AAAI 2023 Papers — Page 4
AAAI Conference on Artificial Intelligence · 1578 papers
Curriculum Temperature for Knowledge Distillation
Zheng Li (Nankai University), Jian Yang (Nankai University)
ClassificationObject DetectionKnowledge DistillationImage
🎯 What it does: A learnable, curriculum learning-based temperature regulation method CTKD is proposed to dynamically adjust the temperature during the knowledge distillation process, thereby gradually increasing the learning difficulty for the student network.
Cyclically Disentangled Feature Translation for Face Anti-spoofing
Haixiao Yue (Baidu Inc.), Jingdong Wang (Baidu Inc.)
ClassificationRecognitionDomain AdaptationGenerative Adversarial NetworkImage
🎯 What it does: This paper proposes a Circular Decoupled Feature Translation Network (CDFTN) that generates pseudo-labeled samples and trains a robust classifier by exchanging domain-invariant liveness features with domain-specific content features, thereby achieving cross-scene face anti-spoofing.
DACOM: Learning Delay-Aware Communication for Multi-Agent Reinforcement Learning
Tingting Yuan (University of Gttingen), Xiaoming Fu (Beijing University of Posts and Telecommunications)
Reinforcement Learning
🎯 What it does: This paper proposes a delay-aware communication model called DACOM, aimed at enhancing cooperative performance in multi-agent reinforcement learning by learning appropriate waiting times to balance communication benefits and delay costs.
DAMix: Exploiting Deep Autoregressive Model Zoo for Improving Lossless Compression Generalization
Qishi Dong (Hong Kong Baptist University), Zhenguo Li (Huawei)
CompressionMixture of ExpertsImage
🎯 What it does: A multi-model autoregressive model library (DAMix) has been constructed, achieving efficient lossless compression of out-of-distribution (OoD) data through rapid model selection and unbiased density estimation based on von Mises-Fisher filtering during compression.
DarkFeat: Noise-Robust Feature Detector and Descriptor for Extremely Low-Light RAW Images
Yuze He (Tsinghua University), Jiangtao Wen (Research Institute of Tsinghua University in Shenzhen)
Object DetectionConvolutional Neural NetworkImage
🎯 What it does: An end-to-end DarkFeat model has been developed, capable of detecting and describing local features directly from extremely low-light RAW images.
DARL: Distance-Aware Uncertainty Estimation for Offline Reinforcement Learning
Hongchang Zhang (Tsinghua University), Xiangyang Ji (Tsinghua University)
Reinforcement LearningTabular
🎯 What it does: A distance-aware uncertainty estimation method is proposed, which is combined with an adaptive truncated distributed evaluator to form the DARL framework, aimed at enhancing the performance of offline reinforcement learning.
Darwinian Model Upgrades: Model Evolving with Selective Compatibility
Binjie Zhang (Tencent), Ying Shan (Tencent)
RecognitionRetrievalConvolutional Neural NetworkContrastive LearningImage
🎯 What it does: This paper proposes a backward-compatible model upgrade framework called Darwinian Model Upgrades (DMU), which enables continuous model upgrades for large-scale retrieval systems without the need to recalculate old features (backfill).
DASH: A Distributed and Parallelizable Algorithm for Size-Constrained Submodular Maximization
Tonmoy Dey (Florida State University), Alan Kuhnle (Texas A&M University)
OptimizationImage
🎯 What it does: This paper proposes three distributed algorithms, R-DASH, T-DASH, and G-DASH, for submodular maximization under the MapReduce framework, achieving constant factor approximation in a single round of MR with sublinear adaptive complexity.
Data Imputation with Iterative Graph Reconstruction
Jiajun Zhong (Central South University), Weiwei Ye (Central South University)
Graph Neural NetworkTabular
🎯 What it does: An iterative graph generation and reconstruction framework (IGRM) is proposed for table completion with missing data.
Data-Efficient Image Quality Assessment with Attention-Panel Decoder
Guanyi Qin (Tsinghua University), Yan Zhang (Xiamen University)
TransformerImage
🎯 What it does: A blind image quality assessment method based on a Transformer encoder-decoder (DEIQT) has been developed, which re-encodes the CLS token through the decoder and incorporates an attention panel to reduce prediction uncertainty.
DC-Former: Diverse and Compact Transformer for Person Re-identification
Wen Li (Ant Group), Wei Chu (Ant Group)
RecognitionTransformerImage
🎯 What it does: Introduce multi-class Tokens in Vision Transformer and use self-heterogeneous constraints (SDC) to make the embedding subspaces corresponding to different Tokens orthogonal, resulting in diversified and compact feature representations, thereby improving person re-identification performance.
De-biased Teacher: Rethinking IoU Matching for Semi-supervised Object Detection
Kuo Wang (Sun Yat-sen University), Fan Zhou
Object DetectionKnowledge DistillationImage
🎯 What it does: This paper proposes a semi-supervised object detection framework called De-biased Teacher, which eliminates the traditional IoU matching process and directly uses soft labels for consistency regularization to reduce training bias.
DE-net: Dynamic Text-Guided Image Editing Adversarial Networks
Ming Tao (Nanjing University of Posts and Telecommunications), Qi Tian (Huawei Inc)
Image TranslationGenerationConvolutional Neural NetworkGenerative Adversarial NetworkImageText
🎯 What it does: This paper proposes a GAN-based dynamic text-guided image editing framework called DE-Net, which can dynamically adjust editing strategies based on input text and source images to accurately perform various editing tasks such as color, texture, and content addition or deletion.
DeAR: A Deep-Learning-Based Audio Re-recording Resilient Watermarking
Chang Liu (University of Science and Technology of China), Nenghai Yu (University of Science and Technology of China)
Convolutional Neural NetworkAudio
🎯 What it does: This paper proposes a deep learning-based audio watermark embedding and extraction framework called DeAR, specifically designed for robustness against audio re-recording (AR) attacks.
Debiased Fine-Tuning for Vision-Language Models by Prompt Regularization
Beier Zhu (Nanyang Technological University), Hanwang Zhang (Nanyang Technological University)
ClassificationDomain AdaptationTransformerSupervised Fine-TuningPrompt EngineeringVision Language ModelImageMultimodality
🎯 What it does: A new fine-tuning paradigm called Prompt Regularization (ProReg) is proposed, which balances pre-trained knowledge and downstream task knowledge by using the prompt predictions of pre-trained models as regularization, achieving unbiased fine-tuning.
Decentralized Riemannian Algorithm for Nonconvex Minimax Problems
Xidong Wu (University of Pittsburgh), Heng Huang (University of Pittsburgh)
OptimizationFederated LearningImage
🎯 What it does: A decentralized Riemannian gradient descent ascent algorithm is proposed to solve distributed non-convex-strongly convex min-max problems, constrained on the Stiefel manifold.
Decentralized Stochastic Multi-Player Multi-Armed Walking Bandits
Guojun Xiong (Binghamton University), Jian Li (Binghamton University)
OptimizationReinforcement Learning from Human FeedbackTabular
🎯 What it does: A multi-player multi-armed walking bandit (MPMAB-WA) model is proposed, and a decentralized UCB algorithm MPMAB-WA-UCB is designed, which can achieve near-optimal cumulative rewards under the condition that each player can only access a dynamic local arm set.
Decision-Making Context Interaction Network for Click-Through Rate Prediction
Xiang Li (Meituan), Dong Wang (Meituan)
Recommendation SystemTabular
🎯 What it does: This paper proposes a Decision-Making Context Interaction Network (DCIN) to enhance the accuracy of Click-Through Rate (CTR) prediction.
DeCOM: Decomposed Policy for Constrained Cooperative Multi-Agent Reinforcement Learning
Zhaoxing Yang (Shanghai Jiao Tong University), Chenghu Zhou (Shanghai Jiao Tong University)
OptimizationReinforcement Learning
🎯 What it does: A constrained cooperative multi-agent reinforcement learning framework named DeCOM is proposed, aiming to maximize team average returns while satisfying team average cost constraints.
Deconstructed Generation-Based Zero-Shot Model
Dubing Chen (Nanjing University of Science and Technology), Philip H.S. Torr (University of Oxford)
GenerationData SynthesisGenerative Adversarial NetworkImage
🎯 What it does: Decomposes the generative zero-shot learning framework and proposes a simplified method based on attribute generalization and bias correction.
Decorate the Newcomers: Visual Domain Prompt for Continual Test Time Adaptation
Yulu Gan (Peking University), Lin Luo (Peking University)
Domain AdaptationPrompt EngineeringImage
🎯 What it does: Proposes to achieve continuous testing domain adaptation by adding lightweight visual domain prompts (domain-specific and domain-agnostic) on the input images while keeping the source model unchanged;
Deep Attentive Model for Knowledge Tracing
Xinping Wang (East China Normal University), Min Zhang (East China Normal University)
Recommendation SystemRecurrent Neural NetworkTabularTime SeriesSequential
🎯 What it does: A new knowledge tracing model called Deep Attentive Knowledge Tracing Network (DAKTN) is proposed, which combines deep neural networks with students' historical learning behavior sequences. It extracts sequence information using pooling layers and local attention units, and then predicts performance through embedding layers and multilayer perceptrons.
Deep Digging into the Generalization of Self-Supervised Monocular Depth Estimation
Jinwoo Bae (DGIST), Sunghoon Im (DGIST)
Depth EstimationDomain AdaptationAutonomous DrivingConvolutional Neural NetworkTransformerImage
🎯 What it does: This study investigates the generalization performance of self-supervised monocular depth estimation, systematically evaluating the performance of CNNs, Transformers, and their hybrid networks under different texture/shape biases, and proposes the MonoFormer network to enhance generalization.
Deep Equilibrium Models for Snapshot Compressive Imaging
Yaping Zhao (Westlake University), Xin Yuan (Westlake University)
RestorationCompressionRecurrent Neural NetworkVideo
🎯 What it does: For the inverse problem of video snapshot compression imaging (SCI), a reconstruction framework based on the Deep Equilibrium Model (DEQ) is proposed, which implicitly learns non-expansive operators and analytically solves fixed points, achieving infinite iterations, infinite network depth, while keeping memory consumption constant.
Deep Graph Structural Infomax
Wenting Zhao (Nanjing University of Science and Technology), Tong Zhang (Nanyang Technological University)
Representation LearningGraph Neural NetworkContrastive LearningGraph
🎯 What it does: A self-supervised graph node representation learning framework DGSI based on information bottleneck is proposed, which can simultaneously capture structural and semantic information.
Deep Latent Regularity Network for Modeling Stochastic Partial Differential Equations
Shiqi Gong (Academy of Mathematics and Systems Science, Chinese Academy of Sciences), Tie-Yan Liu (Microsoft Research AI4Science)
OptimizationComputational EfficiencyTime SeriesPhysics RelatedStochastic Differential Equation
🎯 What it does: Construct a Deep Latent Regularization Network (DLR-Net) that utilizes the regularity structure theory to learn the solution operator of stochastic partial differential equations (SPDEs).
Deep Manifold Attack on Point Clouds via Parameter Plane Stretching
Keke Tang (Guangzhou University), Wenping Wang (Texas A&M University)
Adversarial AttackAuto EncoderPoint Cloud
🎯 What it does: Designed a manifold attack based on parameter plane stretching, generating adversarial samples by perturbing the intrinsic two-dimensional manifold surface of point clouds.
Deep Parametric 3D Filters for Joint Video Denoising and Illumination Enhancement in Video Super Resolution
Xiaogang Xu (Chinese University of Hong Kong), Jiaya Jia (Chinese University of Hong Kong)
RestorationSuper ResolutionConvolutional Neural NetworkVideo
🎯 What it does: Proposed a single end-to-end network DP3DF, achieving super-resolution, denoising, and exposure enhancement for low-light noisy videos;
Deep Spiking Neural Networks with High Representation Similarity Model Visual Pathways of Macaque and Mouse
Liwei Huang (Peking University), Yonghong Tian (Peng Cheng Laboratory)
Representation LearningConvolutional Neural NetworkSpiking Neural NetworkTransformerImage
🎯 What it does: This study investigates the effectiveness of deep spiking neural networks (SNN) in modeling human and rodent visual cortices, comparing it with mainstream CNNs and ViTs.
Deep Visual Forced Alignment: Learning to Align Transcription with Talking Face Video
Minsu Kim (KAIST), Yong Man Ro (KAIST)
Anomaly DetectionTransformerVideoText
🎯 What it does: This paper studies a visual forced alignment method (DVFA) that uses only video without relying on audio to achieve temporal alignment between text and speaker video.
Deepfake Video Detection via Facial Action Dependencies Estimation
Lingfeng Tan (Beihang University), Yuanfang Guo (Beihang University)
ClassificationRecognitionGraph Neural NetworkVideo
🎯 What it does: The FADE framework is proposed, modeling deepfake video detection as a facial action unit (AU) graph classification task, significantly improving detection performance through a Multi-Dependency Graph Module (MDGM).
Defending against Backdoor Attacks in Natural Language Generation
Xiaofei Sun (Zhejiang University), Tianwei Zhang (Nanyang Technological University)
GenerationAdversarial AttackTransformerText
🎯 What it does: This paper studies backdoor attacks on natural language generation models and their defense mechanisms, focusing on machine translation and dialogue generation tasks.
Defending Backdoor Attacks on Vision Transformer via Patch Processing
Khoa D. Doan (VinUniversity), Ping Li (LinkedIn Corporation)
ClassificationTransformerImage
🎯 What it does: This study investigates the vulnerability of Vision Transformer (ViT) to backdoor attacks and proposes defense methods based on image patch processing (PatchDrop and PatchShuffle).
Defending Black-Box Skeleton-Based Human Activity Classifiers
He Wang (University of Leeds), Guodong Guo (Baidu Research)
ClassificationRecognitionAdversarial AttackVideo
🎯 What it does: A black-box defense framework BEAT for skeletal action recognition is proposed.
DeFL: Defending against Model Poisoning Attacks in Federated Learning via Critical Learning Periods Awareness
Gang Yan (Binghamton University), Jian Li (Binghamton University)
Anomaly DetectionFederated LearningImage
🎯 What it does: Defending against model poisoning attacks in federated learning by identifying key learning phases and fine-grained gradient differences.
Delving Deep into Pixel Alignment Feature for Accurate Multi-View Human Mesh Recovery
Kai Jia (Tsinghua University), Yebin Liu (Tsinghua University)
Pose EstimationConvolutional Neural NetworkTransformerMesh
🎯 What it does: This paper proposes the Pixel-aligned Feedback Fusion (PaFF) method, which combines iterative regression with pixel-aligned feedback features to achieve high-precision human mesh recovery from multiple viewpoints.
Delving into the Adversarial Robustness of Federated Learning
Jie Zhang (Zhejiang University), Chao Wu (Youtu Lab Tencent)
Federated LearningAdversarial AttackImage
🎯 What it does: This paper systematically evaluates the robustness of federated learning under adversarial attacks, finding that traditional adversarial training significantly reduces clean accuracy in non-IID scenarios. It proposes a decision boundary-based federated adversarial training (DBFAT) scheme that enhances accuracy and robustness through local reweighting and global regularization.
DeMT: Deformable Mixer Transformer for Multi-Task Learning of Dense Prediction
Yangyang Xu (Wuhan University), Lefei Zhang (Wuhan University)
SegmentationDepth EstimationTransformerImage
🎯 What it does: Proposed the DeMT model to achieve multi-task dense prediction.
Demystifying Randomly Initialized Networks for Evaluating Generative Models
Junghyuk Lee (Yonsei University), Jong-Seok Lee (Yonsei University)
GenerationData SynthesisRetrievalConvolutional Neural NetworkTransformerGenerative Adversarial NetworkImage
🎯 What it does: A thorough study was conducted on the evaluation of generative models, focusing on the impact of the feature space of randomly initialized networks on the evaluation;
DENet: Disentangled Embedding Network for Visible Watermark Removal
Ruizhou Sun (South China University of Technology), Qingyao Wu (South China University of Technology)
RestorationConvolutional Neural NetworkAuto EncoderContrastive LearningImage
🎯 What it does: A network named DENet is proposed, which separates watermarked and non-watermarked images through contrastive learning of high-order embeddings, and combines a self-attention enhancement module to achieve the removal of visible watermarks.
Denoising after Entropy-Based Debiasing a Robust Training Method for Dataset Bias with Noisy Labels
Sumyeong Ahn (Kim Jaechul Graduate School of Artificial Intelligence), Se-Young Yun (Kim Jaechul Graduate School of Artificial Intelligence)
ClassificationRecognitionData-Centric LearningImage
🎯 What it does: A training framework named DENEB is proposed to enhance the unbiased generalization performance of models in the presence of dataset bias and noisy labels.
Denoising Multi-Similarity Formulation: A Self-Paced Curriculum-Driven Approach for Robust Metric Learning
Chenkang Zhang (Nanjing University of Information Science and Technology), Bin Gu (Nanjing University of Science and Technology)
RetrievalOptimizationImage
🎯 What it does: A balanced self-paced learning metric learning algorithm based on adaptive learning (BSPML) is proposed to remove noisy samples and enhance the robustness of deep metric learning.
Denoising Pre-training for Machine Translation Quality Estimation with Curriculum Learning
Xiang Geng (Nanjing University), Jiajun Chen (Nanjing University)
TransformerSupervised Fine-TuningText
🎯 What it does: This paper proposes CLQE—a denoising pre-training framework based on curriculum learning for machine translation quality estimation (QE), which enhances model performance by gradually introducing noisy pseudo QE data.
Design Amortization for Bayesian Optimal Experimental Design
Noble Kennamer (University of California Irvine), Alexander Ihler (University of California Irvine)
OptimizationComputational EfficiencyFlow-based ModelTabular
🎯 What it does: A deep learning architecture is proposed for accelerating the estimation of expected information gain (EIG) in Bayesian optimal experimental design, achieving 'design amortization' for all possible designs.
DesNet: Decomposed Scale-Consistent Network for Unsupervised Depth Completion
Zhiqiang Yan (Nanjing University of Science and Technology), Jian Yang (Nanjing University of Science and Technology)
RestorationDepth EstimationConvolutional Neural NetworkPoint Cloud
🎯 What it does: DesNet is proposed for unsupervised depth completion, first decomposing absolute depth into relative depth prediction and global scale estimation, and then generating dense references through a global depth guidance module and propagating under dense-sparse attention.
Detecting and Grounding Important Characters in Visual Stories
Danyang Liu (University of Edinburgh), Frank Keller (University of Edinburgh)
RecognitionObject DetectionVision Language ModelContrastive LearningImageTextMultimodalityBenchmark
🎯 What it does: Introduced a character-centered VIST-Character dataset for visual story generation, and proposed two new tasks: important character detection and character localization;
Detecting Multivariate Time Series Anomalies with Zero Known Label
Qihang Zhou (NetEase Fuxi AI Lab), Wenchao Meng (Zhejiang University)
Anomaly DetectionRecurrent Neural NetworkGraph Neural NetworkFlow-based ModelTime Series
🎯 What it does: This paper proposes an unlabeled multivariate time series anomaly detection method called MTGFlow, which can automatically identify anomalies in a training set that contains anomalous samples.
Detecting Sources of Healthcare Associated Infections
Hankyu Jang (University of Iowa), Sriram V. Pemmaraju (University of Iowa)
OptimizationGraphBiomedical DataElectronic Health Records
🎯 What it does: In the hospital-acquired infection (HAI) transmission model, a new source detection method is proposed, utilizing a load sharing model to identify the source of infection.
DHGE: Dual-View Hyper-Relational Knowledge Graph Embedding for Link Prediction and Entity Typing
Haoran Luo (Beijing University of Posts and Telecommunications), Kaiyang Wan (Beijing University of Posts and Telecommunications)
Knowledge DistillationRepresentation LearningGraph Neural NetworkGraphBiomedical Data
🎯 What it does: Proposes a dual-view hyper-relational knowledge graph embedding model DHGE, addressing the issue of traditional single-view hyper-relational KG embeddings neglecting hierarchical structures;
Dialogue Rewriting via Skeleton-Guided Generation
Chunlei Xin (Chinese Information Processing Laboratory, Institute of Software, Chinese Academy of Sciences), Le Sun (Xiaomi AI Lab)
GenerationTransformerLarge Language ModelSupervised Fine-TuningTextBenchmark
🎯 What it does: This paper proposes a dialogue rewriting framework that can convert incomplete and noisy utterances in multi-turn dialogues into complete, context-independent natural language sentences.
Dialogue State Distillation Network with Inter-slot Contrastive Learning for Dialogue State Tracking
Jing Xu (Beijing Institute of Technology), Jian Xie (Baidu Inc.)
Knowledge DistillationTransformerContrastive LearningText
🎯 What it does: This paper proposes the Dialogue State Distillation Network (DSDN) and a cross-slot contrastive learning framework for dialogue state tracking in task-oriented dialogue systems.
DICNet: Deep Instance-Level Contrastive Network for Double Incomplete Multi-View Multi-Label Classification
Chengliang Liu (Harbin Institute of Technology), Yong Xu (Sun Yat-sen University)
ClassificationAuto EncoderContrastive LearningImage
🎯 What it does: This paper addresses the problem of dual incomplete multi-view multi-label classification and proposes a deep instance-level contrastive network, DICNet.
DiFA: Differentiable Feature Acquisition
Aritra Ghosh (University of Massachusetts Amherst), Andrew Lan (University of Massachusetts Amherst)
OptimizationReinforcement LearningImageTabular
🎯 What it does: This paper proposes a differentiable feature acquisition method called DiFA, which dynamically selects a small number of features for prediction tasks.
Diffeomorphic Information Neural Estimation
Bao Duong (Deakin University), Thin Nguyen (Deakin University)
Flow-based ModelTabular
🎯 What it does: This paper proposes the DINE (Diffeomorphic Information Neural Estimator) method for accurately estimating the conditional mutual information (CMI) of continuous random variables, and further applies it to conditional independence testing.
Differentiable Meta Multigraph Search with Partial Message Propagation on Heterogeneous Information Networks
Chao Li (Huazhong University of Science and Technology), Kun He (Huazhong University of Science and Technology)
ClassificationRecommendation SystemNeural Architecture SearchGraph Neural NetworkGraph
🎯 What it does: This paper proposes a differentiable search-based heterogeneous information network graph neural network architecture search method called PMMM, which automatically learns multi-graph elements (meta-multigraph) and stable partial message propagation strategies.
Differentially Private Condorcet Voting
Zhechen Li (Peking University), Hanpin Wang (Guangzhou University)
Safty and Privacy
🎯 What it does: Proposes three types of differential privacy voting rules based on the Condorcet method, analyzing the trade-off between privacy and voting axioms;
Differentially Private Fair Division
Pasin Manurangsi (Google Research), Warut Suksompong (National University of Singapore)
OptimizationSafty and Privacy
🎯 What it does: In the framework of fair allocation of indivisible resources, this paper studies how differential privacy (DP) and fairness (such as approximate envy-freeness and approximate proportionality) can be achieved simultaneously, and provides algorithms and lower bounds for two types of adjacency relations (agent-level and agent×item-level).
Differentially Private Heatmaps
Badih Ghazi (Google Research), Nachiappan Valliappan (Google Research)
Safty and PrivacyTabular
🎯 What it does: This paper designs a differential privacy heatmap generation algorithm based on Earth Mover's Distance (EMD), which can aggregate sparse distributions while ensuring user privacy.
Differentially Private Learning with Per-Sample Adaptive Clipping
Tianyu Xia (Tsinghua University), Xing Fu (Ant Group)
OptimizationSafty and PrivacyImageText
🎯 What it does: A differential privacy learning algorithm DP-PSAC based on sample-adaptive clipping is proposed to reduce gradient bias using a non-monotonic weight function.
Differentially Private Nonlinear Causal Discovery from Numerical Data
Hao Zhang (Fudan University), Shuigeng Zhou (Fudan University)
OptimizationSafty and PrivacyTabular
🎯 What it does: A differential privacy nonlinear numerical causal discovery method based on regression-based conditional independence testing (RCIT) is proposed, which can handle nonlinear relationships while maintaining privacy.
DiffMD: A Geometric Diffusion Model for Molecular Dynamics Simulations
Fang Wu (Tsinghua University), Stan Z. Li (Westlake University)
GenerationComputational EfficiencyDrug DiscoveryTransformerDiffusion modelScore-based ModelTime SeriesStochastic Differential Equation
🎯 What it does: The DIFFMD model is proposed, which utilizes score-based denoising diffusion to generate molecular dynamics trajectories, directly predicting the next frame coordinates, eliminating the need for energy/force field intermediate variables, thus improving inference efficiency.
Diffuser: Efficient Transformers with Multi-Hop Attention Diffusion for Long Sequences
Aosong Feng (Yale University), Rex Ying (Yale University)
ClassificationGenerationComputational EfficiencyTransformerImageText
🎯 What it does: This paper presents Diffuser, an efficient Transformer that combines sparse attention with multi-hop attention diffusion, enabling global information propagation within a single layer and addressing the trade-off between memory and expressiveness in traditional sparse Transformers.
Diffusing Gaussian Mixtures for Generating Categorical Data
Florence Regol (McGill University), Mark Coates (McGill University)
GenerationData SynthesisProtein Structure PredictionTransformerDiffusion modelTabularSequentialBiomedical Data
🎯 What it does: A nominal classification data generation method based on a diffusion probability model is proposed—GMCD, which utilizes spherical packing encoding to map categories to continuous space and combines a Gaussian mixture structure during the denoising phase;
Diffusion Models Beat GANs on Topology Optimization
François Mazé (Massachusetts Institute of Technology), Faez Ahmed (Massachusetts Institute of Technology)
GenerationOptimizationDiffusion modelImage
🎯 What it does: The paper proposes a topology optimization framework called TopoDiff based on a conditional diffusion model, which is used to generate structures that meet volume, load, and boundary conditions while satisfying mechanical performance and manufacturability.
DINet: Deformation Inpainting Network for Realistic Face Visually Dubbing on High Resolution Video
Zhimeng Zhang (Netease Fuxi AI Lab), Yu Ding (Netease Fuxi AI Lab)
Image TranslationRestorationGenerationConvolutional Neural NetworkGenerative Adversarial NetworkVideo
🎯 What it does: This paper proposes the Deformation Inpainting Network (DINet), which achieves high-resolution facial audio-visual dubbing by spatially deforming and filling in the features of reference images.
Direct Heterogeneous Causal Learning for Resource Allocation Problems in Marketing
Hao Zhou (Meituan), Dong Wang (Meituan)
Recommendation SystemOptimizationTabularFinance Related
🎯 What it does: A direct causal learning framework is proposed, utilizing decision factors as learning objectives to directly predict decision outcomes in resource allocation problems, avoiding the error accumulation of traditional two-stage methods.
Directed Acyclic Graph Structure Learning from Dynamic Graphs
Shaohua Fan (Beijing University of Posts and Telecommunications), Chuan Shi (Peng Cheng Laboratory)
Graph Neural NetworkGraphTime Series
🎯 What it does: This paper proposes a directed acyclic graph (DAG) structure learning method for dynamic graphs called GraphNOTEARS, which aims to estimate the feature generation mechanisms both within the same time slice (intra-slice) and across time slices (inter-slice);
Discriminability and Transferability Estimation: A Bayesian Source Importance Estimation Approach for Multi-Source-Free Domain Adaptation
Zhongyi Han (Shandong University), Yilong Yin (Shandong University)
Domain AdaptationImage
🎯 What it does: This paper proposes a source model importance estimation method based on Bayesian inference, called DATE, which can assign weights for multi-source free domain adaptation and improve performance without accessing the source data.
Disentangle and Remerge: Interventional Knowledge Distillation for Few-Shot Object Detection from a Conditional Causal Perspective
Jiangmeng Li (University of Chinese Academy of Sciences), Fuchun Sun (Tsinghua University)
Object DetectionKnowledge DistillationImage
🎯 What it does: This paper introduces knowledge distillation in few-shot object detection and identifies and eliminates the negative impact of teacher errors on the student model through a structural causal model.
Disentangled CVAEs with Contrastive Learning for Explainable Recommendation
Linlin Wang (East China Normal University), Liang He (East China Normal University)
Recommendation SystemExplainability and InterpretabilityTransformerAuto EncoderContrastive LearningText
🎯 What it does: This paper proposes an explanation generation framework that utilizes Disentangled Conditional Variational Autoencoders (CVAE) combined with self-supervised contrastive learning, capable of learning interpretable latent preferences from user, item ID, and rating information, and generating high-quality, personalized natural language explanations.
Disentangled Representation for Causal Mediation Analysis
Ziqi Xu (University of South Australia), Ke Wang (Simon Fraser University)
Representation LearningAuto EncoderTabular
🎯 What it does: A new causal mediation analysis method called DMAVAE is proposed, aimed at accurately estimating natural direct effects, natural indirect effects, and total effects.
Disentangling Reafferent Effects by Doing Nothing
Benedict Wilkins (Royal Holloway University of London), Kostas Stathis (Royal Holloway University of London)
Reinforcement LearningImageSequential
🎯 What it does: A framework based on causal inference is proposed, which separates the spontaneous effects (reafference) and external effects (exafference) in agent perception by comparing 'doing nothing' (no action) and provides corresponding learning algorithms.
DisGUIDE: Disagreement-Guided Data-Free Model Extraction
Jonathan Rosenthal (Purdue University), Lin Tan (Purdue University)
ClassificationKnowledge DistillationData-Centric LearningConvolutional Neural NetworkGenerative Adversarial NetworkImage
🎯 What it does: DisGUIDE proposes a data-independent model extraction framework that simultaneously trains two clone models and uses disagreement loss to generate query samples that better force the clone models to diverge, thereby improving extraction efficiency.
Dish-TS: A General Paradigm for Alleviating Distribution Shift in Time Series Forecasting
Wei Fan (University of Central Florida), Yanjie Fu (University of Central Florida)
Recurrent Neural NetworkTransformerTime Series
🎯 What it does: This paper proposes a general neural network paradigm called Dish-TS, which uses dual coefficient networks (BACKCONET and HORICONET) to normalize and denormalize distribution shifts in time series forecasting, thereby enhancing the model's generalization performance.
Distantly-Supervised Named Entity Recognition with Adaptive Teacher Learning and Fine-Grained Student Ensemble
Xiaoye Qu (Huawei Cloud), Pan Zhou (Hubei Key Laboratory of Distributed System Security Hubei Engineering Research Center on Big Data Security)
RecognitionKnowledge DistillationTransformerSupervised Fine-TuningText
🎯 What it does: A self-supervised teacher-student framework ATSEN is proposed for denoising remote supervised named entity recognition.
Distributed Projection-Free Online Learning for Smooth and Convex Losses
Yibo Wang (Nanjing University), Lijun Zhang (Nanjing University)
OptimizationTabular
🎯 What it does: This study investigates the computational bottleneck of projection operations in distributed online convex optimization and proposes a projection-free distributed algorithm D-OSPA, which achieves a better regret upper bound by utilizing the smoothness of the loss function.
Distributed Spectrum-Based Fault Localization
Avraham Natan (Ben Gurion University of the Negev), Meir Kalech (Ben Gurion University of the Negev)
Safty and PrivacyTabular
🎯 What it does: A Distributed Spectrum Fault Localization (DSFL) framework is proposed, and two distributed SFL algorithms (single fault DSFLA-SINGLE and multiple fault DSFLA-MULTI) are designed to achieve the same diagnostic results as centralized SFL while ensuring privacy protection.
Distributionally Robust Optimization with Probabilistic Group
Soumya Suvra Ghosal (University of Wisconsin), Yixuan Li (University of Wisconsin)
OptimizationConvolutional Neural NetworkImageText
🎯 What it does: This paper proposes a distributionally robust optimization framework based on probabilistic groups (PG-DRO) aimed at addressing the robustness issues of machine learning models against spurious correlations.
Diversified and Realistic 3D Augmentation via Iterative Construction, Random Placement, and HPR Occlusion
Jungwook Shin (Seoul National University), Wonjong Rhee (Seoul National University)
Object DetectionAutonomous DrivingPoint Cloud
🎯 What it does: In 3D point cloud detection, a data augmentation method called DR.CPO is proposed, which achieves self-occlusion and external occlusion through iterative construction of complete targets, random placement and rotation, and HPR (Hidden Point Removal).
Diversity Maximization in the Presence of Outliers
Daichi Amagata (Osaka University)
Anomaly DetectionOptimizationTabular
🎯 What it does: This paper studies the solution methods for the maximum-minimum diversity problem in the presence of outliers and proposes two approximation algorithms.
DM²: Decentralized Multi-Agent Reinforcement Learning via Distribution Matching
Caroline Wang (University of Texas at Austin), Peter Stone (University of Texas at Austin)
Reinforcement LearningGenerative Adversarial Network
🎯 What it does: A completely decentralized multi-agent reinforcement learning framework (DM2) is proposed, achieving cooperation through independent target distribution matching by each agent.
DMIS: Dynamic Mesh-Based Importance Sampling for Training Physics-Informed Neural Networks
Zijiang Yang (University of Science and Technology Beijing), Dongmei Fu (University of Science and Technology Beijing)
OptimizationComputational EfficiencyMeshBenchmarkPhysics Related
🎯 What it does: A dynamic mesh importance sampling (DMIS) scheme is proposed to accelerate the training of physics-informed neural networks (PINNs) and improve the accuracy of solutions.
DNG: Taxonomy Expansion by Exploring the Intrinsic Directed Structure on Non-gaussian Space
Songlin Zhai (Southeast University), Yuan Meng (Southeast University)
Graph
🎯 What it does: This paper proposes a novel semantic tree structure expansion method called DNG, which constructs representations for each node by explicitly distinguishing between inherited features and supplementary features, and utilizes non-Gaussian constraints to achieve the directionality of is-a relationships.
Do Invariances in Deep Neural Networks Align with Human Perception?
Vedant Nanda (University of Maryland), Adrian Weller (University of Cambridge)
ClassificationRecognitionRepresentation LearningConvolutional Neural NetworkContrastive LearningImage
🎯 What it does: Using the gradient inversion method to generate equivalent representations of different models' inputs (IRI), and measuring their consistency with human perception through human evaluation and LPIPS perceptual distance.
DocEdit: Language-Guided Document Editing
Puneet Mathur (University of Maryland), Vlad I. Morariu (Adobe Research)
TransformerVision Language ModelImageTextMultimodality
🎯 What it does: This paper proposes a language-guided document editing system that can automatically generate executable instructions in real document editing software based on users' natural language editing requests, and locate local areas in document images.
Does It Pay to Optimize AUC?
Baojian Zhou (Fudan University), Steven Skiena (Stony Brook University)
OptimizationTabular
🎯 What it does: This paper proposes an efficient algorithm AUC-opt for precisely optimizing AUC linear classifiers in two-dimensional space, and provides a recursively extendable implementation to higher dimensions; it also proves that linear AUC optimization is NP-complete when the dimension is not fixed.
Domain Adaptation with Adversarial Training on Penultimate Activations
Tao Sun (Stony Brook University), Haibin Ling (Stony Brook University)
Domain AdaptationGenerative Adversarial NetworkImage
🎯 What it does: A framework for adversarial training of the penultimate activation in unsupervised domain adaptation (APA) is proposed, along with two normalized variants derived from it.
Domain Decorrelation with Potential Energy Ranking
Sen Pei (Institute of Automation, Chinese Academy of Sciences), Gaofeng Meng (Institute of Automation, Chinese Academy of Sciences)
Domain AdaptationConvolutional Neural NetworkTransformerImage
🎯 What it does: A domain decorrelation method based on Potential Energy Ranking (PoER) is proposed, which explicitly separates label and domain information using ranking and clustering losses at both shallow and deep levels.
Domain Generalised Faster R-CNN
Karthik Seemakurthy (University of Lincoln), Petra Bosilj (University of Lincoln)
Object DetectionDomain AdaptationConvolutional Neural NetworkImage
🎯 What it does: This paper proposes Domain Generalised Faster R-CNN, which addresses the domain transfer problem in object detection by introducing a new regularization term into Faster R-CNN, particularly enhancing the generalization ability to unknown domains without assuming covariate shift.
Domain-Adapted Dependency Parsing for Cross-Domain Named Entity Recognition
Chenxiao Dou (Nanhu Academy of Electronics and Information Technology), Xiangang Li (Beike)
RecognitionDomain AdaptationRecurrent Neural NetworkLarge Language ModelText
🎯 What it does: This paper proposes a method that utilizes cross-domain dependency parsing (DP) as an auxiliary task to enhance named entity recognition (NER) in low-resource domains.
Domain-General Crowd Counting in Unseen Scenarios
Zhipeng Du (King's College London), Miaojing Shi (Tongji University)
Domain AdaptationMeta LearningConvolutional Neural NetworkImage
🎯 What it does: This paper studies the problem of domain generalization for crowd counting in unseen scenarios, proposing a meta-learning-based domain general model that implements dynamic subdomain partitioning of source domain data, re-encoding of domain-invariant and domain-specific memory modules, and corresponding loss constraints.
Don’t Be So Sure! Boosting ASR Decoding via Confidence Relaxation
Tomer Wullach (OriginAI), Shlomo E. Chazan (OriginAI)
RecognitionTransformerSupervised Fine-TuningAudio
🎯 What it does: To address the issue of overconfidence in self-supervised learning (SSL) pre-trained speech recognition models, a hierarchical aggregation method is proposed to relax confidence and apply it to beam search decoding, thereby improving recognition accuracy.
Don’t Predict Counterfactual Values, Predict Expected Values Instead
Jeremiasz Wołosiuk (Deepsolver), Jacek Mańdziuk (Warsaw University of Technology)
Reinforcement LearningTabular
🎯 What it does: This paper studies an improved method for estimating counterfactual values (CFV) in poker games, proposing to obtain CFV by predicting expected value (EV) and then multiplying it by matching probability, instead of the traditional direct prediction of CFV.
Doodle to Object: Practical Zero-Shot Sketch-Based 3D Shape Retrieval
Bingrui Wang (Tianjin University), Yuan Zhou (Tianjin University)
RetrievalConvolutional Neural NetworkRecurrent Neural NetworkContrastive LearningImagePoint CloudMesh
🎯 What it does: A zero-shot hand-drawn sketch to 3D model retrieval method is proposed, and a large-scale Doodle2Object (D2O) dataset is constructed.
Double Doubly Robust Thompson Sampling for Generalized Linear Contextual Bandits
Wonyoung Kim (Columbia University), Myunghee Cho Paik (Seoul National University)
Recommendation SystemOptimizationReinforcement LearningTabular
🎯 What it does: A dual double robust Thompson sampling algorithm (DDRTS-GLM) is proposed for the multi-armed bandit problem in the context of generalized linear models, providing a theoretically optimal upper bound on regret.
DPText-DETR: Towards Better Scene Text Detection with Dynamic Points in Transformer
Maoyuan Ye (Wuhan University), Dacheng Tao (The University of Sydney)
Object DetectionTransformerImage
🎯 What it does: Proposes the DPText-DETR model, which utilizes dynamic points to achieve more efficient and accurate scene text detection in the Transformer framework;
DQ-DETR: Dual Query Detection Transformer for Phrase Extraction and Grounding
Shilong Liu (Tsinghua University), Lei Zhang (International Digital Economy Academy)
Object DetectionSegmentationTransformerContrastive LearningImageText
🎯 What it does: This paper proposes an end-to-end visual grounding method that jointly extracts phrases from sentences and locates the corresponding image regions, addressing the shortcomings of traditional methods when phrases are unknown.
Dream to Generalize: Zero-Shot Model-Based Reinforcement Learning for Unseen Visual Distractions
Jeongsoo Ha (Samsung Electronics), Yusung Kim (Sungkyunkwan University)
Robotic IntelligenceReinforcement LearningContrastive LearningWorld ModelVideo
🎯 What it does: A foundational reinforcement learning framework for Dr. G zero-shot modeling is proposed, utilizing self-supervised dual contrastive learning and recursive state inverse dynamics to train encoders and world models, achieving strong generalization in visual control tasks under unseen visual disturbances.
DRGCN: Dynamic Evolving Initial Residual for Deep Graph Convolutional Networks
Lei Zhang (Ant Group), Wei Chu (Ant Group)
Knowledge DistillationGraph Neural NetworkGraph
🎯 What it does: A deep graph convolutional network DRGCN is proposed to alleviate the over-smoothing problem through a dynamically evolving initial residual mechanism.
Drop Clause: Enhancing Performance, Robustness and Pattern Recognition Capabilities of the Tsetlin Machine
Jivitesh Sharma (University of Agder), Lei Jiao (University of Agder)
ClassificationRecognitionImageText
🎯 What it does: The Drop Clause mechanism is introduced in the Tsetlin Machine (TM), randomly discarding some clauses during each training cycle to enhance learning diversity and generalization performance.