AAAI 2024 Papers — Page 5
AAAI Conference on Artificial Intelligence · 2331 papers
Coverage-Guaranteed Prediction Sets for Out-of-Distribution Data
Xin Zou (Wuhan University), Weiwei Liu (Wuhan University)
Domain AdaptationSupervised Fine-TuningTabular
🎯 What it does: A trustworthy prediction set method under the OOD generalization setting is proposed, and its coverage guarantee is proven.
CoVR: Learning Composed Video Retrieval from Web Video Captions
Lucas Ventura (Université Gustave Eiffel), Gül Varol (Inria)
RetrievalTransformerLarge Language ModelContrastive LearningVideoTextMultimodalityBenchmark
🎯 What it does: A scalable automatically generated composed video retrieval (CoVR) dataset has been constructed, and a model has been trained based on this dataset. The model was then transferred to the composed image retrieval (CoIR) task, achieving leading performance in both zero-shot and fine-tuning settings.
CPN: Complementary Proposal Network for Unconstrained Text Detection
Longhuang Wu (Shopee Pte. Ltd.), Pengfei Xiong (Shopee Pte. Ltd.)
Object DetectionSegmentationConvolutional Neural NetworkImage
🎯 What it does: A Complementary Proposal Network (CPN) that combines segmentation-based and anchor-based methods is proposed for scene text detection.
CR-SAM: Curvature Regularized Sharpness-Aware Minimization
Tao Wu (Missouri University of Science and Technology), Donald C. Wunsch II (Missouri University of Science and Technology)
ClassificationOptimizationConvolutional Neural NetworkTransformerImage
🎯 What it does: A curvature regularization method based on normalized Hessian trace, CR-SAM, is proposed to enhance the generalization ability of deep networks.
CRA-PCN: Point Cloud Completion with Intra- and Inter-level Cross-Resolution Transformers
Yi Rong (Nanjing University), Tong Lu (Nanjing University)
GenerationData SynthesisTransformerPoint Cloud
🎯 What it does: This paper proposes a point cloud completion network CRA-PCN based on Cross-Resolution Transformer, which can effectively aggregate features across different resolution levels and gradually generate complete point clouds.
CREAD: A Classification-Restoration Framework with Error Adaptive Discretization for Watch Time Prediction in Video Recommender Systems
Jie Sun (Kuaishou Technology), Ben Wang (Kuaishou Technology)
Recommendation SystemVideo
🎯 What it does: The CREAD framework is proposed, which predicts video viewing duration using a classification-recovery approach, and addresses the contradiction between learning error and recovery error caused by traditional discretization through Error Adaptive Discretization (EAD).
Critic-Guided Decision Transformer for Offline Reinforcement Learning
Yuanfu Wang (Shanghai Jiao Tong University), Yu Qiao (Shanghai Artificial Intelligence Laboratory)
TransformerReinforcement LearningSequential
🎯 What it does: Proposes the Critic-Guided Decision Transformer (CGDT), which combines RCSL and value function-guided policy learning to address the performance degradation in offline RL caused by the inconsistency between target returns and actual expected returns.
Cross-Class Feature Augmentation for Class Incremental Learning
Taehoon Kim (Seoul National University), Bohyung Han (Seoul National University)
ClassificationKnowledge DistillationAdversarial AttackImage
🎯 What it does: This paper proposes an incremental learning method based on Cross-Class Feature Augmentation (CCFA), which generates samples of new classes by applying adversarial perturbations to features in the feature space of the old model, thereby compensating for the collapse of decision boundaries caused by insufficient samples from old tasks.
Cross-Constrained Progressive Inference for 3D Hand Pose Estimation with Dynamic Observer-Decision-Adjuster Networks
Zhehan Kan (Southern University of Science and Technology), Zhihai He (Pengcheng Laboratory)
Pose EstimationConvolutional Neural NetworkGraph Neural NetworkTransformerPoint CloudMesh
🎯 What it does: A dynamic progressive reasoning framework based on cross constraints (CCPI) is proposed, achieving adaptive improvement in 3D hand pose estimation through an observation-decision-adjustment network.
Cross-Covariate Gait Recognition: A Benchmark
Shinan Zou (Central South University), Jin Tang (Central South University)
RecognitionConvolutional Neural NetworkVideoBenchmark
🎯 What it does: A CCGR cross-covariate gait recognition benchmark is proposed, providing a dataset with millions of samples, 53 types of covariates, and 33 viewpoints;
Cross-Domain Contrastive Learning for Time Series Clustering
Furong Peng (Shanxi University), Feijiang Li (Zhengzhou University of Aeronautics)
Representation LearningConvolutional Neural NetworkRecurrent Neural NetworkContrastive LearningTime Series
🎯 What it does: An end-to-end cross-domain contrastive learning framework CDCC is proposed for clustering time series data.
Cross-Gate MLP with Protein Complex Invariant Embedding Is a One-Shot Antibody Designer
Cheng Tan (Westlake University), Stan Z. Li (Westlake University)
Drug DiscoveryProtein Structure PredictionBiomedical Data
🎯 What it does: A single-step antibody CDR sequence and structure co-design model based on invariant embedding of protein complexes and cross-gated MLP has been designed.
Cross-Layer and Cross-Sample Feature Optimization Network for Few-Shot Fine-Grained Image Classification
Zhen-Xiang Ma (Shandong University), Xin-Shun Xu (Shandong University)
ClassificationMeta LearningConvolutional Neural NetworkImage
🎯 What it does: A cross-layer and cross-sample feature optimization network C2-Net is proposed for few-shot fine-grained image classification, addressing feature noise and matching mismatch issues.
Cross-Modal and Uni-Modal Soft-Label Alignment for Image-Text Retrieval
Hailang Huang (Beihang University), Ziyu Shang (Southeast University)
RetrievalContrastive LearningImageTextMultimodality
🎯 What it does: A CUSA framework is proposed, which uses the soft labels of a single-modal pre-trained model to align cross-modal and single-modal soft labels for the image-text retrieval model, thereby addressing the issues of false negative samples and single-modal semantic loss.
Cross-Modal Feature Distribution Calibration for Few-Shot Visual Question Answering
Jing Zhang (East China University of Science and Technology), Zhe Wang (East China University of Science and Technology)
RecognitionObject DetectionTransformerVision Language ModelImageTextMultimodality
🎯 What it does: This paper proposes a Cross-Modal Feature Distribution Calibration Inference Network (CDCIN), which calibrates multimodal feature distributions using visual information entropy and enhances few-shot visual question answering performance through cross-modal fine-grained interaction.
Cross-Modal Match for Language Conditioned 3D Object Grounding
Yachao Zhang (Tsinghua University), Xiu Li (Tsinghua University)
Object DetectionKnowledge DistillationRepresentation LearningTransformerVision Language ModelTextPoint Cloud
🎯 What it does: This paper proposes a cross-modal matching framework called xM Match for 3D object localization under language conditions, primarily addressing the inconsistency in matching between local visual representations and global sentence descriptions, as well as between visual features and label word feature spaces.
Cross-Sentence Gloss Consistency for Continuous Sign Language Recognition
Qi Rao (University of Technology Sydney), Bang Zhang (Alibaba Group)
RecognitionRecurrent Neural NetworkContrastive LearningVideoMultimodality
🎯 What it does: A cross-sentence holographic consistency method is proposed in continuous sign language recognition, constructing a gloss prototype and enhancing the discriminative power of gloss representations through contrastive learning.
CrossBind: Collaborative Cross-Modal Identification of Protein Nucleic-Acid-Binding Residues
Linglin Jing (Shanghai Artificial Intelligence Laboratory), Siqi Sun (Research Institute of Intelligent Complex Systems, Fudan University)
RecognitionProtein Structure PredictionConvolutional Neural NetworkLarge Language ModelContrastive LearningMultimodalityPoint CloudBiomedical Data
🎯 What it does: A cross-modal framework called CrossBind is proposed, which combines protein structure (atomic point clouds) and sequence (ESM-2 language model) to identify nucleic acid binding residues.
CrystalBox: Future-Based Explanations for Input-Driven Deep RL Systems
Sagar Patel (University of California), Nina Narodytska (VMware Research)
Explainability and InterpretabilitySupervised Fine-TuningReinforcement LearningSequential
🎯 What it does: The CrystalBox framework is proposed, providing post-hoc interpretability for input-driven deep reinforcement learning controllers and generating future causal explanations.
CSL: Class-Agnostic Structure-Constrained Learning for Segmentation Including the Unseen
Hao Zhang (University of Illinois at Urbana-Champaign), Narendra Ahuja (University of Illinois at Urbana-Champaign)
SegmentationDomain AdaptationAnomaly DetectionTransformerImage
🎯 What it does: A pluggable Class-Agnostic Structure-Constrained Learning (CSL) framework is proposed for segmenting unseen categories in OOD, zero-shot semantic segmentation, and domain adaptation tasks.
CTO-SLAM: Contour Tracking for Object-Level Robust 4D SLAM
Xiaohan Li (University of Science and Technology of China), Jun Wu (Fudan University)
Object TrackingPose EstimationDepth EstimationAutonomous DrivingSimultaneous Localization and MappingImageVideo
🎯 What it does: This paper proposes a contour tracking-based object-level 4D SLAM system called CTO-SLAM, which can simultaneously estimate camera pose and dynamic object motion in dynamic scenes, and generate a sparse 4D map.
CUDC: A Curiosity-Driven Unsupervised Data Collection Method with Adaptive Temporal Distances for Offline Reinforcement Learning
Chenyu Sun (Nanyang Technological University), Chunyan Miao (Agency for Science Technology and Research)
Reinforcement LearningContrastive LearningSequential
🎯 What it does: This paper proposes a curiosity-driven unsupervised data collection method called CUDC, which utilizes an accessibility module to adaptively adjust the time step, enhancing the data quality and generalization ability of multi-task offline reinforcement learning.
Cumulative Difference Learning VAE for Time-Series with Temporally Correlated Inflow-Outflow
Tianchun Li (Purdue University), Xiaoqian Wang (Purdue University)
GenerationData SynthesisRecurrent Neural NetworkAuto EncoderGenerative Adversarial NetworkTime SeriesSequentialFinance Related
🎯 What it does: A differential learning-based variational autoencoder (DT-VAE) is proposed, which addresses the issue of error accumulation in time series generation by introducing an inflow-outflow structure, and can be combined with GANs to enhance generation quality.
Cumulative Regret Analysis of the Piyavskii–Shubert Algorithm and Its Variants for Global Optimization
Kaan Gokcesu (Regrify AI), Hakan Gökcesu
Optimization
🎯 What it does: Theoretical analysis of cumulative regret for Piyavskii-Shubert and its variants in global optimization, and a simpler variant is proposed that can achieve the same optimal bound.
Curriculum-Enhanced Residual Soft An-Isotropic Normalization for Over-Smoothness in Deep GNNs
Jin Li (Fuzhou University), Yang-Geng Fu (Fuzhou University)
ClassificationOptimizationGraph Neural NetworkGraph
🎯 What it does: This paper proposes a Residual Enhanced Soft Graph Normalization Layer (R-SoftGraphAIN) and a label smoothing-based curriculum learning framework (SmoothCurriculum) to address the issues of over-smoothing and optimization difficulties in deep graph neural networks.
Curvature-Invariant Adversarial Attacks for 3D Point Clouds
Jianping Zhang (Chinese University of Hong Kong), Michael R. Lyu (Chinese University of Hong Kong)
Adversarial AttackPoint Cloud
🎯 What it does: A 3D point cloud adversarial attack method based on local curvature invariance (CIM) is designed, which regularizes the gradient on the tangent plane and normal direction after the point cloud coordinate transformation to maintain local geometric shapes, achieving more imperceptible adversarial perturbations.
Curved Representation Space of Vision Transformers
Juyeop Kim (Yonsei University), Jong-Seok Lee (Yonsei University)
Representation LearningAdversarial AttackConvolutional Neural NetworkTransformerImage
🎯 What it does: This paper explores the output representation space curvature of Vision Transformers under input linear perturbations and uses this to explain their high robustness and low confidence phenomena.
Customizing Language Model Responses with Contrastive In-Context Learning
Xiang Gao (Intuit), Kamalika Das (Intuit)
TransformerLarge Language ModelPrompt EngineeringContrastive LearningTextChain-of-Thought
🎯 What it does: A context learning method is proposed that utilizes positive and negative contrast examples and incorporates reasoning analysis to guide large language models in generating answers that better align with user preferences.
CutFreq: Cut-and-Swap Frequency Components for Low-Level Vision Augmentation
Hongyang Chen (Xi'an Jiaotong University), Kaisheng Ma (Tsinghua University)
RestorationSegmentationImage
🎯 What it does: A CutFreq data augmentation method is proposed for cutting and exchanging frequency components in the frequency domain to enhance the performance of low-level visual tasks.
CUTS+: High-Dimensional Causal Discovery from Irregular Time-Series
Yuxiao Cheng (Tsinghua University), Qionghai Dai (Tsinghua University)
Graph Neural NetworkTime Series
🎯 What it does: This paper proposes CUTS+, a causal discovery framework for high-dimensional irregular time series, which integrates coarse-fine level discovery (C2FD) and message passing graph neural networks (MPGNN) to achieve scalable causal graph learning and missing value imputation.
Cycle Self-Refinement for Multi-Source Domain Adaptation
Chaoyang Zhou (Wuhan University), Yong Luo (Wuhan University)
ClassificationDomain AdaptationConvolutional Neural NetworkImage
🎯 What it does: A multi-source domain adaptation method with a cyclic self-improvement approach is proposed, utilizing source-specific networks and domain integration networks to generate high-confidence pseudo-labels through instance-level voting, and mutually guiding each other in a cycle to enhance target domain performance.
Cycle-Consistency Learning for Captioning and Grounding
Ning Wang (Huawei Inc), Mingbo Jia (Huawei Inc)
Object DetectionGenerationTransformerVision Language ModelImageTextMultimodality
🎯 What it does: The CyCo framework is proposed, which trains visual grounding and regional image captioning through cyclic consistency for collaborative learning of both tasks.
CycleVTON: A Cycle Mapping Framework for Parser-Free Virtual Try-On
Chenghu Du (Wuhan University of Technology), Shengwu Xiong (Wuhan University of Technology)
Image TranslationGenerationConvolutional Neural NetworkGenerative Adversarial NetworkImage
🎯 What it does: A parser-free cyclic mapping framework CycleVTON is proposed, which implements clothing deformation and virtual fitting using a single network.
D3: A Methodological Exploration of Domain Division, Modeling, and Balance in Multi-Domain Recommendations
Pengyue Jia (City University of Hong Kong), Ruiming Tang (Huawei Noah's Ark Lab)
Recommendation SystemTabular
🎯 What it does: A general flexible framework D3 is proposed to automatically partition domains in multi-domain recommendation, integrate common and differential features, and dynamically adjust loss weights during training.
DA-Net: A Disentangled and Adaptive Network for Multi-Source Cross-Lingual Transfer Learning
Ling Ge (Beihang University), Hong Zhang (National Computer Network Emergency Response Technical Team Coordination Center of China)
ClassificationDomain AdaptationTransformerContrastive LearningText
🎯 What it does: A multi-source cross-lingual transfer learning framework DA-Net is proposed to address the issues of source language interference caused by shared encoders and the source-target language gap.
DAG-Aware Variational Autoencoder for Social Propagation Graph Generation
Dongpeng Hou (Northwestern Polytechnical University), Zhen Wang (Northwestern Polytechnical University)
GenerationData SynthesisRecurrent Neural NetworkGraph Neural NetworkAuto EncoderGraph
🎯 What it does: This paper proposes a user feature attention-based DAG-Aware Variational Autoencoder (DAVA), which utilizes nearly a million real social network propagation data to generate large-scale, highly realistic propagation DAG graphs.
DALDet: Depth-Aware Learning Based Object Detection for Autonomous Driving
Ke Hu (University of Science and Technology of China), Yi Kang (University of Science and Technology of China)
Object DetectionAutonomous DrivingConvolutional Neural NetworkImage
🎯 What it does: This paper proposes an efficient 2D object detection framework called DALDet, which utilizes depth maps to output target distance information based on 2D detection.
DanceAnyWay: Synthesizing Beat-Guided 3D Dances with Randomized Temporal Contrastive Learning
Aneesh Bhattacharya (Indian Institute of Information Technology Naya Raipur), Aniket Bera (Adobe Research)
GenerationData SynthesisTransformerGenerative Adversarial NetworkContrastive LearningVideoAudio
🎯 What it does: This work proposes DanceAnyWay, a two-stage hierarchical generation framework that synchronously generates 3D dance movements using the beat information from audio.
DanceMVP: Self-Supervised Learning for Multi-Task Primitive-Based Dance Performance Assessment via Transformer Text Prompting
Yun Zhong (Imperial College London), Yiannis Demiris (Imperial College London)
RecognitionRecurrent Neural NetworkTransformerPrompt EngineeringContrastive LearningVideoMultimodality
🎯 What it does: A multi-task dance evaluation framework called DanceMVP based on self-supervised learning is proposed, utilizing Transformer text prompts to accomplish dance vocabulary recognition, scoring, and rhythm assessment.
DART: Dual-Modal Adaptive Online Prompting and Knowledge Retention for Test-Time Adaptation
Zichen Liu (Peking University), Jiahuan Zhou (Peking University)
Domain AdaptationTransformerPrompt EngineeringVision Language ModelContrastive LearningImageMultimodality
🎯 What it does: A dual-modal online prompting and knowledge retention method (DART) is proposed, which adaptively adjusts CLIP during testing, enhancing the model's performance under distribution shifts by learning text and image prompts.
Data Adaptive Traceback for Vision-Language Foundation Models in Image Classification
Wenshuo Peng (OpenGVLab Shanghai AI Laboratory), Yu Qiao (OpenGVLab Shanghai AI Laboratory)
ClassificationPrompt EngineeringContrastive LearningImageText
🎯 What it does: The Data Adaptive Traceback (DAT) framework is proposed, which enhances downstream image classification performance by reusing pre-trained image-text data through adaptive sampling, semi-supervised learning, and contrastive learning.
Data Augmented Graph Neural Networks for Personality Detection
Yangfu Zhu (Beijing University of Posts and Telecommunications), Bin Wu (Beijing University of Posts and Telecommunications)
ClassificationRecognitionGraph Neural NetworkTextGraph
🎯 What it does: Proposes Semi-PerGCN, a semi-supervised graph neural network that utilizes multi-view graph enhancement and consistency learning from a large number of unlabeled users for personality detection.
Data Disparity and Temporal Unavailability Aware Asynchronous Federated Learning for Predictive Maintenance on Transportation Fleets
Leonie von Wahl (Volkswagen Group), Nicolas Tempelmeier (Penn State University)
Federated LearningRecurrent Neural NetworkTabularTime Series
🎯 What it does: This paper proposes an asynchronous federated learning framework for predictive maintenance of transportation fleets, called DAAFL, which addresses the issues of uneven data distribution and the unavailability of vehicle time series data.
Data Distribution Distilled Generative Model for Generalized Zero-Shot Recognition
Yijie Wang (Chongqing University), Sheng Huang (Chongqing University)
RecognitionGenerationKnowledge DistillationDiffusion modelGenerative Adversarial NetworkImage
🎯 What it does: A novel end-to-end generalized zero-shot learning framework, D3GZSL, is designed and implemented, which enhances the performance of generative models under real distributions through the combination of knowledge distillation and anomaly distribution detection.
Data Poisoning to Fake a Nash Equilibria for Markov Games
Young Wu (University of Wisconsin Madison), Qiaomin Xie (University of Wisconsin Madison)
OptimizationAdversarial AttackReinforcement LearningTabular
🎯 What it does: This paper studies offline multi-agent reinforcement learning (two-player zero-sum Markov games), utilizing data pollution to induce learners to achieve a specified (possibly not real) unique Markov perfect Nash equilibrium.
Data Roaming and Quality Assessment for Composed Image Retrieval
Matan Levy (Hebrew University of Jerusalem), Dani Lischinski (Hebrew University of Jerusalem)
RetrievalTransformerVision Language ModelImageTextMultimodality
🎯 What it does: This paper proposes a combination image retrieval (CoIR) dataset LaSCo, which is ten times larger than existing datasets, and generates CoIR triplets from VQA data through an automated method (Data Roaming); it also introduces a modality redundancy detection method and an early fusion model CASE based on BLIP, achieving stronger retrieval performance with the reverse query target.
Data Shunt: Collaboration of Small and Large Models for Lower Costs and Better Performance
Dong Chen (Zhejiang University), Siliang Tang (Ant Group)
ClassificationComputational EfficiencyKnowledge DistillationTransformerLarge Language ModelPrompt EngineeringImageText
🎯 What it does: This paper proposes the Data Shunt paradigm, which uses the confidence of a small model to determine whether to invoke a large model, and achieves collaboration between small and large models through Prompt Pruning and 2-Stage Confidence Distillation, significantly improving performance while reducing the cost of invoking the large model.
Data-Augmented Curriculum Graph Neural Architecture Search under Distribution Shifts
Yang Yao (Tsinghua University), Hong Mei (Peking University)
ClassificationNeural Architecture SearchGraph Neural NetworkDiffusion modelGraph
🎯 What it does: This paper proposes a data augmentation and curriculum-based graph neural network architecture search (DCGAS) to address the issue of distribution shift in graph classification, enabling the learning of customized architectures for each graph.
Data-Driven Knowledge-Aware Inference of Private Information in Continuous Double Auctions
Lvye Cui (Beijing Institute of Technology), Haoran Yu (Beijing Institute of Technology)
Recurrent Neural NetworkTabular
🎯 What it does: This paper proposes a knowledge-aware machine learning framework that infers the private cost vector of sellers in continuous double auctions by utilizing sellers' inquiry behavior and the global inquiry-cost ratio distribution.
Data-Free Generalized Zero-Shot Learning
Bowen Tang (Beihang University), Dong Xu (The University of Hong Kong)
ClassificationRecognitionGenerationTransformerVision Language ModelGenerative Adversarial NetworkImage
🎯 What it does: A data-free zero-shot learning (DFZSL) framework is proposed without any real image data, achieving recognition of new categories through the recovery, alignment, and generation of the CLIP base classifier.
Data-Free Hard-Label Robustness Stealing Attack
Xiaojian Yuan (University of Science and Technology of China), Nenghai Yu (University of Science and Technology of China)
Adversarial AttackConvolutional Neural NetworkGenerative Adversarial NetworkImage
🎯 What it does: A data-independent model stealing attack that only queries hard labels is proposed, capable of simultaneously stealing the target model's accuracy and robustness.
DC-NAS: Divide-and-Conquer Neural Architecture Search for Multi-Modal Classification
Xinyan Liang (Shanxi University), Yuhua Qian (Shanxi University)
ClassificationNeural Architecture SearchMultimodality
🎯 What it does: A novel evolutionary-based multimodal classification neural architecture search method DC-NAS is proposed, which employs partitioned-parallel subpopulations and knowledge transfer to accelerate the search while maintaining performance.
DCLP: Neural Architecture Predictor with Curriculum Contrastive Learning
Shenghe Zheng (Harbin Institute of Technology), Tianyu Mu (Harbin Institute of Technology)
Neural Architecture SearchRecurrent Neural NetworkGraph Neural NetworkContrastive LearningImage
🎯 What it does: A neural network predictor DCLP based on contrastive learning is designed and implemented, and the difficulty of positive samples is scheduled through curriculum learning, thereby improving prediction performance under the condition of using only a very small amount of trained network data.
DDAE: Towards Deep Dynamic Vision BERT Pretraining
Honghao Chen (Institute of Automation, Chinese Academy of Sciences), Kaiqi Huang (CAS Center for Excellence in Brain Science and Intelligence Technology)
Object DetectionSegmentationKnowledge DistillationRepresentation LearningTransformerContrastive LearningImage
🎯 What it does: This paper proposes a DDAE framework based on deep dynamic supervision and dynamic loss, improving the self-supervised pre-training process in Masked Image Modeling (MIM).
DDDM-VC: Decoupled Denoising Diffusion Models with Disentangled Representation and Prior Mixup for Verified Robust Voice Conversion
Ha-Yeong Choi (Korea University), Seong-Whan Lee (Korea University)
GenerationData SynthesisDiffusion modelStochastic Differential EquationAudio
🎯 What it does: This paper proposes a Decoupled Denoising Diffusion Model (DDDM) and its application in voice conversion (DDDM-VC), achieving decoupling and style transfer of attributes such as content, pitch, and speaker.
De-biased Attention Supervision for Text Classification with Causality
Yiquan Wu (Zhejiang University), Kun Kuang (Zhejiang University)
ClassificationExplainability and InterpretabilityRecurrent Neural NetworkTransformerText
🎯 What it does: To address the label bias and word frequency bias in attention supervision (AS) for text classification tasks, a Debiased Attention Supervision (DAS) method is proposed.
Dealing with Numeric and Metric Time Constraints in PDDL3 via Compilation to Numeric Planning
Luigi Bonassi (University of Brescia), Enrico Scala (University of Brescia)
Tabular
🎯 What it does: This study investigates a technique for compiling PDDL3 planning problems that include mixed propositional and numerical conditions, as well as metric time constraints, into unconstrained numerical planning problems, enabling the use of existing numerical planners for solving them.
Debiased Novel Category Discovering and Localization
Juexiao Feng (Tsinghua University), Guiguang Ding (Tsinghua University)
Object DetectionContrastive LearningImageBenchmark
🎯 What it does: This paper proposes Debiased Region Mining and semi-supervised instance contrastive learning, constructing a target detection framework that can simultaneously detect known categories and actively discover, locate, and cluster unknown categories.
Debiasing Multimodal Sarcasm Detection with Contrastive Learning
Mengzhao Jia (Shandong University), Liqiang Jing (University of Texas at Dallas)
ClassificationDomain AdaptationTransformerLarge Language ModelContrastive LearningTextMultimodality
🎯 What it does: A debiased multimodal sarcasm detection framework is proposed, and an OOD evaluation task is designed to address the generalization problem caused by text bias.
DeblurSR: Event-Based Motion Deblurring under the Spiking Representation
Chen Song (University of Texas at Austin), Qixing Huang (University of Texas at Austin)
RestorationSuper ResolutionConvolutional Neural NetworkSpiking Neural NetworkImageVideo
🎯 What it does: Using event camera data, convert single-frame blurred images into high frame rate sharpened videos;
Decentralized Gradient-Free Methods for Stochastic Non-smooth Non-convex Optimization
Zhenwei Lin (Shanghai University of Finance and Economics), Luo Luo (Fudan University)
OptimizationAdversarial AttackImageTabular
🎯 What it does: This paper proposes two distributed zeroth-order optimization algorithms, DGFM and DGFM+, for solving stochastic non-convex optimization problems that are Lipschitz continuous but not smooth or convex.
Decentralized Monte Carlo Tree Search for Partially Observable Multi-Agent Pathfinding
Alexey Skrynnik (Federal Research Center for Computer Science and Control of Russian Academy of Sciences), Aleksandr Panov (Federal Research Center for Computer Science and Control of Russian Academy of Sciences)
OptimizationReinforcement Learning
🎯 What it does: A decentralized lifelong multi-agent path planning method based on MCTS and a lightweight learnable policy (MATS-LP) is proposed.
Decentralized Scheduling with QoS Constraints: Achieving O(1) QoS Regret of Multi-Player Bandits
Qingsong Liu (Tsinghua University), Zhixuan Fang (Shanghai Qi Zhi Institute)
Optimization
🎯 What it does: This paper proposes a completely decentralized multi-player multi-armed bandit (MP-MAB) algorithm, aiming to ensure that each player achieves a preset QoS (Quality of Service) threshold in terms of long-term average rewards, and proves that its QoS regret can reach a constant order (O(1)), thereby achieving fair and stable resource allocation in multi-user systems.
Decentralized Sum-of-Nonconvex Optimization
Zhuanghua Liu (CNRS at CREATE), Bryan Kian Hsiang Low (National University of Singapore)
OptimizationTabular
🎯 What it does: This paper addresses the sum-of-nonconvex optimization problem in decentralized environments. It first provides a linear convergence analysis of PMGT-SVRG and proposes a new accelerated stochastic decentralized first-order algorithm, PMGT-KatyushaX.
Deciphering Compatibility Relationships with Textual Descriptions via Extraction and Explanation
Yu Wang (University of California San Diego), Julian McAuley (University of California San Diego)
Recommendation SystemExplainability and InterpretabilityTransformerLarge Language ModelSupervised Fine-TuningText
🎯 What it does: A PFE dataset was constructed and a two-stage pipeline model was proposed to generate compatible natural language explanations for clothing pairs.
Deciphering Raw Data in Neuro-Symbolic Learning with Provable Guarantees
Lue Tao (Nanjing University), Yuan Jiang (Nanjing University)
ClassificationImage
🎯 What it does: The paper studies the impact of knowledge bases on learning outcomes in neural symbolic learning, proposing a theoretical evaluation of knowledge bases through probability matrices and rank criteria, and validating their feasibility across different tasks.
Decoding AI’s Nudge: A Unified Framework to Predict Human Behavior in AI-Assisted Decision Making
Zhuoyan Li (Purdue University), Ming Yin (Purdue University)
Recommendation SystemExplainability and InterpretabilitySupervised Fine-TuningTabular
🎯 What it does: Designed and validated an interpretable unified framework for predicting and explaining human decision-making behavior under different forms of AI assistance.
Decoding Global Preferences: Temporal and Cooperative Dependency Modeling in Multi-Agent Preference-Based Reinforcement Learning
Tianchen Zhu (Beihang University), Jianxin Li (Beihang University)
TransformerReinforcement LearningSequential
🎯 What it does: A Transformer-based multi-agent preference learning framework called MAPT is proposed to learn reward functions from human preferences.
Decomposing Constraint Networks for Calculating c-Representations
Marco Wilhelm (TU Dortmund University), Gabriele Kern-Isberner (TU Dortmund University)
🎯 What it does: This paper proposes a framework based on Safe Covers and Constraint Networks for locally computing the c-representations of Conditional Belief Bases, and proves that global skeptical c-inference can be achieved through local computation.
Decomposing Semantic Shifts for Composed Image Retrieval
Xingyu Yang (Wuhan University), Jing Zhang (University of Sydney)
RetrievalTransformerVision Language ModelImageText
🎯 What it does: This paper proposes treating text in composite image retrieval as instructions and splits the retrieval process into two steps: degradation (reference → visual prototype) and upgrading (visual prototype → target) through the Semantic Shift Network (SSN). In this process, the text is deconstructed in both forward (upgrading) and backward (degradation) directions, guiding the generation of visual prototypes and ultimately obtaining representations of target images. End-to-end training is achieved through cross-entropy retrieval loss and KL regularization.
Decomposing Temporal Equilibrium Strategy for Coordinated Distributed Multi-Agent Reinforcement Learning
Chenyang Zhu (Changzhou University), Zhihao Jiang (ShanghaiTech University)
Reinforcement LearningTabular
🎯 What it does: A hierarchical multi-agent reinforcement learning framework named MATEA is proposed, which first synthesizes time-balanced strategies through parity games and then decomposes them into reward machines for each agent to achieve distributed training.
Decouple Content and Motion for Conditional Image-to-Video Generation
Cuifeng Shen (Peking University), Jinzhi Wang (Peking University)
GenerationData SynthesisCompressionDiffusion modelAuto EncoderVideoText
🎯 What it does: A decoupled content and motion conditional image-to-video generation framework (D-VDM and ED-VDM) is proposed, which achieves efficient and temporally consistent video synthesis by splitting the video into the first frame (content) and motion vectors + residuals (motion) and performing diffusion generation in a compressed space.
Decoupled Contrastive Learning for Long-Tailed Recognition
Shiyu Xuan (Peking University), Shiliang Zhang (Peking University)
ClassificationRecognitionKnowledge DistillationContrastive LearningImage
🎯 What it does: For long-tail visual classification, we propose Decoupled Supervised Contrastive Loss (DSCL) and Patch-Based Self Distillation (PBSD), which enhance model performance by decoupling the gradients of two types of positive samples and leveraging shared visual patterns between head and tail classes.
Decoupled Contrastive Multi-View Clustering with High-Order Random Walks
Yiding Lu (Sichuan University), Xi Peng (Sichuan University)
Representation LearningContrastive LearningImageText
🎯 What it does: This paper proposes DIVIDE, a robust multi-view clustering method that utilizes random walks to identify high-order neighbors and simultaneously corrects false negatives and false positives through decoupled contrastive learning.
Decoupled Optimisation for Long-Tailed Visual Recognition
Cong Cong (University of New South Wales), Yang Song (Macquarie University)
ClassificationRecognitionOptimizationConvolutional Neural NetworkMixture of ExpertsImage
🎯 What it does: This paper proposes a multi-stage parameter decoupling and optimization (DO) framework to balance the importance of different category groups (diverse, ordinary, minority) in long-tail datasets, enhancing the recognition performance of each category.
Decoupled Textual Embeddings for Customized Image Generation
Yufei Cai (Chinese Academy of Sciences), Wangmeng Zuo (Harbin Institute of Technology)
GenerationData SynthesisDiffusion modelImageText
🎯 What it does: Proposes the DETEX method, which utilizes multi-word embeddings to achieve text-customized image generation.
Decoupled Training: Return of Frustratingly Easy Multi-Domain Learning
Ximei Wang (Tencent Inc), Jie Jiang (Tsinghua University)
Domain AdaptationRecommendation SystemImageTextMultimodality
🎯 What it does: A three-stage separation training method called D-Train is proposed, which first pre-trains using a shared bottom network, then splits multiple heads for each domain, and finally freezes the bottom layer for domain-specific fine-tuning, achieving multi-domain learning.
Decoupling Degradations with Recurrent Network for Video Restoration in Under-Display Camera
Chengxu Liu (Xi'an Jiaotong University), Xueming Qian (MEGVII Technology)
RestorationRecurrent Neural NetworkOptical FlowVideo
🎯 What it does: To address the distortion caused by diffraction in camera (UDC) videos on full-screen displays, the authors propose a video restoration network called D2RNet.
Decoupling Meta-Reinforcement Learning with Gaussian Task Contexts and Skills
Hongcai He (University of Electronic Science and Technology of China), Jie Shao (Sichuan Artificial Intelligence Research Institute)
Meta LearningReinforcement LearningContrastive LearningSequential
🎯 What it does: This paper studies an offline meta reinforcement learning framework called DCMRL, which enhances the adaptability to unseen tasks by implementing separate exploration and learning through contrastive learning of task context and skills, along with Gaussian quantization VAE.
Decoupling Representation and Knowledge for Few-Shot Intent Classification and Slot Filling
Jie Han (Huazhong University of Science and Technology), Ruixuan Li
ClassificationDomain AdaptationRepresentation LearningTransformerSupervised Fine-TuningText
🎯 What it does: This paper studies a model called JMRM that simultaneously performs intent classification and slot filling in few-shot scenarios, and directly transfers from the source domain to the target domain after training.
Deep Active Learning with Noise Stability
Xingjian Li (Carnegie Mellon University), Chengzhong Xu (University of Macau)
ClassificationSegmentationConvolutional Neural NetworkContrastive LearningImageText
🎯 What it does: A deep active learning algorithm based on noise stability is proposed, which estimates uncertainty and performs diversity sampling by utilizing the feature bias caused by perturbations to the model parameters.
Deep Contrastive Graph Learning with Clustering-Oriented Guidance
Mulin Chen (Northwestern Polytechnical University), Xuelong Li (Northwestern Polytechnical University)
Representation LearningGraph Neural NetworkAuto EncoderContrastive LearningImageText
🎯 What it does: A deep contrastive learning-based framework for clustering non-graph data, DCGL, is proposed, which utilizes a pseudo dual-branch network to simultaneously learn structural and attribute representations, and achieves adaptive graph learning through the fusion of local and global graphs.
Deep Copula-Based Survival Analysis for Dependent Censoring with Identifiability Guarantees
Weijia Zhang (University of Newcastle), Xuanhui Zhang (Nanjing University)
TabularTime SeriesElectronic Health Records
🎯 What it does: A deep Copula-based survival analysis method is proposed, which can automatically learn the dependency structure from data without pre-specifying the Copula in the presence of dependent censoring;
Deep Hierarchical Video Compression
Ming Lu (Nanjing University), Zhan Ma (Purdue University)
CompressionAuto EncoderVideo
🎯 What it does: A learning-based video compression framework (DHVC) based on hierarchical probabilistic prediction is designed and implemented, which models the conditional probability of future frames in the latent space using multi-scale variational autoencoders, thus achieving video coding without motion estimation.
Deep Homography Estimation for Visual Place Recognition
Feng Lu (Tsinghua University), Chun Yuan (Tsinghua University)
RecognitionRetrievalTransformerImage
🎯 What it does: A deep homography estimation network (DHE) based on Transformer is proposed for two-stage retrieval in visual location recognition, replacing RANSAC with differentiable homography fitting for geometric verification, and designing an unsupervised REI loss for end-to-end joint training.
Deep Incomplete Multi-View Learning Network with Insufficient Label Information
Zhangqi Jiang (National University of Defense Technology), Xinyan Liang (Shanxi University)
ClassificationRepresentation LearningGraph Neural NetworkAuto EncoderContrastive LearningImage
🎯 What it does: This paper proposes a deep network named DIMvLN to address the multi-view semi-supervised classification problem with simultaneously missing view features and scarce labels;
Deep Linear Array Pushbroom Image Restoration: A Degradation Pipeline and Jitter-Aware Restoration Network
Zida Chen (Zhejiang University), Shiqi Chen (Zhejiang University)
RestorationOptical FlowImage
🎯 What it does: A two-stage JARNet network is proposed, which combines optical flow correction and spatial frequency domain residual modules to recover distortions and blurriness in line-scan camera images.
Deep Quantum Error Correction
Yoni Choukroun (Blavatnik School of Computer Science Tel Aviv University), Lior Wolf (Blavatnik School of Computer Science Tel Aviv University)
TransformerPhysics Related
🎯 What it does: An end-to-end deep quantum error correction decoder (QECCT) based on Transformer is proposed, which first predicts noise and then refines it step by step, allowing for differentiable optimization of logical error rates.
Deep Semantic Graph Transformer for Multi-View 3D Human Pose Estimation
Lijun Zhang (Chongqing Institute of Green and Intelligent Technology, Chinese Academy of Sciences), Yu Shi (Chongqing Institute of Green and Intelligent Technology, Chinese Academy of Sciences)
Pose EstimationGraph Neural NetworkTransformerPoint Cloud
🎯 What it does: A multi-view 3D human pose estimation framework based on a deep semantic graph Transformer is proposed, which integrates multi-view semantic features and gradually fuses spatial-temporal information.
Deep Structural Knowledge Exploitation and Synergy for Estimating Node Importance Value on Heterogeneous Information Networks
Yankai Chen (Chinese University of Hong Kong), Irwin King (University of Copenhagen)
Graph Neural NetworkGraph
🎯 What it does: This paper proposes the SKES framework, which utilizes structural prior knowledge and self-attention mechanisms to model the distribution of node features in heterogeneous information networks (HIN), and measures the difference between node features and a random reference distribution using the 1-Wasserstein distance to infer the importance values of nodes.
Deep Unfolded Network with Intrinsic Supervision for Pan-Sharpening
Hebaixu Wang (Wuhan University), Jiayi Ma (Wuhan University)
RestorationConvolutional Neural NetworkImage
🎯 What it does: An interpretable deep unfolding network is proposed, which achieves the fusion of PAN and MS images (pansharpening) through spatial consistency and spectral projection priors.
Deep Variational Incomplete Multi-View Clustering: Exploring Shared Clustering Structures
Gehui Xu (Harbin Institute of Technology), Wei Wang (Harbin Institute of Technology)
ClassificationMixture of ExpertsAuto EncoderMultimodality
🎯 What it does: A depth incomplete multi-view clustering method without interpolation, DVIMC, is designed to jointly learn shared latent representations and achieve clustering through variational autoencoders, Product-of-Experts, and VaDE techniques.
DeepAccident: A Motion and Accident Prediction Benchmark for V2X Autonomous Driving
Tianqi Wang (University of Hong Kong), Ping Luo (University of Hong Kong)
Autonomous DrivingTransformerImagePoint CloudBenchmark
🎯 What it does: The first large-scale V2X autonomous driving accident dataset, DeepAccident, has been constructed, and an end-to-end motion and accident prediction task has been proposed along with the baseline model V2XFormer.
DeepBranchTracer: A Generally-Applicable Approach to Curvilinear Structure Reconstruction Using Multi-Feature Learning
Chao Liu (Zhejiang University), Nenggan Zheng (Zhejiang University)
Object DetectionSegmentationConvolutional Neural NetworkRecurrent Neural NetworkImage
🎯 What it does: This paper proposes DeepBranchTracer, a method for curve structure reconstruction using multi-feature learning.
DeepCalliFont: Few-Shot Chinese Calligraphy Font Synthesis by Integrating Dual-Modality Generative Models
Yitian Liu (Peking University), Zhouhui Lian (Peking University)
GenerationData SynthesisConvolutional Neural NetworkTransformerContrastive LearningImageMultimodality
🎯 What it does: This paper studies a few-shot Chinese calligraphy font synthesis method based on a dual-modal generative model called DeepCalliFont.
DeepPointMap: Advancing LiDAR SLAM with Unified Neural Descriptors
Xiaze Zhang (Fudan University), Rui Feng (Fudan University)
Autonomous DrivingTransformerContrastive LearningSimultaneous Localization and MappingPoint Cloud
🎯 What it does: DeepPointMap proposes a unified neural descriptor framework that achieves high-precision LiDAR SLAM and lightweight map construction using sparse neural features;
DeepSaDe: Learning Neural Networks That Guarantee Domain Constraint Satisfaction
Kshitij Goyal (KU Leuven), Hendrik Blockeel (KU Leuven)
OptimizationConvolutional Neural NetworkReinforcement LearningTabularFinance Related
🎯 What it does: The DeepSaDe method is proposed, which combines MaxSMT solving and gradient descent to ensure that the network satisfies given domain constraints for all inputs during the training of feedforward neural networks.
DeepSpeed Data Efficiency: Improving Deep Learning Model Quality and Training Efficiency via Efficient Data Sampling and Routing
Conglong Li (Microsoft), Yuxiong He (Microsoft)
Computational EfficiencyData-Centric LearningTransformerLarge Language ModelSupervised Fine-TuningMixture of ExpertsText
🎯 What it does: The DeepSpeed Data Efficiency framework is proposed, which enhances training data and computational efficiency through scalable curriculum learning sampling and random layer token dropping techniques, applied to the pre-training and fine-tuning of large language models and vision models.
Defeasible Normative Reasoning: A Proof-Theoretic Integration of Logical Argumentation
Ofer Arieli (Tel-Aviv Academic College), Christian Straßer (Ruhr University Bochum)
🎯 What it does: A new non-monotonic normative reasoning framework based on proof theory is proposed, which explicitly represents the acceptability and rejection of derivations in a sequenced system using annotated reasoning rules, and establishes a correspondence with formal argumentation and input/output logic.
Defying Imbalanced Forgetting in Class Incremental Learning
Shixiong Xu (Chinese Academy of Sciences), Shiming Xiang (University of Science and Technology Beijing)
ClassificationRecognitionConvolutional Neural NetworkImage
🎯 What it does: This paper proposes a new method called CLAD to address the issue of uneven forgetting among different old classes in class-incremental learning.