation and Replay algorithm (PGDR) to address the dual challenges of label ambiguity and catastrophic forgetting.
Towards Safer Pretraining: Analyzing and Filtering Harmful Content in Webscale Datasets for Responsible LLMs
Sai Krishna Mendu (Microsoft), Parag Agrawal (Microsoft)
Safty and PrivacyTransformerLarge Language ModelPrompt EngineeringTextBenchmarkChain-of-Thought
🎯 What it does: This paper conducts systematic audits of large-scale web datasets such as Common Crawl, C4, and FineWeb for harmful content, constructing a three-dimensional (Safe, Topical, Toxic) harm classification framework. It generates TTP prompts using GPT-4 Omni Prompt, trains a Longformer-based HarmFormer model to achieve high-accuracy filtering of long texts, and releases the TTP-Eval evaluation set and HAVOC multi-harm generation benchmark.
Towards VLM-based Hybrid Explainable Prompt Enhancement for Zero-Shot Industrial Anomaly Detection
Weichao Cai (Xiamen University), Jie Wen (Harbin Institute of Technology)
Anomaly DetectionExplainability and InterpretabilityTransformerLarge Language ModelPrompt EngineeringVision Language ModelImageMultimodality
🎯 What it does: Proposed a multi-stage prompt generation agent and a hybrid interpretable prompt enhancement framework for zero-shot industrial anomaly detection, leveraging multi-modal large language models (MLLM) and vision foundation models (VFM) to generate fine-grained text prompts and attention prompts, thereby enhancing CLIP's detection and localization performance on unseen anomaly categories.
TP-Eval: Tap Multimodal LLMs' Potential in Evaluation by Customizing Prompts
Yuxuan Xie (Shanghai Artificial Intelligence Laboratory), Kaipeng Zhang (Shanghai Artificial Intelligence Laboratory)
OptimizationTransformerLarge Language ModelPrompt EngineeringMultimodalityBenchmark
🎯 What it does: Proposed a TP-Eval framework for evaluating multimodal large language models (MLLMs) based on automatic prompt optimization, which can customize optimal prompts for each model to alleviate underestimation and bias caused by prompt sensitivity.
Trace: Structural Riemannian Bridge Matching for Transferable Source Localization in Information Propagation
Li Sun (North China Electric Power University), Philip S. Yu (University of Illinois at Chicago)
Graph Neural NetworkGraphStochastic Differential Equation
🎯 What it does: Proposes a transferable source localization method called Trace, which directly models the mapping between the source and the final distribution on a manifold using a structural Riemannian Schrödinger bridge.
Training-free Fourier Phase Diffusion for Style Transfer
Siyuan Zhang (Beijing University of Technology), Hongbin Zha (Peking University)
Image TranslationGenerationDiffusion modelImageText
🎯 What it does: Propose a training-free Fourier phase diffusion model that uses the Fourier phase spectrum of the content image as a condition to modulate intermediate samples during the diffusion process, achieving high-quality style transfer while preserving the content structure.
TrajCogn: Leveraging LLMs for Cognizing Movement Patterns and Travel Purposes from Trajectories
Zeyu Zhou (Beijing Jiaotong University), Huaiyu Wan (Beijing Jiaotong University)
ClassificationRetrievalTransformerLarge Language ModelSupervised Fine-TuningPrompt EngineeringTime SeriesSequential
🎯 What it does: Proposed the TrajCogn model, which leverages LLM (GPT-2) to learn spatiotemporal trajectories through trajectory prompts and a trajectory semantic embedder, enabling the model to recognize motion patterns and travel purposes from trajectories, and directly apply to multiple tasks (trip duration prediction, destination prediction, similar trajectory retrieval, trajectory classification).
Trajectory-Dependent Generalization Bounds for Pairwise Learning with φ-mixing Samples
Liyuan Liu (Huazhong Agricultural University), Yulong Wang (Huazhong Agricultural University)
Sequential
🎯 What it does: This paper proposes a generalization error upper bound for trajectory-dependent dual learning with φ-mixing samples;
Transferable Relativistic Predictor: Mitigating Cross-Task Cold-Start Issue in NAS
Nan Li (Northeastern University), Mengjie Zhang (Victoria University of Wellington)
Neural Architecture SearchContrastive LearningImage
🎯 What it does: Proposes a transferable relative predictor (TRP) that pre-trains using soft labels from multiple zero-training metrics and dynamically determines the required evaluation samples for fine-tuning via Chebyshev interpolation, significantly reducing the cold-start evaluation cost in NAS.
Tree-of-AdEditor: Heuristic Tree Reasoning for Automated Video Advertisement Editing with Large Language Model
Yuqi Zhang (Northwestern Polytechnical University), Qing Li (Hong Kong Polytechnic University)
GenerationTransformerLarge Language ModelVision Language ModelVideoTextMultimodalityChain-of-Thought
🎯 What it does: Propose the Tree-of-AdEditor (ToAE) framework, which uses large language models to construct a tree-like reasoning process for automatic ad editing of product videos.
TreeKV: Smooth Key-Value Cache Compression with Tree Structures
Ziwei He (Shanghai Jiao Tong University), Bo Jiang (Shanghai Jiao Tong University)
Computational EfficiencyTransformerLarge Language ModelText
🎯 What it does: Propose TreeKV, a training-free, tree-structure-based KV cache compression method that balances prefilling and decoding stages;
TsCA: On the Semantic Consistency Alignment via Conditional Transport for Compositional Zero-Shot Learning
Miaoge Li (Hong Kong Polytechnic University), Song Guo (Hong Kong University of Science and Technology)
ClassificationRecognitionRepresentation LearningTransformerPrompt EngineeringVision Language ModelContrastive LearningMultimodalityBenchmark
🎯 What it does: By constructing three groups of distributions of image patches, combinations, and primitives, and leveraging conditional transport with cyclic consistency regularization, the TsCA framework achieves fine-grained alignment and reasoning for Compositional Zero-Shot Learning (CZSL).
TSTAI: A Time-varying Brain Effective Connectivity Network Construction Method Combining with Brain Active Information
Qi Chen (Northeastern University), Junchang Xin (Northeastern University)
Biomedical Data
🎯 What it does: Propose a two-stage high-order non-stationary brain effective connectivity network construction method (TSTAI) based on two types of brain activity information, and introduce hierarchical screening, improved scoring functions, and pruning strategies in structural learning.
Two-Stage Feature Generation with Transformer and Reinforcement Learning
Wanfu Gao (Jilin University), Kunpeng Liu (Portland State University)
Representation LearningTransformerReinforcement LearningTabular
🎯 What it does: Propose a two-stage feature generation framework TSFG, first using a Transformer encoder-decoder for pre-training to generate candidate features, then fine-tuning with PPO reinforcement learning to make the generated features more aligned with downstream tasks;
Two-stage Risk Control with Application to Ranked Retrieval
Yunpeng Xu (New Jersey Institute of Technology), Zhi Wei (New Jersey Institute of Technology)
RetrievalText
🎯 What it does: Propose a two-stage risk control method specifically designed for ranking retrieval systems, capable of controlling expected risk simultaneously during the retrieval and ranking phases, along with corresponding theoretical guarantees.
UltraModel: A Modeling Paradigm for Industrial Objects
Haoran Yang (Zhejiang University), Wenhai Wang (Zhejiang University)
OptimizationGraph Neural NetworkTabular
🎯 What it does: Proposed UltraModel, a general modeling paradigm for the industrial object modeling (MIO) task;
Uncertainty-guided Graph Contrastive Learning from a Unified Perspective
Zhiqiang Li (Shanxi University), Jiye Liang (Shanxi University)
Representation LearningGraph Neural NetworkContrastive LearningGraph
🎯 What it does: This paper proposes a graph contrastive learning framework called UGCL based on sample uncertainty, which unifies and coordinates data augmentation with contrastive objectives to enhance graph embedding quality.
Underground Diagnosis in 3D GPR Data by Learning in CuCoRes Model Space
Xiren Zhou (University of Science and Technology of China), Huanhuan Chen (University of Science and Technology of China)
Anomaly DetectionRecurrent Neural NetworkPoint Cloud
🎯 What it does: Propose a method for detecting and classifying underground defects using three-dimensional ground-penetrating radar (GPR) data within the CuCoRes model space;
Understanding Matters: Semantic-Structural Determined Visual Relocalization for Large Scenes
Jingyi Nie (Beihang University), Zhong Zhou (Beihang University)
Pose EstimationConvolutional Neural NetworkSimultaneous Localization and MappingImage
🎯 What it does: Propose a visual localization method based on semantic-structural partitioning, which first divides large scenes into semantically consistent and structurally continuous sub-scenes, then uses sampling learning to quickly predict partition labels, and finally achieves pose refinement through partition-aware feature fusion and discriminative point selection.
Understanding Visual Detail Hallucinations of Large Vision-Language Models
Xiaoxi Sun (Peking University), Dongyan Zhao (Peking University)
Object DetectionAnomaly DetectionExplainability and InterpretabilityLarge Language ModelVision Language ModelImageMultimodalityChain-of-Thought
🎯 What it does: This study constructs a specialized dataset of visual detail illusions to systematically evaluate the performance of large visual-language models (LVLMs) in identifying and interpreting small targets, and explores the impact of various training-free schemes and visual encoder designs on small target illusions.
UniCT Depth: Event-Image Fusion Based Monocular Depth Estimation with Convolution-Compensated ViT Dual SA Block
Luoxi Jing (Peking University), Jianqiang Xia
Depth EstimationConvolutional Neural NetworkTransformerImageMultimodality
🎯 What it does: Propose a monocular depth estimation framework named UniCT Depth that integrates event cameras with traditional images.
Universal Backdoor Defense via Label Consistency in Vertical Federated Learning
Peng Chen, Wanchun Dou (Nanjing University)
Federated LearningAdversarial AttackConvolutional Neural NetworkSupervised Fine-TuningImageMultimodality
🎯 What it does: A general backdoor defense framework (UBD) is proposed under the vertical federated learning environment. It identifies backdoor targets and reverses triggers through label consistency clustering (LCC), and then fine-tunes the model using linear probing (LP) combined with BN statistical constraints, thereby simultaneously detecting and eliminating backdoor attacks.
Universal Graph Self-Contrastive Learning
Liang Yang (Hebei University of Technology), Zhen Wang (Northwestern Polytechnical University)
Representation LearningGraph Neural NetworkContrastive LearningGraph
🎯 What it does: Designed and proposed an augmentation-free, node-attribute self-contrast learning framework called GRASS, achieving graph representation learning through self-contrast between nodes and attributes.
Unleashing the Potential of Transformer Flow for Photorealistic Face Restoration
Kepeng Xu (Xidian University), Wenxin Yu (Southwest University of Science and Technology)
RestorationTransformerDiffusion modelFlow-based ModelAuto EncoderImage
🎯 What it does: Propose OmniFace, a face restoration framework based on Transformer Flow, combining large-parameter Transformers, C-Projector ControlNet, and adaptive training strategies to achieve high-resolution, realistic face restoration.
Unleashing the Semantic Adaptability of Controlled Diffusion Model for Image Colorization
Xiangcheng Du (Fudan University), Cheng Jin (Fudan University)
RestorationDiffusion modelImage
🎯 What it does: This paper proposes a semantics-adaptive controlled diffusion model called SeAda, which automatically restores grayscale images into colorful, semantically consistent color images.
Unlocking Dark Vision Potential for Medical Image Segmentation
Hongpeng Yang (University of South Carolina), Fei Guo (Central South University)
SegmentationConvolutional Neural NetworkImageBiomedical DataMagnetic Resonance Imaging
🎯 What it does: Proposed a plug-and-play network called DVNet based on the principle of dark vision, utilizing wavelet transform and Mamba fusion modules to enhance the accuracy of medical image segmentation.
Unlocking the Potential of Lightweight Quantized Models for Deepfake Detection
Renshuai Tao (Beijing Jiaotong University), Wei Wang (Beijing Jiaotong University)
Anomaly DetectionComputational EfficiencyConvolutional Neural NetworkImage
🎯 What it does: Propose a low-bit quantization framework (including Connected Quantized Block and Shifted Logarithmic Redistribution Quantizer) for efficiently performing deepfake detection on resource-constrained edge devices.
Unsupervised Feature Transformation via In-context Generation, Generator-critic LLM Agents, and Duet-play Teaming
Nanxu Gong (Arizona State University), Yanjie Fu (Arizona State University)
Representation LearningTransformerLarge Language ModelAgentic AIPrompt EngineeringTabular
🎯 What it does: Propose an unsupervised feature transformation framework that leverages two LLMs, a generator and a critic, to generate feature transformations through dialogue and iteratively optimize them;
Unveiling Maternity and Infant Care Conversations: A Chinese Dialogue Dataset for Enhanced Parenting Support
Bo Xu (Dalian University of Technology), Hongfei Lin (Dalian University of Technology)
Graph Neural NetworkTransformerLarge Language ModelPrompt EngineeringTextGraphRetrieval-Augmented Generation
🎯 What it does: Construct a Chinese maternal and infant care dialogue dataset called MicDialogue and propose a knowledge-driven dialogue generation model called Kng-BART
Unveiling the Power of Noise Priors: Enhancing Diffusion Models for Mobile Traffic Prediction
Zhi Sheng (Tsinghua University), Yong Li (Tsinghua University)
Recurrent Neural NetworkDiffusion modelTime Series
🎯 What it does: Propose the NPDiff framework, which leverages noise priors to enhance the performance of diffusion models in mobile traffic prediction
Variational Graph Auto-Encoder Driven Graph Enhancement for Sequential Recommendation
Yuwen Liu (China University of Petroleum (East China)), Xiaokang Zhou (Kansai University)
Recommendation SystemGraph Neural NetworkAuto EncoderGraphSequential
🎯 What it does: In the sequential recommendation task, a graph-enhanced method called VGAE‑GE based on VGAE is proposed, which reconstructs the item transition graph using a graph autoencoder and combines it with Mamba4Rec to achieve robust sequential recommendation.
Variational Multi-Modal Hypergraph Attention Network for Multi-Modal Relation Extraction
Qian Li (Beijing University of Posts and Telecommunications), Shangguang Wang (Beijing University of Posts and Telecommunications)
Computational EfficiencyRepresentation LearningGraph Neural NetworkMultimodality
🎯 What it does: Proposed a Variational Multimodal Hypergraph Attention Network (VM-HAN) for Multimodal Relation Extraction.
Variational Offline Multi-agent Skill Discovery
Jiayu Chen (Carnegie Mellon University), Vaneet Aggarwal
Representation LearningReinforcement LearningAuto EncoderSequential
🎯 What it does: This paper proposes a framework (VO-MASD) for automatically learning multi-agent skills from offline multi-task data, which constructs 3D or hierarchical codebooks through dynamic grouping and VQ-VAE, enabling simultaneous capture of subgroup and temporal hierarchical collaborative abstractions.
Variety-Seeking Jump Games on Graphs
Lata Narayanan (Concordia University), Alexandros A. Voudouris (University of Essex)
Graph
🎯 What it does: Study a jumping game where players occupy nodes in a graph, aiming to maximize the number of different types in their neighborhood, and explore the existence and quality of equilibria in this game.
VeRecycle: Reclaiming Guarantees from Probabilistic Certificates for Stochastic Dynamical Systems after Change
Sterre Lutz (Delft University of Technology), Anna Lukina (Delft University of Technology)
Computational EfficiencyRobotic IntelligenceReinforcement LearningBenchmark
🎯 What it does: Propose the VeRecycle framework, which automatically recovers safety guarantees from existing probabilistic neural Lyapunov certificates after local changes in system dynamics, avoiding the need to re-verify the entire system.
Verified Certificates via SAT and Computer Algebra Systems for the Ramsey R(3,8) and R(3,9) Problems
Zhengyu Li (Georgia Institute of Technology), Vijay Ganesh (Georgia Institute of Technology)
Computational EfficiencyGraph
🎯 What it does: Verifiable complete search certificates for Ramsey numbers R(3,8)=28 and R(3,9)=36 were generated using the SAT+CAS method (MATHCHECK combined with CADICAL and CAS).
Verifying Quantized Graph Neural Networks is PSPACE-complete
Marco Sälzer, Nicolas Troquard (Gran Sasso Science Institute)
Explainability and InterpretabilityGraph Neural NetworkGraph
🎯 What it does: This paper proposes a formal verification framework for quantized graph neural networks (GNNs), including the linear constraint validity problem (LVP) and a new logical language L quantGNN, along with the corresponding proof system.
VideoHumanMIB: Unlocking Appearance Decoupling for Video Human Motion In-betweening
Haiwei Xue (Tsinghua University), Zhiyong Wu (Tsinghua University)
GenerationPose EstimationDiffusion modelAuto EncoderOptical FlowVideo
🎯 What it does: Propose the VideoHumanMIB framework to achieve video human motion in-betweening, generating continuous and natural human action sequences between two frames.
VidEvo: Evolving Video Editing through Exhaustive Temporal Modeling
Sizhe Dang (Xi'an Jiaotong University), Jingdong Wang (Baidu Inc)
GenerationVision Language ModelDiffusion modelVideoText
🎯 What it does: Propose a single-shot text-guided video editing framework called VidEvo, which balances global and local temporal consistency.
View-Association-Guided Dynamic Multi-View Classification
Xinyan Liang (Shanxi University), Lu Chen (Shanxi University)
ClassificationGraph Neural NetworkAuto Encoder
🎯 What it does: Propose a view association guided dynamic multi-view classification method (AssoDMVC), which enhances classification performance by explicitly modeling inter-view relationships and dynamically adjusting view weights.
VimGeo: Efficient Cross-View Geo-Localization with Vision Mamba Architecture
Jinglin Huang (Guangdong University of Technology), Rong Yu (Guangdong University of Technology)
RetrievalComputational EfficiencyRepresentation LearningImageBenchmark
🎯 What it does: In the cross-view geolocation task, the VimGeo network is proposed, combining Vision Mamba backbone, Channel Group Pooling (CGP), and Dynamic Weighted Batch-tuple Loss (DWBL) to achieve efficient and accurate localization.
Viral Marketing and Convergence Properties in Generalised Voter Model
Abhiram Manohara (Indian Institute of Science), Ahad N. Zehmakan (Australian National University)
Graph
🎯 What it does: Proposed an extended voting model (GVM) that allows directed weighted graphs, nodes without initial opinions, and stubborn nodes, and studied the problem of selecting seed blue nodes to maximize the expected number of blue nodes in this model; simultaneously analyzed the convergence period and convergence time of this process;
Visual Perturbation and Adaptive Hard Negative Contrastive Learning for Compositional Reasoning in Vision-Language Models
Xin Huang (Nanyang Normal University), Ya Wang (Nanyang Normal University)
Representation LearningSupervised Fine-TuningVision Language ModelContrastive LearningMultimodality
🎯 What it does: By mapping the hard negative semantic shifts in the text layer to the visual space, generating image-level hard negatives, and employing adaptive hard negative contrastive learning to enhance VLM performance on compositional reasoning tasks.
Volumetric Axial Disentanglement Enabling Advancing in Medical Image Segmentation
Xingru Huang (Hangzhou Dianzi University), Xiaoshuai Zhang (Ocean University of China)
SegmentationConvolutional Neural NetworkTransformerBiomedical Data
🎯 What it does: The paper proposes a post-refinement module based on volume axial decoupling (PaR) to improve the results of 3D medical image segmentation.
VQCounter: Designing Visual Prompt Queue for Accurate Open-World Counting
Fanfan Ye (Zhejiang University), Mingli Song (Zhejiang University)
Object DetectionTransformerPrompt EngineeringVision Language ModelMultimodality
🎯 What it does: Propose VQCounter, an open-world counting framework based on visual prompt queues and Voronoi matching, addressing issues such as insufficient visual prompt diversity, low matching efficiency, limitations of single-modal training, and inadequate localization error evaluation.
Wave-driven Graph Neural Networks with Energy Dynamics for Over-smoothing Mitigation
Peihan Wu (Shanghai Normal University), Qin Zhao (Shanghai Normal University)
Representation LearningGraph Neural NetworkGraphOrdinary Differential Equation
🎯 What it does: This paper proposes a graph neural network framework based on the wave equation, combined with an energy dynamic regulation mechanism to alleviate the over-smoothing problem in deep GNNs.
Wave-wise Discriminative Tracking by Phase-Amplitude Separation, Augmentation and Mixture
Huibin Tan (National University of Defense Technology), Mengzhu Wang (Hebei University of Technology)
Object TrackingTransformerContrastive LearningImageVideo
🎯 What it does: Propose treating image patches as wave functions, constructing the Wave-wise Discriminative Tracker (WDT) through phase-amplitude separation, enhancement, and mixing, and training/deploying it on visual object tracking tasks.
Wavelet Multi-scale Region-Enhanced Network for Medical Image Segmentation
Hang Lu (Anhui University), Peng Zhou (Anhui University)
SegmentationConvolutional Neural NetworkBiomedical DataMagnetic Resonance ImagingComputed Tomography
🎯 What it does: Propose a UNet-style multi-scale medical image segmentation network (WMREN) that integrates a dual-branch encoder combining wavelet transform and residual networks, along with spatial adaptive fusion and contrast enhancement modules.
WDMIR: Wavelet-Driven Multimodal Intent Recognition
Weiyin Gong (University of Science and Technology of China), Linbo Zhu (Hefei Comprehensive National Science Center)
RecognitionRecurrent Neural NetworkTransformerVideoTextMultimodalityAudio
🎯 What it does: Proposed and implemented the WDMIR framework, integrating frequency domain fusion via wavelet transform, cross-modal collaborative representation, and layer-wise evolutionary fusion, enabling joint intent recognition from three modalities (text, video, audio).
Weakly-supervised Audio Temporal Forgery Localization via Progressive Audio-language Co-learning Network
Junyan Wu (Sun Yat-sen University), Shize Guo (State Key Laboratory of Mathematical Engineering and Advanced Computing)
Anomaly DetectionTransformerLarge Language ModelPrompt EngineeringContrastive LearningMultimodalityAudio
🎯 What it does: Propose a weakly supervised audio temporal forgery localization method called LOCO, which can locate forged segments using only sentence-level labels.
WenyanGPT: A Large Language Model for Classical Chinese Tasks
Xinyu Yao (Minzu University of China), Xiaobing Zhao (Minzu University of China)
TransformerLarge Language ModelSupervised Fine-TuningTextBenchmark
🎯 What it does: Proposed WenyanGPT, a specialized large language model for ancient Chinese texts, and constructed the WenyanBENCH benchmark;