AAAI 2025 Papers — Page 14
AAAI Conference on Artificial Intelligence · 3028 papers
Granularity-Adaptive Spatial Evidence Tokenization for Video Question Answering
Hao Jiang (Peking University), Yadong Mu (Kuaishou Technology)
RecognitionGenerationOptimizationComputational EfficiencyTransformerLarge Language ModelSupervised Fine-TuningVision Language ModelVideoText
🎯 What it does: A granularity-adaptive spatial evidence tokenization framework is proposed, which utilizes question guidance to locate key areas at low resolution and performs fine-grained encoding only on these areas at high resolution, thereby better capturing detailed information in video question answering.
Graph Agent Network: Empowering Nodes with Inference Capabilities for Adversarial Resilience
Ao Liu (Sichuan University), Hanyuan Huang (Sichuan University)
ClassificationAdversarial AttackGraph Neural NetworkAgentic AIGraph
🎯 What it does: This paper proposes the Graph Agent Network (GAgN), which treats nodes as agents with limited perspectives, utilizing decentralized communication to achieve global perception, thereby completing robust node classification in the face of edge perturbation attacks.
Graph Coarsening via Supervised Granular-Ball for Scalable Graph Neural Network Training
Shuyin Xia (Chongqing University of Posts and Telecommunications), Guoyin Wang (Chongqing Normal University)
ClassificationComputational EfficiencyGraph Neural NetworkGraph
🎯 What it does: This study investigates the use of Granular-Ball for supervised graph coarsening to enhance the scalability of GNN training.
Graph Consistency and Diversity Measurement for Federated Multi-View Clustering
Bohang Sun (Beijing University of Technology), Gengyu Lyu (Idealism Beijing Technology Co., Ltd.)
Federated LearningSafty and PrivacyGraph Neural NetworkAuto EncoderContrastive LearningGraph
🎯 What it does: A consistency and diversity graph learning framework MGCD is proposed for unsupervised clustering of multi-view data in a federated learning environment, strictly protecting data privacy.
Graph Contrastive Learning with Joint Spectral Augmentation of Attribute and Topology
Liang Yang (Hebei University of Technology), Xiaochun Cao (Sun Yat-sen University)
ClassificationRepresentation LearningGraph Neural NetworkContrastive LearningGraph
🎯 What it does: The GCL-JAM framework is proposed, treating attributes as nodes to construct an attribute interpolation graph, and performing joint spectral enhancement on this graph to improve the representation effect of graph contrastive learning.
Graph Mixture of Experts and Memory-augmented Routers for Multivariate Time Series Anomaly Detection
Xiaoyu Huang (University of Science and Technology of China), Zhendong Mao (University of Science and Technology of China)
Anomaly DetectionGraph Neural NetworkMixture of ExpertsTime Series
🎯 What it does: A multivariate time series anomaly detection framework based on Graph-MoE (Graph Mixture of Experts) is proposed, which can adaptively aggregate structural information from short-range to long-range in multi-layer GNNs and capture global historical features using a memory-enhanced router.
Graph Segmentation and Contrastive Enhanced Explainer for Graph Neural Networks
Zhiqiang Wang (Shanxi University), Jiarong Zhang (Shanxi University)
Explainability and InterpretabilityGraph Neural NetworkContrastive LearningGraph
🎯 What it does: A GNN interpretability method called GSCExplainer based on graph segmentation and contrastive learning is proposed, which can automatically divide the input graph into important subgraphs and redundant subgraphs, enhancing the quality of explanations through contrastive learning.
Graph Structure Learning for Spatial-Temporal Imputation: Adapting to Node and Feature Scales
Xinyu Yang (Nankai University), Xiaojie Yuan (Nankai University)
OptimizationRepresentation LearningGraph Neural NetworkTransformerTime Series
🎯 What it does: Design and propose a multi-scale graph structure learning framework GSLI for spatial-temporal missing value imputation.
Graph Structure Refinement with Energy-based Contrastive Learning
Xianlin Zeng (Beihang University), Baochang Zhang (Beihang University)
OptimizationRepresentation LearningGraph Neural NetworkContrastive LearningGraph
🎯 What it does: This paper proposes an energy contrastive learning (ECL) based graph structure refinement framework, ECL-GSR, for simultaneously learning graph structure and node representations under unsupervised conditions.
Graph-Based Cross-Domain Knowledge Distillation for Cross-Dataset Text-to-Image Person Retrieval
Bingjun Luo (Tsinghua University), Xibin Zhao (Tsinghua University)
RetrievalDomain AdaptationKnowledge DistillationGraph Neural NetworkContrastive LearningImageTextMultimodality
🎯 What it does: This paper proposes an unsupervised domain adaptation framework GCKD, which jointly learns shared feature representations for cross-domain text-image retrieval using graph networks and momentum knowledge distillation.
GraphAvatar: Compact Head Avatars with GNN-Generated 3D Gaussians
Xiaobao Wei (Institute of Software, Chinese Academy of Sciences), Feng Tian (Institute of Software, Chinese Academy of Sciences)
GenerationOptimizationGraph Neural NetworkVideo
🎯 What it does: This paper presents GraphAvatar, which utilizes graph neural networks to generate 3D Gaussian points and combines graph-guided optimization and 3D-aware post-processing to achieve real-time high-quality rendering of facial avatars, requiring only a 10 MB GNN model without storing massive Gaussian points.
Graphic Design with Large Multimodal Model
Yutao Cheng (ByteDance Inc), Jie Shao (ByteDance Inc)
GenerationTransformerLarge Language ModelMultimodality
🎯 What it does: The Hierarchical Layout Generation (HLG) task is proposed, which can automatically generate graphic layouts from a collection of design elements in any order;
GraphMoRE: Mitigating Topological Heterogeneity via Mixture of Riemannian Experts
Zihao Guo (Beihang University), Jianxin Li (Beihang University)
Graph Neural NetworkMixture of ExpertsGraph
🎯 What it does: The GraphMoRE framework is proposed, which dynamically generates personalized mixed curvature spaces for each node through various Riemannian experts and a topology distortion-based gating mechanism, effectively addressing the topological heterogeneity of graphs.
GraSP: Simple Yet Effective Graph Similarity Predictions
Haoran Zheng (Hong Kong Baptist University), Renchi Yang (Hong Kong Baptist University)
OptimizationGraph Neural NetworkGraph
🎯 What it does: This paper proposes a graph similarity prediction framework called GRASP, which estimates graph edit distance (GED) and maximum common subgraph (MCS) without cross-graph node interactions.
GRICP: Granular-Ball Iterative Closest Point with Multikernel Correntropy for Point Cloud Fine Registration
Yihao (Southwest University), Shukai Duan (Southwest University)
OptimizationPoint Cloud
🎯 What it does: A globally robust ICP registration framework GRICP is proposed, which transforms point clouds into particle sphere clouds and uses multi-core mutual information optimization.
Grimm: A Plug-and-Play Perturbation Rectifier for Graph Neural Networks Defending Against Poisoning Attacks
Ao Liu (Sichuan University), Pan Zhou (Huazhong University of Science and Technology)
Anomaly DetectionAdversarial AttackGraph Neural NetworkGraph
🎯 What it does: A plugin-based defense model named GRIMM is proposed, which can monitor, detect, and correct edge perturbations caused by poisoning attacks on GNNs during training without modifying the original GNN structure.
Grounded Multi-Hop VideoQA in Long-Form Egocentric Videos
Qirui Chen (Shanghai Jiao Tong University), Weidi Xie (Shanghai Jiao Tong University)
TransformerLarge Language ModelVideoTextMultimodalityBenchmark
🎯 What it does: This paper proposes the Multi-hop Long-term Autonomous Perspective Video Question Answering (MH-VidQA) task and generates a large number of multi-hop Q&A samples through an automated pipeline, constructing the MULTIHOP-EGOQA benchmark dataset. It also designs the GeLM model, which utilizes location tokens to achieve multi-hop time segment retrieval and reasoning.
GRPose: Learning Graph Relations for Human Image Generation with Pose Priors
Xiangchen Yin (University of Science and Technology of China), Xun Yang (Space AI, Li Auto)
GenerationPose EstimationGraph Neural NetworkDiffusion modelImage
🎯 What it does: A pose-guided portrait generation framework called GRPose is proposed, which captures high-order correlations of pose priors using graph structures.
GRSN: Gated Recurrent Spiking Neurons for POMDPs and MARL
Lang Qin (Zhejiang University), Huajin Tang (Zhejiang University)
Recurrent Neural NetworkSpiking Neural NetworkReinforcement LearningSequential
🎯 What it does: This paper proposes the Time Alignment Paradigm (TAP) and Gated Recurrent Spiking Neurons (GRSN), enabling spiking neural networks to perform single-step updates corresponding to single-step decisions in POMDP and multi-agent reinforcement learning, significantly reducing time steps and lowering energy consumption.
GSDiff: Synthesizing Vector Floorplans via Geometry-enhanced Structural Graph Generation
Sizhe Hu (Hefei University of Technology), Liping Zheng (Hefei University of Technology)
GenerationData SynthesisTransformerDiffusion modelGraph
🎯 What it does: The GSDiff framework is proposed, which transforms floor plans into structural graphs and divides the process into two stages: node generation and edge prediction, to directly synthesize vector floor plans.
GTDE: Grouped Training with Decentralized Execution for Multi-agent Actor-Critic
Mengxian Li (Institute of Computing Technology Chinese Academy of Sciences), Yongjun Xu (Institute of Computing Technology Chinese Academy of Sciences)
Graph Neural NetworkReinforcement LearningGraph
🎯 What it does: A multi-agent reinforcement learning framework called 'Group Training and Decentralized Execution (GTDE)' is proposed and implemented, aiming to address the performance degradation issues of traditional Centralized Training with Decentralized Execution (CTDE) and Decentralized Training with Decentralized Execution (DTDE) in large-scale scenarios.
GTG: Generalizable Trajectory Generation Model for Urban Mobility
Jingyuan Wang (Beihang University), Yudong Li (Beihang University)
GenerationDomain AdaptationGraph Neural NetworkGenerative Adversarial NetworkGraphTime Series
🎯 What it does: A transferable urban trajectory generation model GTG is proposed, utilizing spatial syntax and deep learning to learn city-invariant movement patterns, enabling cross-city trajectory generation.
Guaranteeing Out-Of-Distribution Detection in Deep RL via Transition Estimation
Mohit Prashant (Nanyang Technological University), Michael Yuhas (Nanyang Technological University)
Anomaly DetectionReinforcement LearningSequential
🎯 What it does: This work achieves out-of-distribution (OOD) detection for deep reinforcement learning agents by learning the transfer distribution and can actively interrupt execution during deployment.
Guided and Variance-Corrected Fusion with One-shot Style Alignment for Large-Content Image Generation
Shoukun Sun (University of Idaho), Luca Capriotti (Idaho National Laboratory)
GenerationData SynthesisDiffusion modelImageStochastic Differential Equation
🎯 What it does: Three methods are proposed: Guided Fusion, Variance-Corrected Fusion, and One-shot Style Alignment, to merge overlapping patches when generating large images using small diffusion models, eliminating seams, blurriness, and incoherent objects.
Guided Real Image Dehazing Using YCbCr Color Space
Wenxuan Fang (Nanjing University of Science and Technology), Jun Li (Nanjing University of Science and Technology)
RestorationImage
🎯 What it does: A structure-guided dehazing network (SGDN) based on RGB and YCbCr color spaces is proposed, which effectively removes haze from real hazy images through a dual-color guidance bridge and a color enhancement module.
GuideNER: Annotation Guidelines Are Better than Examples for In-Context Named Entity Recognition
Shizhou Huang (East China Normal University), Xin Alex Lin (East China Normal University)
RecognitionTransformerLarge Language ModelText
🎯 What it does: This paper proposes a GuideNER framework based on LLM, which achieves unsupervised and efficient named entity recognition by summarizing the label patterns of the training set into annotation guidelines.
GURecon: Learning Detailed 3D Geometric Uncertainties for Neural Surface Reconstruction
Zesong Yang (Zhejiang University), Zhaopeng Cui (Simon Fraser University)
GenerationKnowledge DistillationRepresentation LearningNeural Radiance FieldPoint Cloud
🎯 What it does: Learning a 3D geometric uncertainty field in neural surface reconstruction, with unsupervised estimation based on multi-view consistency.
GVMGen: A General Video-to-Music Generation Model with Hierarchical Attentions
Heda Zuo (Zhejiang University), Lingyun Sun (Zhejiang University of Science and Technology)
GenerationData SynthesisTransformerContrastive LearningVideoMultimodalityAudio
🎯 What it does: The GVMGen model is proposed, which automatically generates multi-style audio background music that is highly relevant to video input.
H-MBA: Hierarchical MamBa Adaptation for Multi-Modal Video Understanding in Autonomous Driving
Siran Chen (Shenzhen Institutes of Advanced Technology, Chinese Academy of Sciences), Yali Wang (Shenzhen Institutes of Advanced Technology, Chinese Academy of Sciences)
Autonomous DrivingTransformerLarge Language ModelVideoMultimodality
🎯 What it does: A hierarchical Mamba adaptation framework named H-MBA is proposed to enhance the understanding capability of multimodal videos in autonomous driving scenarios.
HaCore: Efficient Coreset Construction with Locality Sensitive Hashing for Vertical Federated Learning
Qinbo Zhang (Wuhan University), Jiawei Jiang (Wuhan University)
Federated LearningComputational EfficiencyTabular
🎯 What it does: Designed and implemented the HaCore algorithm, which constructs binary signatures through local sensitive hashing and selects representative samples via clustering in vertical federated learning, thereby building an efficient core set.
Hand1000: Generating Realistic Hands from Text with Only 1,000 Images
Haozhuo Zhang (Peking University), Yanbin Hao (University of Science and Technology of China)
GenerationData SynthesisLarge Language ModelSupervised Fine-TuningDiffusion modelImageText
🎯 What it does: Proposes the Hand1000 method, which uses only 1000 gesture images to train a Diffusion model to generate realistic hand images that match text descriptions.
HandDiffuse: Generative Controllers for Two-Hand Interactions via Diffusion Models
Pei Lin (ShanghaiTech University)
GenerationData SynthesisPose EstimationTransformerDiffusion modelImageVideo
🎯 What it does: This paper presents the HandDiffuse 12.5M two-hand interaction action dataset and a controllable generation method based on diffusion models, aimed at generating highly interactive two-hand motion sequences.
Hansel: Output Length Controlling Framework for Large Language Models
Seoha Song (Samsung Research), Hyeonmok Ko (Samsung Research)
GenerationTransformerLarge Language ModelSupervised Fine-TuningPrompt EngineeringText
🎯 What it does: The Hansel framework is introduced during the fine-tuning phase of large language models (LLMs) by periodically inserting hidden special tokens (e.g., |x⟩<y|) in the output to convey remaining length information, thereby achieving precise control over the length of generated text.
Harmonious Music-driven Group Choreography with Trajectory-Controllable Diffusion
Yuqin Dai (Nanjing University of Science and Technology), Jian Yang (Nanjing University of Science and Technology)
Diffusion modelVideo
🎯 What it does: Proposes a trajectory controllable diffusion framework TCDiff for music-driven multi-dancer choreography;
Harmonizing Visual and Textual Embeddings for Zero-Shot Text-to-Image Customization
Yeji Song (Seoul National University), Nojun Kwak (Seoul National University)
Image TranslationGenerationPose EstimationTransformerVision Language ModelDiffusion modelImageTextMultimodality
🎯 What it does: This paper proposes a zero-shot text-to-image customization method to address the pose-identity entanglement problem between visual and text embeddings.
Harnessing Event Sensory Data for Error Pattern Prediction in Vehicles: A Language Model Approach
Hugo Math (Augsburg University), Robin Schön (Augsburg University)
Anomaly DetectionAutonomous DrivingTransformerTime SeriesSequential
🎯 What it does: This paper proposes using a Transformer language model to model vehicle diagnostic event sequences to predict the occurrence time and type of future fault patterns.
Harnessing Language Model for Cross-Heterogeneity Graph Knowledge Transfer
Jinyu Yang (Beijing University of Posts and Telecommunications), Chuan Shi (Beijing University of Posts and Telecommunications)
Knowledge DistillationRepresentation LearningGraph Neural NetworkLarge Language ModelSupervised Fine-TuningGraph
🎯 What it does: A cross-heterogeneous graph knowledge transfer framework LMCH is proposed, which converts the meta-paths of heterogeneous graphs into language text, utilizes language models to extract general knowledge, and performs iterative self-supervised training on the target graph.
Harnessing Large Language Models for Knowledge Graph Question Answering via Adaptive Multi-Aspect Retrieval-Augmentation
Derong Xu (University of Science and Technology of China), Enhong Chen (University of Science and Technology of China)
RetrievalTransformerLarge Language ModelSupervised Fine-TuningPrompt EngineeringTextGraphRetrieval-Augmented Generation
🎯 What it does: The AMAR framework is proposed, which converts knowledge graph information from multi-perspective (entities, relations, subgraphs) retrieval into prompt embeddings through self-alignment and relevance gating, enhancing the reasoning and answer accuracy of large models in KGQA tasks.
Harnessing Multimodal Large Language Models for Multimodal Sequential Recommendation
Yuyang Ye (Rutgers University), Hui Xiong (Hong Kong University of Science and Technology)
Recommendation SystemTransformerLarge Language ModelSupervised Fine-TuningPrompt EngineeringMultimodalitySequential
🎯 What it does: This paper proposes a multi-modal large language model enhanced sequence recommendation framework (MLLM‑MSR), which achieves sequential recommendations of multi-modal information (images + text) through a two-stage user preference summarization and supervised fine-tuning.
HC-LLM: Historical-Constrained Large Language Models for Radiology Report Generation
Tengfei Liu (Beijing University of Technology), Baocai Yin (University of Science and Technology of China)
GenerationTransformerLarge Language ModelContrastive LearningTextBiomedical Data
🎯 What it does: The HC-LLM framework is proposed, utilizing historically constrained large language models to achieve longitudinal generation of chest X-ray reports.
HDLayout: Hierarchical and Directional Layout Planning for Arbitrary Shaped Visual Text Generation
Tonghui Feng (Xidian University), Xiaotian Qiao (Xidian University)
GenerationTransformerDiffusion modelImageText
🎯 What it does: A visual text generation framework based on hierarchical and directional layout, HDLayout, is proposed, which can automatically generate text in any shape solely based on text prompts.
HDT: Hierarchical Discrete Transformer for Multivariate Time Series Forecasting
Feng Shibo, Zhiqi Shen (Nanyang Technological University)
TransformerGenerative Adversarial NetworkTime Series
🎯 What it does: This paper proposes a two-stage hierarchical discrete Transformer (HDT) that compresses high-dimensional multivariate time series into discrete tokens through vector quantization. It first generates a discrete representation of the downsampled target at a lower level, and then generates the complete target conditionally autoregressively at a higher level, achieving long-term forecasting.
Hedging and Approximate Truthfulness in Traditional Forecasting Competitions
Mary Monroe (University of Colorado Boulder), Rafael Frongillo (University of Colorado Boulder)
🎯 What it does: This paper conducts the first systematic strategic analysis of the traditional prediction competition mechanism Simple Max, demonstrating its incentive flaws and failure of truthfulness in multi-event scenarios.
HeGTa: Leveraging Heterogeneous Graph-enhanced Large Language Models for Few-shot Complex Table Understanding
Rihui Jin (Southeast University), Sheng Bi (Southeast University)
Graph Neural NetworkTransformerLarge Language ModelPrompt EngineeringTabular
🎯 What it does: This paper studies a framework called HeGTa that combines heterogeneous graphs with large language models for complex table understanding tasks with few samples.
HeMeNet: Heterogeneous Multichannel Equivariant Network for Protein Multi-task Learning
Rong Han (Tsinghua University), Ting Chen (Tsinghua University)
Drug DiscoveryProtein Structure PredictionGraph Neural NetworkGraphBiomedical DataBenchmark
🎯 What it does: This paper proposes a fully atomic heterogeneous multi-channel E(3) equivariant graph neural network called HeMeNet, which jointly predicts six tasks: ligand binding affinity (LBA), protein-protein affinity (PPA), as well as the enzymatic classification number (EC) and gene ontology (GO) functions (MF, BP, CC). A unified Protein-MT multi-task benchmark set is constructed.
HEP-NAS: Towards Efficient Few-shot Neural Architecture Search via Hierarchical Edge Partitioning
Jianfeng Li (Wuhan University), Lianbo Ma (Northeastern University)
OptimizationKnowledge DistillationNeural Architecture SearchImage
🎯 What it does: This paper proposes HEP-NAS, which achieves more accurate performance evaluation and higher precision network search by hierarchically partitioning hypernetwork edges and gradually shrinking the search space.
HePa: Heterogeneous Graph Prompting for All-Level Classification Tasks
Jia Jinghong (Southeast University), Youyong Kong (Southeast University)
ClassificationGraph Neural NetworkPrompt EngineeringGraph
🎯 What it does: The HePa framework is proposed, which implements a unified prompt template on heterogeneous graphs, supporting few-shot in-context learning for three types of tasks: nodes, graphs, and edges.
HEROS-GAN: Honed-Energy Regularized and Optimal Supervised GAN for Enhancing Accuracy and Range of Low-Cost Accelerometers
Yifeng Wang (Harbin Institute of Technology), Yi Zhao (Harbin Institute of Technology)
GenerationData SynthesisOptimizationGenerative Adversarial NetworkTime Series
🎯 What it does: This paper proposes a HEROS-GAN, which enhances the accuracy and range of low-cost accelerometers through a generative model of unpaired low-cost and high-cost accelerometer signals.
HeterGP: Bridging Heterogeneity in Graph Neural Networks with Multi-View Prompting
Fengyu Yan (Tianjin University), Di Jin (Tianjin University)
ClassificationGraph Neural NetworkPrompt EngineeringGraph
🎯 What it does: The HeterGP framework is proposed, which enhances node and graph classification with a small amount of labeled data by sampling homologous and heterologous subgraphs through random walk dual-view sampling and constructing a dual prompt graph.
Heterogeneous Graph Neural Network on Semantic Tree
Mingyu Guan (Georgia Institute of Technology), Taesoo Kim (Georgia Institute of Technology)
ClassificationGraph Neural NetworkGraph
🎯 What it does: A new heterogeneous graph neural network called HETTREE is designed to simultaneously utilize graph structure and heterogeneous characteristics. It performs offline aggregation of features and labels for all meta-paths before training, constructs a semantic tree structure, and encodes it using subtree attention to obtain node representations and complete the node classification task.
Heterogeneous Multi-Agent Bandits with Parsimonious Hints
Amirmahdi Mirfakhar (University of Massachusetts), Mohammad Hajiesmaili (University of Massachusetts)
Tabular
🎯 What it does: This paper proposes and studies an algorithm that utilizes low-cost hints to reduce exploration costs and achieve time-independent regret in the heterogeneous multi-agent multi-armed bandit (HMA2B) problem, covering both centralized and decentralized settings.
Heterogeneous Multi-Robot Graph Coverage with Proximity and Movement Constraints
Dolev Mutzari (Bar Ilan University), Sarit Kraus (Bar Ilan University)
OptimizationRobotic IntelligenceSimultaneous Localization and MappingGraph
🎯 What it does: This paper proposes a general model for the heterogeneous multi-robot graph coverage problem, considering collaborative coverage under proximity and motion constraints, and provides the shortest traversal scheme.
Heterogeneous Prompt-Guided Entity Inferring and Distilling for Scene-Text Aware Cross-Modal Retrieval
Zhiqian Zhao (Hangzhou Dianzi University), Shaowei Jiang (Hangzhou Dianzi University)
RetrievalKnowledge DistillationRecurrent Neural NetworkGraph Neural NetworkPrompt EngineeringContrastive LearningImageTextMultimodality
🎯 What it does: This study proposes a HOPID network that guides entity reasoning and distillation through heterogeneous prompts, achieving alignment and attribute-centered representation of scene text in images and captions, thereby enhancing cross-modal retrieval performance.
Heuristic-free Knowledge Distillation for Streaming ASR via Multi-modal Training
Ji Won Yoon (Chung-Ang University)
Knowledge DistillationRecurrent Neural NetworkSupervised Fine-TuningTextMultimodalityAudio
🎯 What it does: This paper proposes a self-distillation framework called Heuristic-free KD for knowledge distillation in streaming speech recognition models, which does not require an offline teacher and does not use time offsets.
HFF-Tracker: A Hierarchical Fine-grained Fusion Tracker for Referring Multi-Object Tracking
Zeyong Zhao (Tencent), Xi Chen (Tencent)
Object TrackingConvolutional Neural NetworkVision Language ModelImageTextMultimodality
🎯 What it does: This paper proposes HFF-Tracker, which aims at the Referring Multi-Object Tracking task and achieves precise tracking through hierarchical fine-grained fusion of text-image features.
HGSFusion: Radar-Camera Fusion with Hybrid Generation and Synchronization for 3D Object Detection
Zijian Gu (Southeast University), Wei Hong (Southeast University)
Object DetectionAutonomous DrivingImageMultimodalityPoint Cloud
🎯 What it does: The HGSFusion network is proposed to address the issues of sparse radar point clouds and angular errors through the fusion of radar and camera data.
HHAN: Comprehensive Infectious Disease Source Tracing via Heterogeneous Hypergraph Neural Network
Qiang He (Northeastern University), Hao Sun (Northeastern University)
Graph Neural NetworkGraph
🎯 What it does: A source tracking model HHAN based on heterogeneous hypergraph attention networks is designed to track the spread of infectious disease sources in heterogeneous networks.
HI-DR: Exploiting Health Status-Aware Attention and an EHR Graph+ for Effective Medication Recommendation
Taeri Kim (Hanyang University), Sang-Wook Kim (Hanyang University)
Recommendation SystemDrug DiscoveryGraph Neural NetworkTransformerMultimodalityGraphBiomedical DataElectronic Health Records
🎯 What it does: A new drug recommendation framework, HI-DR, is proposed, utilizing a health-status-aware attention mechanism and an enhanced EHR graph (EHR Graph+) to improve the accuracy and safety of drug recommendations.
HiCM²: Hierarchical Compact Memory Modeling for Dense Video Captioning
Minkuk Kim (Kyung Hee University), Seong Tae Kim (Kyung Hee University)
GenerationRetrievalTransformerLarge Language ModelVideoTextMultimodalityRetrieval-Augmented Generation
🎯 What it does: The HiCM 2 model is proposed, which enhances the accuracy and fluency of dense video subtitle generation by constructing a compressed memory structure at the hierarchical level of human memory and employing a top-down retrieval mechanism.
HieraFashDiff: Hierarchical Fashion Design with Multi-stage Diffusion Models
Zhifeng Xie (Shanghai University), Ying Cao (Giant Network)
GenerationData SynthesisDiffusion modelImageMultimodality
🎯 What it does: A multi-stage diffusion framework is constructed, capable of first generating clothing sketches using high-level concepts, and then gradually refining them through low-level attributes, supporting generation and local editing to be completed in a unified manner.
Hierarchical Alignment-enhanced Adaptive Grounding Network for Generalized Referring Expression Comprehension
Yaxian Wang (Xi'an Jiaotong University), Jun Liu (Xi'an Jiaotong University)
RecognitionObject DetectionTransformerVision Language ModelContrastive LearningImageTextMultimodality
🎯 What it does: A hierarchical alignment enhanced adaptive localization network HieA2G is proposed, which can flexibly detect any number of targets under single-target, multi-target, or no-target free text expressions.
Hierarchical Classification Auxiliary Network for Time Series Forecasting
Yanru Sun (Tianjin University), Qinghua Hu (Tianjin University)
ClassificationOptimizationTransformerTime Series
🎯 What it does: This paper proposes a Hierarchical Classification-based Auxiliary Network (HCAN) to improve time series forecasting.
Hierarchical Consensus Network for Multiview Feature Learning
Chengwei Xia (Lanzhou University), Kun Zhan (Lanzhou University)
Representation LearningAuto EncoderContrastive LearningImage
🎯 What it does: This study investigates multi-view feature learning and proposes a Hierarchical Consensus Network (HCN) that learns unified, unsupervised multi-view representations through clustering consistency, encoding consistency, and global consistency.
Hierarchical Context Pruning: Optimizing Real-World Code Completion with Repository-Level Pretrained Code LLMs
Lei Zhang (Shenzhen Institutes of Advanced Technology, Chinese Academy of Sciences), Min Yang
OptimizationAI Code AssistantLarge Language ModelPrompt EngineeringTextBenchmark
🎯 What it does: This study investigates warehouse-level code completion and proposes the Hierarchical Context Pruning (HCP) strategy to construct high-quality prompts, significantly reducing context length and improving completion accuracy.
Hierarchical Cross-Modal Alignment for Open-Vocabulary 3D Object Detection
Youjun Zhao (City University of Hong Kong), Rynson W. H. Lau
Object DetectionTransformerContrastive LearningImageTextPoint Cloud
🎯 What it does: This paper proposes the HCMA framework, which achieves open vocabulary 3D object detection through hierarchical data integration, interactive cross-modal alignment, and object-focused contextual correction.
Hierarchical Divide-and-Conquer for Fine-Grained Alignment in LLM-Based Medical Evaluation
Shunfan Zheng (East China Normal University), Linlin Wang (East China Normal University)
TransformerLarge Language ModelSupervised Fine-TuningReinforcement LearningTextBiomedical Data
🎯 What it does: A hierarchical split-merge framework HDCEval is proposed and implemented for fine-grained evaluation of medical LLMs, including the construction of specialized evaluation criteria, splitting evaluation tasks, and using expert models along with ADTO training based on preference data.
Hierarchical Gradient-Based Genetic Sampling for Accurate Prediction of Biological Oscillations
Heng Rao (Northeastern University), Minghan Chen (Wake Forest University)
OptimizationData-Centric LearningTime SeriesBiomedical DataOrdinary Differential Equation
🎯 What it does: For the task of predicting biological oscillation frequencies, a Hierarchical Gradient-based Genetic Sampling framework (HGGS) is proposed. By first filtering sensitive boundary points and then performing multi-grid genetic sampling, a more balanced and diverse training set is constructed to improve the prediction accuracy of neural networks.
Hierarchical Mixture of Experts: Generalizable Learning for High-Level Synthesis
Weikai Li (University of California Los Angeles), Yizhou Sun (University of California Los Angeles)
Graph Neural NetworkMixture of ExpertsTabular
🎯 What it does: A dual-layer hierarchical Mixture of Experts (MoE) model is proposed to enhance the cross-domain generalization capability of high-level synthesis (HLS) performance prediction.
Hierarchical Multi-Source Uncertainty Aggregation for Interactive Video Captioning
Ervine Zheng (Rochester Institute of Technology), Qi Yu (Rochester Institute of Technology)
GenerationData-Centric LearningTransformerLarge Language ModelVideoText
🎯 What it does: A hierarchical multi-source uncertainty aggregation framework is proposed for interactive video subtitle generation and active learning.
Hierarchical Vector Quantization for Unsupervised Action Segmentation
Federico Spurio (University of Bonn), Juergen Gall (Toyota Motor Europe)
SegmentationAuto EncoderVideo
🎯 What it does: This study proposes an unsupervised video temporal action segmentation method—Hierarchical Vector Quantization (HVQ), which achieves semantic action segmentation of long videos by learning embeddings, a dual-layer codebook, and clustering from frames to sub-actions and then to actions.
Hierarchically Controlled Deformable 3D Gaussians for Talking Head Synthesis
Zhenhua Wu (Sun Yat-sen University), Guanbin Li (Sun Yat-sen University)
GenerationData SynthesisTransformerGenerative Adversarial NetworkVideoPoint CloudAudio
🎯 What it does: A hierarchical control framework HiCoDe based on deformable 3D Gaussians is proposed for audio-driven talking head synthesis.
Hierarchically-Structured Open-Vocabulary Indoor Scene Synthesis with Pre-trained Large Language Model
Weilin Sun (Shandong University), Lei Meng (Shandong University)
GenerationData SynthesisOptimizationGraph Neural NetworkTransformerLarge Language ModelText
🎯 What it does: This paper proposes a hierarchical indoor scene synthesis pipeline based on large language models (LLMs), where a hierarchical scene description is first generated by the LLM, then a hierarchical-aware graph neural network infers fine-grained relative positions, and finally, a divide-and-conquer optimization is employed to obtain a physically feasible 3D indoor layout.
HiGDA: Hierarchical Graph of Nodes to Learn Local-to-Global Topology for Semi-Supervised Domain Adaptation
Ba Hung Ngo (Chonnam National University), Tae Jong Choi (Chonnam National University)
Domain AdaptationGraph Neural NetworkImage
🎯 What it does: This paper proposes HiGDA—a hierarchical graph node model that captures image sub-block features using local graphs, globally aggregates samples of the same category, and generates pseudo-labels through Graph Active Learning (GAL) to achieve semi-supervised domain adaptation.
High-Fidelity Polarimetric Implicit 3D Reconstruction with View-Dependent Physical Representation
Yu Qiu (Beihang University), Zhiming Zheng (Beihang University)
Data SynthesisDepth EstimationImage
🎯 What it does: A method for implicit 3D reconstruction that combines geometric and polarization information is proposed, utilizing perspective-dependent physical representation to achieve high-fidelity reconstruction of transparent and reflective objects.
High-Resolution Frame Interpolation with Patch-based Cascaded Diffusion
Junhwa Hur (Google), Deqing Sun (University of Toronto)
RestorationGenerationDiffusion modelVideo
🎯 What it does: This paper proposes a high-resolution frame interpolation method called HiFI based on patch-level cascading diffusion, capable of generating high-quality intermediate frames at extremely high resolutions such as 4K/8K.
Higher Order Structures for Graph Explanations
Akshit Sinha (International Institute of Information Technology Hyderabad), Ponnurangam Kumaraguru (International Institute of Information Technology Hyderabad)
Explainability and InterpretabilityGraph Neural NetworkGraphBenchmark
🎯 What it does: The FORGE framework is proposed, which enhances the interpretability of GNNs by elevating graphs to cell complexes.
Highly Efficient Rotation-Invariant Spectral Embedding for Scalable Incomplete Multi-View Clustering
Xinxin Wang (University of Macau), Yicong Zhou (China University of Geosciences)
OptimizationComputational EfficiencyGraph Neural NetworkContrastive LearningMultimodality
🎯 What it does: An efficient rotation-invariant spectral embedding framework RISE is proposed for scalable missing multi-view clustering.
Highly Imperceptible Black-Box Graph Injection Attacks with Reinforcement Learning
Maochang Zhao (Southeast University), Jing Zhang (Southeast University)
Adversarial AttackGraph Neural NetworkReinforcement LearningGraph
🎯 What it does: A black-box node injection attack method based on reinforcement learning, TEANI, is proposed, which can carry out covert attacks on a single target node without knowing the structure of the target GNN.
Highly Parallelized Reinforcement Learning Training with Relaxed Assignment Dependencies
Zhouyu He (National University of Defense Technology), Yusong Tan (National University of Defense Technology)
Reinforcement LearningVideo
🎯 What it does: A distributed reinforcement learning training system called TianJi is proposed, which achieves high-throughput parallel training by relaxing the assignment dependencies between subtasks.
HiPoser: 3D Human Pose Estimation with Hierarchical Shared Learning at Parts-Level Using Inertial Measurement Units
Guorui Liao (Chongqing University), Li Liu (Chongqing University)
Pose EstimationSupervised Fine-TuningTime Series
🎯 What it does: This paper proposes a 3D human pose estimation method called HiPoser based on sparse IMU, which uses a hierarchical shared learning framework to decompose the pose into four sub-tasks: waist, lower limbs, and upper limbs, and sequentially shares motion information.
HiRED: Attention-Guided Token Dropping for Efficient Inference of High-Resolution Vision-Language Models
Kazi Hasan Ibn Arif (Virginia Tech), Bo Ji (University College Dublin)
Computational EfficiencyTransformerVision Language ModelMultimodality
🎯 What it does: This paper proposes HiRED, an early visual token dropping scheme based on CLS attention mechanism for efficient inference of high-resolution Vision-Language Models (VLMs);
HLMEA: Unsupervised Entity Alignment Based on Hybrid Language Models
Xiongnan Jin (Shenzhen University), Jianqiang Li (Shenzhen University)
TransformerLarge Language ModelText
🎯 What it does: A hybrid language model-based unsupervised entity alignment method called HLMEA is proposed.
HOGSA: Bimanual Hand-Object Interaction Understanding with 3D Gaussian Splatting Based Data Augmentation
Wentian Qu (Institute of Software, Chinese Academy of Sciences), Yinda Zhang (Google)
Data SynthesisPose EstimationGaussian SplattingImageMesh
🎯 What it does: A data augmentation framework called HOGSA based on 3D Gaussian Splatting is proposed, which can generate multi-pose and multi-view images of hand-object interactions, significantly expanding the scale and diversity of existing datasets.
HOIMamba: Efficient Mamba-based Disentangled Progressive Learning for HOI Detection
Yongchao Xu (University of Science and Technology of China), Zheng-Jun Zha (University of Science and Technology of China)
RecognitionObject DetectionConvolutional Neural NetworkTransformerSupervised Fine-TuningImage
🎯 What it does: A novel HOI detection framework called HOIMamba is proposed, which utilizes Mamba and LoRA to construct an efficient decoder, and introduces Cross-Enhanced Mamba (CEM) and Detection Context Propagation (DCP) to achieve multi-level progressive learning.
Holistic Correction with Object Prototype for Video Object Segmentation
Shengye Qiao (Beihang University), Jia Li (Beihang University)
Object DetectionSegmentationConvolutional Neural NetworkVideo
🎯 What it does: This paper studies a new Holistic Correction Network (HCNet) that achieves holistic correction in semantics, space, and time through adaptive updates of object prototypes for semi-supervised video object segmentation.
Holistic Semantic Representation for Navigational Trajectory Generation
Ji Cao (Zhejiang University), Mingli Song (Zhejiang University)
GenerationData SynthesisGraph Neural NetworkTransformerTime SeriesSequential
🎯 What it does: This paper proposes a navigation trajectory generation framework called HOSER, which synthesizes high-quality human movement trajectories under the conditions of a given starting point, departure time, and destination.
HomeDiffusion: Zero-Shot Object Customization with Multi-View Representation Learning for Indoor Scenes
Guoqiu Li (Alibaba Group), Yiyun Fei (Alibaba Group)
GenerationRepresentation LearningDiffusion modelContrastive LearningImage
🎯 What it does: A zero-shot object customization framework called HomeDiffusion based on multi-view learning is proposed, which can seamlessly embed reference objects into indoor scenes from the desired perspective while maintaining high detail fidelity.
HomoMatcher: Achieving Dense Feature Matching with Semi-Dense Efficiency by Homography Estimation
Xiaolong Wang (Zhejiang University), Ming Yang (Ant Group)
Pose EstimationTransformerImage
🎯 What it does: A lightweight isomorphic estimation-based fine-grained matching module is proposed, achieving high-precision feature matching under a semi-dense framework through global alignment of patch-to-patch.
HoneypotNet: Backdoor Attacks Against Model Extraction
Yixu Wang (Fudan University), Xingjun Ma (Fudan University)
OptimizationAdversarial AttackConvolutional Neural NetworkImage
🎯 What it does: This paper proposes a new 'Attack as Defense' strategy called HoneypotNet, which utilizes backdoor attack techniques to implant a toxic probability vector in the output layer of the protected model, thereby inducing the attacker’s substitute model to acquire the backdoor during model extraction attacks.
How Do Position Encodings Affect Length Generalization? Case Studies On In-Context Function Learning
Di-Nan Lin (National Cheng Kung University), Hung-Yu Kao (National Tsing Hua University)
Meta LearningTransformerLarge Language ModelTabularSequential
🎯 What it does: In this study, the authors utilized a from-scratch trained GPT-2 model to perform in-context learning (ICL) on synthetic data such as linear regression and Boolean functions, systematically comparing the effects of different position encodings (NoPE, ALiBi, FIRE, Dynamic YaRN) on out-of-distribution (OOD) length extrapolation.
How Does the Smoothness Approximation Method Facilitate Generalization for Federated Adversarial Learning?
Wenjun Ding (Central South University), Zhe Qu (Central South University)
Federated LearningAdversarial AttackImage
🎯 What it does: This paper conducts a theoretical analysis of the generalization error in Federated Adversarial Learning (FAL), providing upper bounds on the generalization error when using three smoothing approximation methods (SSA, RSA, OPSA) under two mainstream algorithms (Vanilla FAL and Slack FAL), and validates the results through experiments.
How Many Lines to Paint the City: Exact Edge-Cover in Temporal Graphs
Argyrios Deligkas (Royal Holloway University of London), Georg Tennigkeit (Hasso Plattner Institute University of Potsdam)
Graph Neural NetworkGraph
🎯 What it does: The study investigates the minimum number of resources required to cover each time edge exactly once using paths, tracks, or walking on time graphs, and systematically analyzes the solvability and unsolvability of the problem.
How Not to Stitch Representations to Measure Similarity: Task Loss Matching Versus Direct Matching
András Balogh (University of Szeged), Márk Jelasity (University of Szeged)
ClassificationRepresentation LearningConvolutional Neural NetworkTransformerImage
🎯 What it does: This paper studies and compares two model stitching methods—Task Loss Matching and Direct Matching—in terms of their performance in measuring the similarity of internal representations in deep networks, and contrasts them with traditional structural similarity metrics (such as CCA, CKA, OPD).
How to Re-enable PDE Loss for Physical Systems Modeling Under Partial Observation
Haodong Feng (Zhejiang University), Dixia Fan (Westlake University)
OptimizationComputational EfficiencyTransformerTime SeriesPhysics RelatedOrdinary Differential Equation
🎯 What it does: The RPLPO framework is proposed, which utilizes an encoding module to reconstruct learnable high-resolution states and uses PDE loss with a transition module to address the challenge of training physical system models under partial observations.
HS-FPN: High Frequency and Spatial Perception FPN for Tiny Object Detection
Zican Shi (Huazhong University of Science and Technology), Junyu Guo (Chinese People's Liberation Army)
Object DetectionImage
🎯 What it does: This paper proposes HS-FPN, which improves the traditional FPN by combining the High-Frequency Perception module (HFP) and the Spatial Dependency Perception module (SDP) to enhance small object detection performance.
HSEvo: Elevating Automatic Heuristic Design with Diversity-Driven Harmony Search and Genetic Algorithm Using LLMs
Pham Vu Tuan Dat (Hanoi University of Science and Technology), Huynh Thi Thanh Binh (George Mason University)
OptimizationTransformerLarge Language ModelText
🎯 What it does: This study investigates the introduction of diversity metrics within the LLM-EPS framework and designs the HSEvo framework to balance diversity and target performance, thereby enhancing the effectiveness of automatic heuristic design.
HSOD-BIT-V2: A Challenging Benchmark for Hyperspectral Salient Object Detection
Yuhao Qiu (Beijing Institute of Technology), Jianan Li (Beijing Institute of Technology)
Object DetectionConvolutional Neural NetworkTransformerImageBenchmark
🎯 What it does: The HSOD-BIT-V2 hyperspectral salient object detection dataset is proposed, and a high-resolution network called Hyper-HRNet is designed to achieve precise hyperspectral salient object detection.
HSRDiff: A Hierarchical Self-Regulation Diffusion Model for Stochastic Semantic Segmentation
Han Yang (Institute of Computing Technology, Chinese Academy of Sciences), Yongjun Xu (Institute of Computing Technology, Chinese Academy of Sciences)
SegmentationGenerationDiffusion modelImageBiomedical DataMagnetic Resonance ImagingComputed Tomography
🎯 What it does: A hierarchical self-regulating diffusion model, HSRDiff, is proposed for generating diverse and reliable semantic segmentation hypotheses in safety-critical domains.
HUANG: A Robust Diffusion Model-based Targeted Adversarial Attack Against Deep Hashing Retrieval
Chihan Huang (Nanjing University of Science and Technology), Xiaobo Shen (Nanjing University of Science and Technology)
RetrievalAdversarial AttackDiffusion modelImage
🎯 What it does: A black-box targeted adversarial attack method based on diffusion models (HUANG) is proposed to disrupt the retrieval effectiveness of deep hashing retrieval systems.
Hubness Change Point Detection
Ikumi Suzuki (Yamagata University), Eiji Murakami (Azbil Kimmon Co., Ltd.)
Anomaly DetectionTime Series
🎯 What it does: A model-free change point detection method based on hubness suppression and uniformity testing, called HubCPD, is proposed.