AAAI 2025 Papers — Page 8
AAAI Conference on Artificial Intelligence · 3028 papers
Designing Specialized Two-Dimensional Graph Spectral Filters for Spatial-Temporal Graph Modeling
Yuxin Chen (Nanjing University of Science and Technology), Hui Yan (Nanjing University of Science and Technology)
Graph Neural NetworkGraphTime Series
🎯 What it does: A new spatio-temporal graph neural network (STSGNN) is proposed, which jointly models the interaction between spatial and temporal dimensions through a two-dimensional graph frequency domain filter;
Destroy and Repair Using Hyper-Graphs for Routing
Ke Li (Southern University of Science and Technology), Qingfu Zhang (City University of Hong Kong)
OptimizationGraph Neural NetworkSupervised Fine-TuningGraphTabular
🎯 What it does: This paper proposes a destruction-repair framework based on hypergraphs (DRHG) for searching large neighborhoods and learning the repair process in large routing problems (TSP, CVRP).
Detail Matters: Mamba-Inspired Joint Unfolding Network for Snapshot Spectral Compressive Imaging
Mengjie Qin (Westlake University), Xin Yuan (Westlake University)
RestorationCompressionTransformerImage
🎯 What it does: A joint unfolding network called MiJUN based on Mamba inspiration is proposed for spectral compressed imaging reconstruction in CASSI systems.
Details Enhancement in Unsigned Distance Field Learning for High-fidelity 3D Surface Reconstruction
Cheng Xu (Institute of Software, Chinese Academy of Sciences), Ying He (Nanyang Technological University)
GenerationOptimizationPoint Cloud
🎯 What it does: Learning unsigned distance fields (UDF) from unstructured point clouds and achieving high-fidelity 3D surface reconstruction
Detecting and Corrupting Convolution-based Unlearnable Examples
Minghui Li (Huazhong University of Science and Technology), Leo Yu Zhang (Griffith University)
Anomaly DetectionAdversarial AttackConvolutional Neural NetworkSupervised Fine-TuningImage
🎯 What it does: This paper proposes a complete framework for identifying and defending against convolutional-based unlearned examples (UEs). It first implements detection through edge pixel statistics (EPD) and then disrupts the convolutional noise distribution using random matrix multiplication (COIN), while also extending two new types of convolutional UEs (VUDA, HUDA).
Detecting and Mitigating Hallucination in Large Vision Language Models via Fine-Grained AI Feedback
Wenyi Xiao (Zhejiang University), Linchao Zhu (Zhejiang University)
GenerationOptimizationReinforcement Learning from Human FeedbackTransformerLarge Language ModelVision Language ModelTextMultimodality
🎯 What it does: This paper proposes a framework for hallucination detection and mitigation based on fine-grained AI feedback for LVLM, which includes automatically generating sentence-level hallucination annotation data, a detect-then-rewrite preference data construction pipeline, and the HSA-DPO method that incorporates hallucination severity into Direct Preference Optimization (DPO);
Detecting Music Performance Errors with Transformers
Benjamin Shiue-Hal Chou (Purdue University), Yung-Hsiang Lu (Purdue University)
ClassificationAnomaly DetectionTransformerAudio
🎯 What it does: Polytune is proposed, an end-to-end model based on Transformer that can directly align performance audio with score audio and label each note as Correct, Missed, or Extra, supporting multiple instruments without the need for traditional manual alignment methods such as Dynamic Time Warping (DTW).
Deterministic Policy Gradient Primal-Dual Methods for Continuous-Space Constrained MDPs
Sergio Rozada (King Juan Carlos University), Alejandro Ribeiro (University of Pennsylvania)
OptimizationRobotic IntelligenceReinforcement LearningSequential
🎯 What it does: A deterministic policy gradient primal-dual algorithm for constrained Markov decision processes with continuous state-action spaces is proposed, along with non-asymptotic convergence guarantees.
DetRF: Detachable Novel Views Synthesis of Dynamic Scenes Using Backdrop-Driven Neural Radiance Fields
Boyu Zhang (SenseTime Research), Wenbo Xu (Waytous)
Data SynthesisAutonomous DrivingTransformerNeural Radiance FieldOptical FlowVideo
🎯 What it does: The DetRF method is proposed, utilizing background-driven neural radiance fields and 6D input NeRF to separate static backgrounds from dynamic foregrounds in monocular videos, generating high-quality arbitrary view synthesis results.
Deviate or Not: Learning Coalition Structures with Multiple-bit Observations in Games
Yixuan Even Xu (Carnegie Mellon University), Fei Fang (Carnegie Mellon University)
🎯 What it does: This paper addresses the problem of learning hidden coalition structures in multi-agent systems through multi-bit observations (a binary vector indicating whether each agent deviates from a specified strategy in each round) and presents a multi-round learning algorithm that approaches the information-theoretic lower bound across various game types (normal form games, congestion games, graphical games, and auctions).
DF-MIA: A Distribution-Free Membership Inference Attack on Fine-Tuned Large Language Models
Zhiheng Huang (Harbin Institute of Technology), Yu Li (ByteDance)
Adversarial AttackTransformerLarge Language ModelSupervised Fine-TuningText
🎯 What it does: The paper proposes a distribution-free membership inference attack framework DF-MIA for fine-tuning large language models.
DFDNet: Disentangling and Filtering Dynamics for Enhanced Video Prediction
Lianqiang Gan (University of Electronic Science and Technology of China), Yi Bin (Tongji University)
GenerationData SynthesisConvolutional Neural NetworkVideo
🎯 What it does: A video prediction network named DFDNet is proposed, which can decompose dynamics in the spatial direction and remove instantaneous high-frequency noise through a learnable threshold filter.
DFF: Decision-Focused Fine-Tuning for Smarter Predict-Then-Optimize with Limited Data
Jiaqi Yang (Tongji University), Kun An (Tongji University)
Recommendation SystemOptimizationData-Centric LearningSupervised Fine-TuningTabularFinance Related
🎯 What it does: A Decision-Focused Fine-tuning (DFF) framework is proposed, utilizing a bias correction layer to perform constrained fine-tuning on existing predictive models within the prediction-optimization pipeline, aiming to enhance decision quality and reduce prediction bias;
DG-Mamba: Robust and Efficient Dynamic Graph Structure Learning with Selective State Space Models
Haonan Yuan (Beihang University), Jianxin Li (Ministry of Public Security)
Computational EfficiencyRepresentation LearningGraph Neural NetworkGraph
🎯 What it does: A robust and efficient dynamic graph structure learning framework, DG-Mamba, is proposed, which can achieve joint learning of graph structure and representation with linear time complexity.
DGFamba: Learning Flow Factorized State Space for Visual Domain Generalization
Qi Bi (Jarvis Research Center Tencent YouTu Lab), Yuexiang Li (Faculty of Information Technology Monash University)
ClassificationDomain AdaptationRepresentation LearningConvolutional Neural NetworkFlow-based ModelAuto EncoderImage
🎯 What it does: This paper proposes a state space model DGFamba based on flow factorization for visual domain generalization;
DHAKR: Learning Deep Hierarchical Attention-Based Kernelized Representations for Graph Classification
Feifei Qian (Beijing Normal University), Edwin Hancock (University of York)
ClassificationGraph Neural NetworkGraphBiomedical Data
🎯 What it does: A hierarchical attention-based kernel representation (DHAKR) is proposed for graph classification, which constructs a differentiable kernel matrix by aggregating original substructures into composite substructures layer by layer and weighting them using an attention mechanism.
DHMoE: Diffusion Generated Hierarchical Multi-Granular Expertise for Stock Prediction
Weijun Chen (Peking University), Yanze Wang (Peking University)
Recommendation SystemOptimizationTransformerMixture of ExpertsDiffusion modelMultimodalityTime SeriesFinance Related
🎯 What it does: A hierarchical expert mixture model based on diffusion generation (DHMoE) is proposed, achieving recursive parameter generation from top-level experts to multi-modal predictions from bottom-level experts in stock forecasting, and making bottom-up decisions through a sparse synthesis attention mechanism.
DialogDraw: Image Generation and Editing System Based on Multi-Turn Dialogue
Shichao Ma (NetEase Inc), Zhipeng Hu (NetEase Inc)
GenerationLarge Language ModelSupervised Fine-TuningReinforcement LearningDiffusion modelImageTextMultimodality
🎯 What it does: Proposes the DialogDraw multi-turn dialogue-based image generation and editing system.
DiCA: Disambiguated Contrastive Alignment for Cross-Modal Retrieval with Partial Labels
Chao Su (Sichuan University), Xu Wang (Sichuan University)
RetrievalContrastive LearningImageTextMultimodality
🎯 What it does: A cross-modal retrieval method named DiCA is proposed, specifically addressing the retrieval task when only a partial set of labels is given (i.e., the candidate label set contains the true label).
DIDiffGes: Decoupled Semi-Implicit Diffusion Models for Real-time Gesture Generation from Speech
Yongkang Cheng (Tencent AI Lab), Mingming Gong (University of Melbourne)
GenerationData SynthesisTransformerDiffusion modelGenerative Adversarial NetworkVideoMultimodalityAudio
🎯 What it does: A real-time voice-driven gesture generation framework named DIDiffGes has been developed, which synthesizes high-quality and expressive full-body gestures using a semi-implicit diffusion model combined with GAN, achieving this with only 10 sampling steps.
Diff-Shadow: Global-guided Diffusion Model for Shadow Removal
Jinting Luo (Megvii Technology Inc.), Shuaicheng Liu (University of Electronic Science and Technology of China)
RestorationConvolutional Neural NetworkDiffusion modelImage
🎯 What it does: A global-guided diffusion model called Diff-Shadow is proposed for shadow removal without obvious boundaries and consistent lighting.
DiffCalib: Reformulating Monocular Camera Calibration as Diffusion-Based Dense Incident Map Generation
Xiankang He (Zhejiang University of Technology), Dongyan Guo (Zhejiang University of Technology)
GenerationDepth EstimationDiffusion modelImage
🎯 What it does: We propose DiffCalib, which transforms monocular camera calibration into a dense incident image generation task using diffusion models, jointly estimating depth and incident images, and achieving high-precision camera intrinsic parameter estimation by leveraging the visual prior of Stable Diffusion.
DiffCLIP: Few-shot Language-driven Multimodal Classifier
Jiaqing Zhang (Xidian University), Yunsong Li (Shanghai AI Laboratory)
ClassificationRecognitionSegmentationTransformerDiffusion modelContrastive LearningImageMultimodality
🎯 What it does: This paper proposes DiffCLIP, a high-dimensional multimodal remote sensing image classification framework that combines unsupervised mask diffusion pre-training with language-based few-shot classification.
DiffCorr: Conditional Diffusion Model with Reliable Pseudo-Label Guidance for Unsupervised Point Cloud Shape Correspondence
Jiacheng Deng (University of Science and Technology of China), Wenfei Yang (University of Science and Technology of China)
GenerationData SynthesisTransformerDiffusion modelPoint Cloud
🎯 What it does: This paper proposes an unsupervised point cloud shape correspondence method called DiffCorr, which achieves a step-by-step correspondence from coarse to fine through a conditional diffusion model and reliable pseudo-label guidance.
DiffDVC: Accurate Event Detection for Dense Video Captioning via Diffusion Models
Wei Chen (Beihang University), Guogang Zhu (Beihang University)
Object DetectionGenerationTransformerDiffusion modelVideo
🎯 What it does: A Dense Video Captioning method based on diffusion models, called DiffDVC, is proposed, and a boundary-sensitive detector is designed to improve the boundary accuracy of video event detection.
Differentiable Adversarial Attacks for Marked Temporal Point Processes
Pritish Chakraborty (Indian Institute of Technology Bombay), Abir De (Indian Institute of Technology Delhi)
Adversarial AttackTransformerTime SeriesSequential
🎯 What it does: A differentiable adversarial attack framework for Marked Time Point Processes (MTPP) called PERMTPP has been designed and implemented, which significantly reduces the performance of MTPP in time prediction and label prediction while keeping the perturbation low perceptible.
Differentiable Information Enhanced Model-Based Reinforcement Learning
Xiaoyuan Zhang (Peking University), Yaodong Yang (Peking University)
Robotic IntelligenceReinforcement LearningTabular
🎯 What it does: A model-based reinforcement learning framework called MB-MIX is proposed, which combines trajectory length mixing and Sobolev training to enhance the accuracy of dynamic models and the stability of policy gradient estimation.
Differential Alignment for Domain Adaptive Object Detection
Xinyu He (Tianjin University), Xiaojie Guo (Tianjin University)
Object DetectionDomain AdaptationAdversarial AttackImage
🎯 What it does: A differentiated alignment domain adaptation framework for object detection is proposed, designing a predicted difference feedback instance alignment module and an uncertainty foreground-guided image alignment module to achieve weighted alignment for different regions.
Differential Private Stochastic Optimization with Heavy-tailed Data: Towards Optimal Rates
Puning Zhao (Sun Yat-sen University), Qingming Li (Zhejiang University)
OptimizationSafty and Privacy
🎯 What it does: This paper studies convex optimization problems under differential privacy (DP), particularly in the case of heavy-tailed data, and proposes two algorithms to achieve optimal DP optimization rates.
Differentially Private Prototypes for Imbalanced Transfer Learning
Dariush Wahdany (Fraunhofer AISEC), Franziska Boenisch (CISPA Helmholtz Center for Information Security)
ClassificationDomain AdaptationSafty and PrivacyTransformerImage
🎯 What it does: A private transfer learning framework based on differentially private prototypes (DPPL) is proposed, which utilizes publicly pre-trained encoders to extract features and generates publicly releasable class prototypes through a differential privacy mechanism to complete classification tasks on private data.
DiffExp: Efficient Exploration in Reward Fine-tuning for Text-to-Image Diffusion Models
Daewon Chae (Korea University), Kimin Lee (KAIST)
GenerationData SynthesisOptimizationReinforcement LearningPrompt EngineeringDiffusion modelImageText
🎯 What it does: This paper proposes DiffExp, a method that enhances exploration, improves sample diversity, and increases reward convergence efficiency during the reward fine-tuning process of text-to-image diffusion models by dynamically scheduling CFG and weighting random prompts.
DiffGrasp: Whole-Body Grasping Synthesis Guided by Object Motion Using a Diffusion Model
Yonghao Zhang (Institute of Software Chinese Academy of Sciences), Hongan Wang (Institute of Software Chinese Academy of Sciences)
GenerationRobotic IntelligenceTransformerDiffusion modelPoint Cloud
🎯 What it does: This paper proposes a DiffGrasp framework based on diffusion models, designed to generate full-body human grasping action sequences conditioned on given 3D object motion sequences. It can simultaneously generate hand grasping postures and body movements while maintaining the realism of hand-object contact and body coordination.
Difficulty-aware Balancing Margin Loss for Long-tailed Recognition
Minseok Son (Korea Advanced Institute of Science and Technology), Changick Kim (Korea Advanced Institute of Science and Technology)
RecognitionContrastive LearningImage
🎯 What it does: A Difficult Perception Balanced Margin Loss (DBM) is proposed for long-tail visual recognition.
DiffIM: Differentiable Influence Minimization with Surrogate Modeling and Continuous Relaxation
Junghun Lee (Korea Advanced Institute of Science and Technology), Kijung Shin (Korea Advanced Institute of Science and Technology)
OptimizationGraph Neural NetworkGraph
🎯 What it does: A method for impact minimization based on differentiable learning, DIFFIM, is proposed, which combines surrogate models and continuous relaxation to achieve efficient edge deletion.
DiffRetouch: Using Diffusion to Retouch on the Shoulder of Experts
Zheng-Peng Duan (Nankai University), Chongyi Li (Nankai University)
Image HarmonizationRestorationDiffusion modelContrastive LearningImage
🎯 What it does: This paper proposes DiffRetouch, an image retouching method based on diffusion models, which allows for free adjustment of retouching styles through four attribute coefficients.
DiffScene: Diffusion-Based Safety-Critical Scenario Generation for Autonomous Vehicles
Chejian Xu (University of Illinois at Urbana-Champaign), Bo Li (University of Illinois at Urbana-Champaign)
Autonomous DrivingOptimizationReinforcement LearningDiffusion modelSequential
🎯 What it does: This paper proposes DiffScene, a safety-critical scene generation framework based on diffusion models.
DiffuseHigh: Training-Free Progressive High-Resolution Image Synthesis Through Structure Guidance
Younghyun Kim (Sungkyunkwan University), Eunbyung Park (Sungkyunkwan University)
GenerationData SynthesisSuper ResolutionDiffusion modelImage
🎯 What it does: Using a pre-trained text-to-image diffusion model, we propose a training-free progressive high-resolution image generation method called DiffuseHigh, which generates higher resolution images guided by low-resolution images.
Diffusion Model Patching via Mixture-of-Prompts
Seokil Ham (Korea Advanced Institute of Science and Technology), Changick Kim (Korea Advanced Institute of Science and Technology)
GenerationData SynthesisTransformerPrompt EngineeringDiffusion modelImageText
🎯 What it does: By inserting a small number of learnable prompts into the input space of a pre-trained diffusion model and using noise-level-based dynamic gating to achieve time-step mixture-of-prompts, the generation quality of the already converged model is further improved without updating the main model parameters.
Diffusion Models for Attribution
Xiongren Chen (University of South Australia), Anthony Walsh (Green Triangle Forest Industries Hub)
SegmentationExplainability and InterpretabilityConvolutional Neural NetworkDiffusion modelImage
🎯 What it does: Proposes an explainable AI method based on diffusion models, generating pixel-level attribution maps through noise perturbation.
Diffusion Prior Interpolation for Flexibility Real-World Face Super-Resolution
Jiarui Yang (Nankai University), Shu-Tao Xia (Tsinghua University)
RestorationSuper ResolutionDiffusion modelImage
🎯 What it does: Proposes the Diffusion Prior Interpolation (DPI) method, which utilizes the prior of a pre-trained diffusion model to achieve facial super-resolution, and balances consistency and diversity through a staged conditional mask and a condition correction mechanism.
Diffusion-Based Active Learning for Distributed Client Manifolds
Kwang In Kim (POSTECH)
Federated LearningDiffusion modelImage
🎯 What it does: This paper proposes a client-layer active learning algorithm in a privacy-preserving distributed learning environment, which constructs client manifolds using pseudo-parameters and evaluates the uncertainty reduction of labels on the global model through anisotropic diffusion, thereby selecting the most impactful and diverse clients and local samples for labeling.
Diffusion-based Semantic Outlier Generation via Nuisance Awareness for Out-of-Distribution Detection
Suhee Yoon (LG AI Research), Woohyung Lim (LG AI Research)
GenerationAnomaly DetectionDiffusion modelImage
🎯 What it does: This paper proposes the SONA method, which utilizes diffusion models to perform fine-grained perturbations in semantic and irrelevant areas, generating OOD samples that retain ID noise similarity while exhibiting significant semantic differences, thereby enhancing OOD detection performance.
Diffusion-based Synthetic Data Generation for Visible-Infrared Person Re-Identification
Wenbo Dai (Nanjing Tech University), Zhihang Li (Chinese Academy of Sciences)
RecognitionGenerationData SynthesisRetrievalDiffusion modelImageMultimodality
🎯 What it does: A visible-infrared person re-identification data augmentation framework called DiVE based on diffusion models is proposed, which automatically generates a large number of identity-preserving RGB-IR image pairs.
DiffusionREC: Diffusion Model with Adaptive Condition for Referring Expression Comprehension
Jingcheng Ke (Guangdong University of Technology), Jie Wen (Harbin Institute of Technology)
RecognitionObject DetectionTransformerVision Language ModelDiffusion modelImageMultimodality
🎯 What it does: A framework for Referring Expression Comprehension based on diffusion models, called DiffusionREC, is proposed and implemented, which uses random noise boxes to gradually denoise and locate the target.
Digging into Intrinsic Contextual Information for High-fidelity 3D Point Cloud Completion
Jisheng Chu (Harbin Institute of Technology), Xiaopeng Fan (Harbin Institute of Technology)
GenerationData SynthesisGraph Neural NetworkDiffusion modelPoint Cloud
🎯 What it does: A two-stage high-fidelity point cloud completion framework is proposed: first, a diffusion model generates a rough complete point cloud, and then a context-aware refiner (CRef) refines the rough results by combining short-range and long-range contextual information.
DigitalLLaVA: Incorporating Digital Cognition Capability for Physical World Comprehension in Multimodal LLMs
Shiyu Li (Peking University), Jie Chen (Peking University)
TransformerLarge Language ModelVision Language ModelContrastive LearningImageTextMultimodality
🎯 What it does: This paper presents a digital cognitive enhancement multimodal large language model, DigitalLLaVA, which can accurately read physical digital measuring instruments.
Dimension Reduction for Symbolic Regression
Paul Kahlmeyer (Friedrich Schiller University Jena), Joachim Giesen (Friedrich Schiller University Jena)
OptimizationTabularPhysics Related
🎯 What it does: This paper proposes a dimension reduction method based on beam search and function dependency metrics to improve the recovery rate of symbolic regression.
Dimension Reduction with Locally Adjusted Graphs
Yingfan Wang (Duke University), Cynthia Rudin (Duke University)
GraphTabular
🎯 What it does: A dynamic local adjustment graph dimensionality reduction algorithm called LocalMAP is proposed and implemented, which can identify and remove misclassified neighboring edges during the dimensionality reduction process, dynamically resampling distant edges to achieve a clearer cluster distribution.
DiMSOD: A Diffusion-Based Framework for Multi-Modal Salient Object Detection
Shuo Zhang (East China Normal University), Jing Liu (East China Normal University)
Object DetectionTransformerDiffusion modelImageMultimodality
🎯 What it does: This paper proposes DiMSOD, a unified framework based on diffusion models for salient object detection in three modalities: RGB, RGB-D, and RGB-T.
Direct Routing Gradient (DRGrad): A Personalized Information Surgery for Multi-Task Learning (MTL) Recommendations
Yuguang Liu (Whisper Bond Technologies Inc), Luyao Xia (Tongji University)
Recommendation SystemMixture of ExpertsTabular
🎯 What it does: The Direct Routing Gradient (DRGrad) framework is proposed to address the negative transfer and 'oscillation' issues in multi-task learning.
Dis²Booth: Learning Image Distribution with Disentangled Features for Text-to-Image Diffusion Models
Guanqi Ding (University of Chinese Academy of Sciences), Qingming Huang (University of Chinese Academy of Sciences)
GenerationData SynthesisDiffusion modelContrastive LearningImage
🎯 What it does: This paper proposes an unsupervised decomposition method called Dis² Booth, designed to personalize the generation of diverse images containing multiple instances with shared features for diffusion models.
DisCo-DSO: Coupling Discrete and Continuous Optimization for Efficient Generative Design in Hybrid Spaces
Jacob F. Pettit (Lawrence Livermore National Laboratory), Mikel Landajuela (Lawrence Livermore National Laboratory)
OptimizationRecurrent Neural NetworkTransformerReinforcement LearningTabular
🎯 What it does: This paper presents DisCo-DSO, a deep generative model that combines discrete and continuous variables to efficiently solve black-box mixed space optimization problems.
DisCo: Graph-Based Disentangled Contrastive Learning for Cold-Start Cross-Domain Recommendation
Hourun Li (Peking University), Wei Ju (Sichuan University)
Recommendation SystemGraph Neural NetworkContrastive LearningGraph
🎯 What it does: This study addresses the cold start cross-domain recommendation problem and proposes the DisCo framework to learn users' fine-grained intentions and filter out irrelevant collaborative information.
Discovering Conceptual Knowledge with Analytic Ontology Templates for Articulated Objects
Jianhua Sun (Shanghai Jiao Tong University), Cewu Lu (Shanghai Jiao Tong University)
Robotic IntelligenceTransformerReinforcement LearningPoint Cloud
🎯 What it does: This paper proposes a parameterizable and differentiable Analytic Ontology Template (AOT) and the AOTNet based on it, aimed at recognizing and utilizing joint structure and operability information at the conceptual level to achieve perception and interaction with novel joint objects.
Discovering Options That Minimize Average Planning Time
Alexander Ivanov (Brown University), George Konidaris (Brown University)
OptimizationReinforcement Learning from Human Feedback
🎯 What it does: A new option discovery algorithm based on k-medians with penalties (Average Options) is proposed, aiming to minimize the average planning time in multi-objective MDPs.
Discovering Symmetries of ODEs by Symbolic Regression
Paul Kahlmeyer (Friedrich Schiller University Jena), Joachim Giesen (Friedrich Schiller University Jena)
OptimizationExplainability and InterpretabilityTime SeriesPhysics RelatedOrdinary Differential Equation
🎯 What it does: Using a search-based symbolic regression method, we directly seek the Lie point symmetry generators of systems of ordinary differential equations (ODEs), thereby discovering symmetries of nonlinear ODEs that are difficult to solve with traditional computer algebra systems (CAS).
Discrete Curvature Graph Information Bottleneck
Xingcheng Fu (Guangxi Normal University), Xianxian Li (Beihang University)
ClassificationOptimizationExplainability and InterpretabilityGraph Neural NetworkGraph
🎯 What it does: This paper proposes a graph neural network framework called CurvGIB, which combines discrete Ricci curvature with the information bottleneck to learn the optimal information transmission structure relevant to tasks while simultaneously optimizing node representations.
Discrete Prior-Based Temporal-Coherent Content Prediction for Blind Face Video Restoration
Lianxin Xie (South China University of Technology), Hau-San Wong (City University of Hong Kong)
RestorationTransformerGenerative Adversarial NetworkVideo
🎯 What it does: This paper proposes a spatiotemporal consistent content prediction Transformer based on discrete priors (DP-TempCoh) for blind face video restoration, capable of reconstructing high-quality face videos from pre-trained visual and motion dictionaries.
Disentangle Nighttime Lens Flares: Self-supervised Generation-based Lens Flare Removal
Yuwen He (Wuhan University of Science and Technology), Kui Jiang (Harbin Institute of Technology)
RestorationImage
🎯 What it does: A novel unsupervised generative network for removing night lens glare (SGLFR-Net) is proposed, which simultaneously removes glow caused by flash and ghost reflections.
Disentangled Contrastive Bundle Recommendation with Conditional Diffusion
Jiuqiang Li (Southwest Jiaotong University)
Recommendation SystemGraph Neural NetworkDiffusion modelContrastive LearningGraph
🎯 What it does: A bundling recommendation framework DCBR based on conditional diffusion models and dual-layer decoupled contrastive learning has been designed and implemented to denoise the user-bundle interaction graph and learn more robust user and bundle representations.
Disentangled Modeling of Preferences and Social Influence for Group Recommendation
Guangze Ye (East China Normal University), Liang He (East China Normal University)
Recommendation SystemGraph Neural NetworkContrastive LearningGraph
🎯 What it does: A model called DisRec is proposed for group recommendation that simultaneously decouples user preferences and social influences, enhancing group representation through social contrastive learning.
Disentangled Motion Modeling for Video Frame Interpolation
Jaihyun Lew (Seoul National University), Sungroh Yoon (Seoul National University)
Image TranslationRestorationGenerationData SynthesisDiffusion modelOptical FlowVideo
🎯 What it does: The MoMo framework is proposed, using diffusion models to generate bidirectional optical flow for intermediate frame interpolation.
Disentangled Table-Graph Representation for Interpretable Transmission Line Fault Location
Na Yu (Zhejiang University), Mingli Song (Zhejiang University)
OptimizationExplainability and InterpretabilityGraph Neural NetworkGraphTabular
🎯 What it does: The DTG4Power framework is proposed, which converts measurement table data into graphical representations and simultaneously locates fault lines and fault points in the same end-to-end learning.
Disentangling Long-Short Term State Under Unknown Interventions for Online Time Series Forecasting
Ruichu Cai (Guangdong University of Technology), Zhifeng Hao (Guangdong University of Technology)
Anomaly DetectionRecurrent Neural NetworkAuto EncoderTime Series
🎯 What it does: Designed and implemented a Long-Short Term State Decoupling Model (LSTD) for online time series forecasting, capable of decoupling and predicting long-term and short-term latent states in non-stationary environments caused by unknown interventions.
Disentangling Tabular Data Towards Better One-Class Anomaly Detection
Jianan Ye (Xi'an Jiaotong-Liverpool University), Kaizhu Huang (Duke Kunshan University)
Anomaly DetectionTabularFinance Related
🎯 What it does: For a type of classification scenario in table anomaly detection, this paper proposes to implicitly split two non-overlapping and related attribute subsets from normal samples through a bidirectional self-attention module, and to reconstruct the original samples using these subsets, thereby learning the internal attribute associations of normal samples, with the reconstruction error serving as the anomaly discrimination metric.
Disentangling, Amplifying, and Debiasing: Learning Disentangled Representations for Fair Graph Neural Networks
Yeon-Chang Lee (Ulsan National Institute of Science and Technology), Sang-Wook Kim (Hanyang University)
Graph Neural NetworkGraph
🎯 What it does: A graph neural network framework named DAB-GNN is designed to separate, amplify, and remove attribute bias, structural bias, and the potential bias generated by the interaction of both in node embeddings, thereby enhancing the fairness of the model.
Distances Between Top-Truncated Elections of Different Sizes
Piotr Faliszewski (AGH University), Tomasz Wąs (University of Oxford)
Tabular
🎯 What it does: Expanded the voting map framework to handle elections with different scales of candidates/voters and top-truncated voting, and proposed two new distance metrics - positional extension and DAP distance based on diversity, agreement, and polarization characteristics;
Distilling Knowledge from Heterogeneous Architectures for Semantic Segmentation
Yanglin Huang (Xiangtan University), Xieping Gao (Hunan Normal University)
SegmentationKnowledge DistillationConvolutional Neural NetworkTransformerImage
🎯 What it does: This paper proposes a heterogeneous architecture knowledge distillation framework called HeteroAKD, which can efficiently transfer diverse knowledge between different biased networks such as CNNs and Transformers.
Distilling Structured Rationale from Large Language Models to Small Language Models for Abstractive Summarization
Linyong Wang (Northwestern Polytechnical University), Kang Wang
GenerationKnowledge DistillationTransformerLarge Language ModelPrompt EngineeringText
🎯 What it does: A structured reasoning mining and weak gating fusion framework based on large language models (LLM) is proposed (LSR-MWF), which distills three perspectives of structured reasoning (Essential Aspects, Associated Sentences, Triple Entity Relations) generated by a 70B-level LLM into a summary generation model with ≤1B parameters.
Distribution-Consistency-Guided Multi-modal Hashing
Jin-Yu Liu (Beijing Institute of Technology), Rong-Cheng Tu (Beijing Institute of Technology)
RetrievalContrastive LearningImageMultimodality
🎯 What it does: A multi-modal hashing method based on distribution consistency (DCGMH) is proposed to enhance retrieval performance by filtering and reconstructing noisy labels.
Distribution-Driven Dense Retrieval: Modeling Many-to-One Query-Document Relationship
Junfeng Kang (University of Science and Technology of China), Yu Su (Hefei Normal University)
RetrievalKnowledge DistillationGaussian SplattingText
🎯 What it does: Represent queries as vectors and model documents using multivariate Gaussian distributions, calculating the likelihood of the query vector under the document distribution as a relevance score.
Distribution-Level Feature Distancing for Machine Unlearning: Towards a Better Trade-off Between Model Utility and Forgetting
Dasol Choi (Yonsei University), Dongbin Na (POSTECH)
ClassificationRecognitionData-Centric LearningConvolutional Neural NetworkImage
🎯 What it does: The paper proposes a Distribution-Level Feature Distance (DLFD) method to achieve instance-level unlearning (forgetting) in deep networks, effectively forgetting target samples while maintaining the model's performance on the original task.
Distributionally Robust Policy Evaluation and Learning for Continuous Treatment with Observational Data
Cheuk Hang Leung (City University of Hong Kong), Qi Wu (City University of Hong Kong)
OptimizationDrug DiscoveryReinforcement LearningTabularBiomedical Data
🎯 What it does: This paper proposes an offline policy evaluation and learning framework for continuous treatment variables with distribution shift, utilizing a kernelized inverse probability weighting (IPW) method to construct a distribution-robust value estimator and solve for the optimal policy.
DiT4Edit: Diffusion Transformer for Image Editing
Kunyu Feng (Peking University), Zeyu Wang (Hong Kong University of Science and Technology)
Image TranslationGenerationTransformerDiffusion modelImage
🎯 What it does: This paper presents DiT4Edit, an image editing framework based on Diffusion Transformer, which supports high-resolution and arbitrary size editing.
Dive into Aerial Remote Sensing Underwater Depth Estimation with Hyperspectral Imagery
Jiahao Qi (National University of Defense Technology), Ping Zhong (National University of Defense Technology)
Depth EstimationConvolutional Neural NetworkImage
🎯 What it does: A UAV-based hyperspectral-LiDAR dataset ATR-HUDE is proposed, and the first publicly available real-world spectral water depth estimation dataset is constructed.
DiveR-CT: Diversity-enhanced Red Teaming Large Language Model Assistants with Relaxing Constraints
Andrew Zhao (Tsinghua University), Gao Huang (Tsinghua University)
OptimizationAdversarial AttackTransformerLarge Language ModelReinforcement LearningText
🎯 What it does: DiveR-CT is proposed, a framework for automated red team attacks through constraint optimization and dynamic semantic rewards.
DiverSAT: A Novel and Effective Local Search Algorithm for Diverse SAT Problem
Jiaxin Liang (Northeast Normal University), Minghao Yin (Northeast Normal University)
Optimization
🎯 What it does: A local search algorithm called Diver SAT has been developed to find a diverse set of k satisfying solutions in a given CNF formula.
Diverse Rare Sample Generation with Pretrained GANs
Subeen Lee (Korea Advanced Institute of Science and Technology), Jaesik Choi (Korea Advanced Institute of Science and Technology)
GenerationData SynthesisOptimizationFlow-based ModelGenerative Adversarial NetworkImage
🎯 What it does: This paper proposes an algorithm that utilizes a pre-trained GAN and optimizes a multi-objective framework (rarity, diversity, similarity) to generate diverse rare samples without the need for retraining the GAN.
Diversifying Query: Region-Guided Transformer for Temporal Sentence Grounding
Xiaolong Sun (Xi'an Jiaotong University), Gang Hua
RecognitionObject DetectionTransformerVideoText
🎯 What it does: This paper proposes the Region-Guided Transformer (RGTR), which constructs queries through explicitly initialized static and dynamic anchor points, providing regional priors to address the query overlap issue in traditional DETR and achieving diverse, non-overlapping temporal sentence localization.
DivGCL: A Graph Contrastive Learning Model for Diverse Recommendation
Wenwen Gong (China Agricultural University), Lianyong Qi (China University of Petroleum)
Recommendation SystemGraph Neural NetworkContrastive LearningGraph
🎯 What it does: An end-to-end DivGCL model is proposed, which combines DPP likelihood loss with Gaussian noise augmentation in graph contrastive learning, achieving a balance between recommendation accuracy and diversity.
Divide-and-Conquer: Tree-structured Strategy with Answer Distribution Estimator for Goal-Oriented Visual Dialogue
Shuo Cai (University of Chinese Academy of Sciences), Shuhui Wang (Key Lab of Intelligent Information Processing)
TransformerReinforcement LearningVision Language ModelImageText
🎯 What it does: Proposes the Tree-Structured Strategy with Answer Distribution Estimator (TSADE), which guides question generation through a binary search method, allowing the target visual dialogue to converge to a unique candidate object in fewer rounds.
Divide-Solve-Combine: An Interpretable and Accurate Prompting Framework for Zero-shot Multi-Intent Detection
Libo Qin (Central South University), Min Li (National University of Singapore)
Explainability and InterpretabilityTransformerLarge Language ModelPrompt EngineeringText
🎯 What it does: A Divide-Solve-Combine (DSCP) based and interactive DSCP prompting framework is proposed for zero-shot multi-intent detection.
Divide, Conquer and Combine: A Training-Free Framework for High-Resolution Image Perception in Multimodal Large Language Models
Wenbin Wang (Wuhan University), Dacheng Tao (Nanyang Technological University)
RecognitionGenerationRetrievalTransformerLarge Language ModelImageMultimodalityBenchmarkRetrieval-Augmented Generation
🎯 What it does: A training-free framework called Divide-Conquer-Combine (DC2) is proposed and implemented, which enhances the perception and understanding of multimodal large language models for 4K/8K images through recursive chunking, merging similar chunks, generating text descriptions, and utilizing visual memory.
DLF: Disentangled-Language-Focused Multimodal Sentiment Analysis
Pan Wang (University of Pittsburgh), Jingtong Hu (University of Pittsburgh)
ClassificationTransformerVideoMultimodality
🎯 What it does: Proposes the Disentangled-Language-Focused (DLF) framework, which first uses a feature disentanglement module to separate multimodal features into shared and modality-specific spaces, then focuses complementary information from other modalities onto the language modality through the Language-Focused Attractor (LFA), and finally performs hierarchical prediction to enhance the accuracy of multimodal sentiment analysis.
DM-Adapter: Domain-Aware Mixture-of-Adapters for Text-Based Person Retrieval
Yating Liu (Shenzhen International Graduate School Tsinghua University), Qingmin Liao (Shenzhen International Graduate School Tsinghua University)
RetrievalTransformerMixture of ExpertsContrastive LearningText
🎯 What it does: This paper proposes a parameter-efficient transfer learning framework based on CLIP, called DM-Adapter, which uses Sparse Mixture-of-Adapters and Domain-Aware Router to perform fine-grained modeling of portrait features in text retrieval while training only about 16M parameters.
DME-Driver: Integrating Human Decision Logic and 3D Scene Perception in Autonomous Driving
Wencheng Han (University of Macau), Jianbing Shen (University of Macau)
Autonomous DrivingTransformerLarge Language ModelSupervised Fine-TuningVision Language ModelImageMultimodality
🎯 What it does: This paper proposes the DME-Driver two-stage autonomous driving system, which first simulates human driving logic using a large visual language model, and then transforms decisions into precise control commands using a planning-oriented perception model.
DMF-Net: Image-Guided Point Cloud Completion with Dual-Channel Modality Fusion and Shape-Aware Upsampling Transformer
Aihua Mao (South China University of Technology), Ying He (Nanyang Technological University)
RestorationTransformerImagePoint CloudBenchmark
🎯 What it does: This study focuses on the point cloud completion task guided by single-view images, proposing the DMF-Net network to achieve coarse-to-fine completion in two stages.
DMT-RoleBench: A Dynamic Multi-Turn Dialogue Based Benchmark for Role-Playing Evaluation of Large Language Model and Agent
Dingbo Yuan (Ant Group), Song Liu (Ant Group)
TransformerLarge Language ModelTextBenchmark
🎯 What it does: Designed and implemented the DMT-RoleBench benchmark, which evaluates the role-playing capabilities of large language models and agents using dynamic multi-turn dialogues.
Do Not DeepFake Me: Privacy-Preserving Neural 3D Head Reconstruction Without Sensitive Images
Jiayi Kong (Nanyang Technological University), Ying He (Nanyang Technological University)
RecognitionRestorationSafty and PrivacyNeural Radiance FieldImage
🎯 What it does: A two-stage, non-sensitive RGB image 3D head reconstruction method is proposed, which generates rough geometry using rear-view non-sensitive images and then refines the geometry using privacy-removed gradient images.
DocKylin: A Large Multimodal Model for Visual Document Understanding with Efficient Visual Slimming
Jiaxin Zhang (South China University of Technology), Lianwen Jin (South China University of Technology)
RecognitionCompressionComputational EfficiencyTransformerLarge Language ModelSupervised Fine-TuningVision Language ModelImageTextMultimodality
🎯 What it does: DocKylin is proposed, a multimodal large language model specifically designed for visual document understanding (VDU), which combines Adaptive Pixel Slimming (APS) and Dynamic Token Slimming (DTS) to significantly reduce the length of visual token sequences and improve inference efficiency and accuracy.
DocMamba: Efficient Document Pre-training with State Space Model
Pengfei Hu (University of Science and Technology of China), Jianshu Zhang (iFLYTEK Research)
Computational EfficiencyTransformerLarge Language ModelText
🎯 What it does: A document pre-training framework called DocMamba based on a state space model is proposed, achieving linear time complexity for visual text understanding.
Does GCL Need a Large Number of Negative Samples? Enhancing Graph Contrastive Learning with Effective and Efficient Negative Sampling
Yongqi Huang (Tianjin University), Zhen Wang (Northwestern Polytechnical University)
OptimizationRepresentation LearningGraph Neural NetworkContrastive LearningGraph
🎯 What it does: A new negative sample sampling strategy for graph contrastive learning, called E2Neg, is proposed. Theoretical analysis shows that a large number of negative samples can weaken semantic differentiation, so it is changed to only use a small number of high-quality and non-topologically coupled representative negative samples for contrastive learning.
Does GPT Really Get It? A Hierarchical Scale to Quantify Human and AI’s Understanding of Algorithms
Mirabel Reid (Georgia Institute of Technology), Santosh S. Vempala (Georgia Institute of Technology)
TransformerLarge Language ModelTextMultimodality
🎯 What it does: This paper proposes and validates a hierarchical algorithm understanding scale, exploring the understanding of algorithms by large language models and humans.
Does VLM Classification Benefit from LLM Description Semantics?
Pingchuan Ma (CompVis at LMU Munich), Björn Ommer (CompVis at LMU Munich)
ClassificationRecognitionVision Language ModelContrastive LearningImageText
🎯 What it does: This paper proposes a training-free method to select descriptors that can significantly improve the image classification accuracy of Vision-Language models (such as CLIP);
Does Your AI Agent Get You? A Personalizable Framework for Approximating Human Models from Argumentation-based Dialogue Traces
Yinxu Tang (Washington University in St. Louis), William Yeoh (Washington University in St. Louis)
TextSequential
🎯 What it does: This paper proposes the Persona framework, which utilizes probabilistic Bayesian updating and a prospect theory weighting function to dynamically learn and update the psychological model of human users in argumentative dialogues.
DoGA: Enhancing Grounded Object Detection via Grouped Pre-Training with Attributes
Yang Liu (Institute of Computing Technology Chinese Academy of Sciences), Zhiqiang He (Lenovo Ltd)
Object DetectionTransformerLarge Language ModelPrompt EngineeringVision Language ModelImage
🎯 What it does: Proposes the DoGA scheme, which enhances visual-language object detectors through attribute-driven group pre-training, mainly including attribute prompts (context definition and hard negatives), parallel text encoding, entity extraction fusion, and polysemous phrase-level training.
DOGE: LLMs-Enhanced Hyper-Knowledge Graph Recommender for Multimodal Recommendation
Fanshen Meng (Beijing University of Posts and Telecommunications), Budan Wu (Beijing University of Posts and Telecommunications)
Recommendation SystemGraph Neural NetworkLarge Language ModelMultimodalityGraph
🎯 What it does: A DOGE recommendation model is proposed, utilizing large language models to generate semantic modalities and construct a hyper knowledge graph to enhance the performance of multimodal recommendation systems.
DOGR: Leveraging Document-Oriented Contrastive Learning in Generative Retrieval
Penghao Lu (Ant Group), Linjian Mo (Ant Group)
GenerationRetrievalTransformerLarge Language ModelContrastive LearningText
🎯 What it does: A two-stage generative retrieval framework named DOGR is proposed, utilizing document-oriented contrastive learning to enhance generative retrieval performance.
Domain Adaptive Unfolded Graph Neural Networks
Zepeng Zhang (Ecole Polytechnique Federale de Lausanne), Olga Fink (Ecole Polytechnique Federale de Lausanne)
Domain AdaptationGraph Neural NetworkGraph
🎯 What it does: This paper proposes the Cascaded Propagation (CP) strategy, which makes architectural improvements to the Unfolded Graph Neural Network (UGNN) in the unsupervised target domain graph domain adaptation task to enhance cross-domain transfer performance.
Domain Generalized Medical Landmark Detection via Robust Boundary-Aware Pre-Training
Haifan Gong (Shenzhen Research Institute of Big Data), Haofeng Li (Shenzhen Research Institute of Big Data)
Object DetectionDomain AdaptationConvolutional Neural NetworkTransformerSupervised Fine-TuningBiomedical DataMagnetic Resonance ImagingBenchmark
🎯 What it does: A domain generalization framework for medical landmark detection based on single-center data was constructed, utilizing publicly available segmentation data for boundary-aware pre-training, and introducing a mixed loss of log-cosh and MSE during the pre-training phase, followed by fine-tuning in the landmark detection task;