IJCAI 2023 Papers — Page 2
International Joint Conference on Artificial Intelligence · 639 papers
CiT-Net: Convolutional Neural Networks Hand in Hand with Vision Transformers for Medical Image Segmentation
Tao Lei (Shaanxi University of Science and Technology), Asoke Nandi (Brunel University London)
SegmentationConvolutional Neural NetworkTransformerImageBiomedical Data
🎯 What it does: Proposed a dual-branch parallel CiT-Net architecture, combining dynamic deformable convolution with deformable window adaptive complementary attention mechanism for medical image segmentation.
CLE-ViT: Contrastive Learning Encoded Transformer for Ultra-Fine-Grained Visual Categorization
Xiaohan Yu (Griffith University), Yongsheng Gao (Griffith University)
ClassificationTransformerContrastive LearningImageAgriculture Related
🎯 What it does: Propose the CLE-ViT model, combining self-supervised instance-level contrastive learning with Vision Transformer for ultra-fine-grained visual classification.
Clustered-patch Element Connection for Few-shot Learning
Jinxiang Lai (Tencent Youtu Lab), Chengjie Wang (Tencent Youtu Lab)
ClassificationObject DetectionSegmentationMeta LearningGraph Neural NetworkTransformerContrastive LearningImage
🎯 What it does: Propose the Clustered-patch Element Connection (CEC) layer and CECNet, which enhance feature representation in few-shot classification through global-to-local connections, and extend this technique to few-shot semantic segmentation and object detection.
Co-Certificate Learning with SAT Modulo Symmetries
Markus Kirchweger (TU Wien), Stefan Szeider (TU Wien)
OptimizationComputational EfficiencyGraphPhysics Related
🎯 What it does: Proposes a collaborative co-certificate learning method based on SAT modular symmetry (SMS) to search for graphs satisfying co-NP properties in graph space, thereby improving the lower bound of Kochen-Specker vector system sizes.
Co-training with High-Confidence Pseudo Labels for Semi-supervised Medical Image Segmentation
Zhiqiang Shen (Northeastern University), Osmar R. Zaiane (University of Alberta)
SegmentationImageBiomedical DataMagnetic Resonance Imaging
🎯 What it does: Propose a new semi-supervised medical image segmentation framework UCMT, combining collaborative average teacher and uncertainty-based regional mixing to enhance pseudo-label quality while maintaining model differences.
Cognitively Inspired Learning of Incremental Drifting Concepts
Mohammad Rostami (University of Southern California), Aram Galstyan (University of Southern California)
Representation LearningAuto EncoderImage
🎯 What it does: Propose a generative continuous incremental learning framework based on internal distribution GMM, achieving incremental learning for concept drift and new concepts, while mitigating catastrophic forgetting through pseudo replay.
Commonsense Knowledge Enhanced Sentiment Dependency Graph for Sarcasm Detection
Zhe Yu (Tianjin University), Jianwu Dang
ClassificationGraph Neural NetworkTransformerLarge Language ModelTextRetrieval-Augmented Generation
🎯 What it does: This paper proposes a Common Sense and Syntax-aware Dependency Graph Convolutional Network (CSDGCN) for detecting sarcastic text.
Communication-Efficient Stochastic Gradient Descent Ascent with Momentum Algorithms
Yihan Zhang (Temple University), Hongchang Gao (Temple University)
ClassificationOptimizationFederated LearningImage
🎯 What it does: Develop two communication-efficient momentum-augmented SGDA algorithms for distributed min-max optimization applied to AUC maximization in imbalanced classification tasks
Competitive-Cooperative Multi-Agent Reinforcement Learning for Auction-based Federated Learning
Xiaoli Tang (Nanyang Technological University), Han Yu (Nanyang Technological University)
Federated LearningReinforcement LearningImage
🎯 What it does: Proposed a cooperative-competitive auction-based federated learning framework called MARL-AFL based on multi-agent reinforcement learning, used for collaborative scheduling of data consumers' bidding and achieving data resource allocation and multi-task participation under multi-consumer competition.
Complete Instances Mining for Weakly Supervised Instance Segmentation
Zecheng Li (South China University Of Technology), Jin-Gang Yu (South China University Of Technology)
SegmentationImage
🎯 What it does: Weakly supervised instance segmentation using image-level labels, proposing an online refinement proposal-based framework.
Complex Contagion Influence Maximization: A Reinforcement Learning Approach
Haipeng Chen (William & Mary), Milind Tambe (Harvard University)
OptimizationGraph Neural NetworkReinforcement LearningGraph
🎯 What it does: Propose RL4CCIM, a reinforcement learning-based method for complex contagion influence maximization (CCIM), addressing the challenge of sparse rewards caused by non-submodularity.
Complexity of Efficient Outcomes in Binary-Action Polymatrix Games and Implications for Coordination Problems
Argyrios Deligkas (Royal Holloway University Of London), Anders Yeo (University Of Southern Denmark)
OptimizationComputational EfficiencyGraph
🎯 What it does: This paper investigates the computational complexity of finding economically efficient outcomes in binary-action multi-player multi-matrix games. It proposes and proves a new complexity dichotomy for the MWDP graph partitioning problem, and utilizes this result to provide a complete complexity classification for three objectives—social welfare maximization, potential maximization, and welfare-optimal Nash equilibrium—in pure coordination and anti-coordination games.
Compositional Zero-Shot Artistic Font Synthesis
Xiang Li (Shandong University), Xiangxu Meng (Shandong University)
GenerationData SynthesisConvolutional Neural NetworkGenerative Adversarial NetworkContrastive LearningImage
🎯 What it does: Proposed a framework named CAFS-GAN for generating artistic fonts under zero-shot conditions, which can synthesize unseen glyph and effect combinations through modules such as a contrastive style encoder, style similarity attention, and hierarchical dual style AdaIN to achieve decoupling and fusion of glyphs and effects.
Computing (1+epsilon)-Approximate Degeneracy in Sublinear Time
Valerie King (University of Victoria), Quinton Yong (University of Victoria)
Computational EfficiencyGraph
🎯 What it does: Propose a (1+ϵ)-approximate degeneracy algorithm (NESD) using neighbor sampling, achieving sublinear time complexity of O(n log n/ϵ³) under the adjacency list model, and scalable to k-core decomposition.
Computing Abductive Explanations for Boosted Regression Trees
Gilles Audemard (University of Artois), Pierre Marquis (University of Artois)
OptimizationExplainability and InterpretabilityTabular
🎯 What it does: The study proposes two arbitrary time algorithms G (Generation) and E (Evaluation) for generating and evaluating subset minimal abductive explanations on gradient boosting regression trees, where the explanations are represented as Boolean conditions and satisfy a specified prediction interval.
Computing Twin-width with SAT and Branch & Bound
André Schidler, Stefan Szeider (TU Wien)
OptimizationComputational EfficiencyGraphBenchmark
🎯 What it does: This paper proposes two new algorithms for computing the twin-width of a graph: a more compact SAT encoding (SAT-CPT), and another algorithm based on branch-and-bound (BB-CCH).
CONGREGATE: Contrastive Graph Clustering in Curvature Spaces
Li Sun (North China Electric Power University), Philip S. Yu (University of Illinois at Chicago)
Representation LearningGraph Neural NetworkContrastive LearningGraph
🎯 What it does: This paper proposes a heterogeneous curvature space model based on Ricci curvature to accomplish end-to-end unsupervised graph clustering tasks.
Constraints First: A New MDD-based Model to Generate Sentences Under Constraints
Alexandre Bonlarron (Université Côte d'Azur, Inria), Jean-Charles Régin (Université Côte d'Azur, CNRS, I3S)
GenerationTransformerLarge Language ModelText
🎯 What it does: For the standardized sentence generation task in the MNREAD vision screening test, a discrete combinatorial optimization method based on multi-valued decision diagrams (MDD) is proposed. The method first enumerates sentences satisfying all hard constraints within the MDD, then filters and ranks the sentences using the GPT-2 language model based on perplexity (PPL).
Contact2Grasp: 3D Grasp Synthesis via Hand-Object Contact Constraint
Haoming Li (Zhejiang University), Qi Ye (Zhejiang University)
GenerationData SynthesisPose EstimationOptimizationRobotic IntelligenceAuto EncoderPoint Cloud
🎯 What it does: Introducing hand-object contact maps as intermediate variables, dividing grasp generation into two stages: contact map generation and pose mapping, and using bias-aware local optimization to enhance grasp quality.
Context-Aware Feature Selection and Classification
Juanyan Wang (Illinois Institute of Technology), Mustafa Bilgic (Illinois Institute of Technology)
ClassificationTabularFinance Related
🎯 What it does: Propose a joint model that first performs context-aware feature selection on each sample and then uses the selected features for classification.
Continuous-Time Graph Learning for Cascade Popularity Prediction
Xiaodong Lu (Beihang University), Tongyu Zhu (Beihang University)
Representation LearningRecurrent Neural NetworkGraph Neural NetworkGraph
🎯 What it does: This paper proposes a cascade popularity prediction framework named CTCP based on continuous-time graph learning, which learns dynamic representations of users and cascades in a temporal order through a global diffusion graph, and generates cascade embeddings by fusing temporal and structural features to predict the incremental popularity of future cascades.
Contour-based Interactive Segmentation
Polina Popenova (Samsung Research), Anton Konushin (Samsung Research)
SegmentationConvolutional Neural NetworkSupervised Fine-TuningImage
🎯 What it does: Proposes an interactive image segmentation method based on a single contour, enabling rapid object segmentation by drawing loose contours.
Contrastive Label Enhancement
Yifei Wang (Xi'an Jiaotong University), Zhiqiang Tian (Xi'an Jiaotong University)
Representation LearningContrastive LearningImageTextBiomedical Data
🎯 What it does: Proposed a new label enhancement method called ConLE, which utilizes contrastive learning to fuse features and logical labels into the same projection space, generating high-order features and then mapping them to a label distribution, thereby achieving the transformation from logical labels to label distributions.
Contrastive Learning and Reward Smoothing for Deep Portfolio Management
Yun-Hsuan Lien (National Yang Ming Chiao Tung University), Yu-Shuen Wang (National Yang Ming Chiao Tung University)
Graph Neural NetworkReinforcement LearningContrastive LearningTime SeriesFinance Related
🎯 What it does: Utilizing deep reinforcement learning (DRL) combined with contrastive learning and reward smoothing to train agents for asset allocation and trading in financial markets.
Contrastive Learning for Sign Language Recognition and Translation
Shiwei Gan (Nanjing University), Sanglu Lu (Nanjing University)
RecognitionImage TranslationConvolutional Neural NetworkTransformerContrastive LearningVideo
🎯 What it does: This paper addresses the CTC peak phenomenon in Continuous Sign Language Recognition (CSLR) and the exposure bias problem in Sign Language Translation (SLT) by introducing visual contrastive learning and semantic contrastive learning, and proposes two new evaluation metrics, Blank Rate (BR) and Consecutive Wrong Word Rate (CWWR).
ContrastMotion: Self-supervised Scene Motion Learning for Large-Scale LiDAR Point Clouds
Xiangze Jia (Nanjing University of Aeronautics and Astronautics), Yuexin Ma (ShanghaiTech University)
Autonomous DrivingContrastive LearningPoint Cloud
🎯 What it does: Proposes ContrastMotion, a self-supervised LiDAR scene motion estimator based on pillar representations, which predicts pillar correspondences through contrastive learning of pillar features to obtain scene motion.
Controlling Neural Style Transfer with Deep Reinforcement Learning
Chengming Feng (Chengdu University of Information Technology), Siwei Lyu (University at Buffalo State University of New York)
Image TranslationConvolutional Neural NetworkRecurrent Neural NetworkReinforcement LearningImageVideo
🎯 What it does: Proposed the RL-NST framework, achieving progressive neural style transfer through reinforcement learning;
Convergence in Multi-Issue Iterative Voting under Uncertainty
Joshua Kavner (Rensselaer Polytechnic Institute), Lirong Xia (Rensselaer Polytechnic Institute)
🎯 What it does: Studied the convergence and cyclicity of iterative voting (IV) in multi-issue uncertain environments, proposed a local dominance improvement (LDI) dynamic model, and analyzed its convergence conditions;
COOL, a Context Outlooker, and Its Application to Question Answering and Other Natural Language Processing Tasks
Fangyi Zhu (National University of Singapore), Stéphane Bressan (National University of Singapore)
TransformerText
🎯 What it does: Proposes the Context Outlooker (COOL) local attention mechanism, which can be inserted into any Transformer model, utilizing multi-window local attention and convolutional blocks to enhance local context information, thereby improving performance on NLP tasks.
CostFormer:Cost Transformer for Cost Aggregation in Multi-view Stereo
Weitao Chen (Alibaba Group), Xuansong Xie (Alibaba Group)
Depth EstimationConvolutional Neural NetworkTransformerImage
🎯 What it does: Proposed CostFormer, a Transformer-based cost aggregation network, to improve the cost volume aggregation process in multi-view stereo (MVS);
CROP: Towards Distributional-Shift Robust Reinforcement Learning Using Compact Reshaped Observation Processing
Philipp Altmann (LMU Munich), Thomy Phan (LMU Munich)
Domain AdaptationReinforcement Learning
🎯 What it does: Design and evaluate Compact Reshaped Observation Processing (CROP), which reduces observational information in reinforcement learning through three manual compression methods (radius, action, object) to improve training speed and robustness to distribution drift.
Cross-community Adapter Learning (CAL) to Understand the Evolving Meanings of Norm Violation
Thiago Freitas dos Santos (Artificial Intelligence Research Institute, CSIC), Marco Schorlemmer (Artificial Intelligence Research Institute, CSIC)
Anomaly DetectionExplainability and InterpretabilityTransformerLarge Language ModelSupervised Fine-TuningText
🎯 What it does: This paper proposes a Cross-Community Adapter Learning framework (CAL), which inserts small adapters into pre-trained language models, and continuously learns and interprets the evolution of meanings related to norm violations (especially hate speech) across different online communities through incremental fine-tuning and Integrated Gradients explanation;
Cross-Domain Facial Expression Recognition via Disentangling Identity Representation
Tong Liu (Wuhan University), Jun Wan (Zhongnan University of Economics and Law)
RecognitionDomain AdaptationImage
🎯 What it does: Propose a domain-generalized facial expression recognition framework that decouples domain, identity, and expression information through feature separation, further enhancing cross-domain robustness by preserving high-level semantics via Fourier phase reconstruction.
Cross-Modal Global Interaction and Local Alignment for Audio-Visual Speech Recognition
Yuchen Hu (Nanyang Technological University), Eng Siong Chng (Nanyang Technological University)
RecognitionTransformerContrastive LearningVideoMultimodalityAudio
🎯 What it does: Proposed a cross-modal global interaction and local alignment (GILA) framework for audio-visual speech recognition (AVSR), achieving deep complementarity and temporal consistency between audio and visual features.
CSGCL: Community-Strength-Enhanced Graph Contrastive Learning
Han Chen (Huazhong University of Science and Technology), Rui Zhang (Tsinghua University)
Representation LearningGraph Neural NetworkContrastive LearningGraph
🎯 What it does: Proposes a graph contrastive learning framework CSGCL that leverages community strength information, incorporating community strength-based attribute voting and edge deletion enhancement, as well as dynamic team collaboration contrastive learning.
CTW: Confident Time-Warping for Time-Series Label-Noise Learning
Peitian Ma (South China University of Technology), Qianli Ma (South China University of Technology)
ClassificationConvolutional Neural NetworkAuto EncoderTime Series
🎯 What it does: Proposes a CTW method for learning with time series label noise, which expands the clean sample distribution by applying time warping on confident samples and enhances model robustness by eliminating class bias through category-normalized loss.
Curriculum Multi-Level Learning for Imbalanced Live-Stream Recommendation
Shuodian Yu (Shanghai Jiao Tong University), Jian Xu (Alibaba Group)
Recommendation SystemOptimizationMeta LearningMixture of ExpertsVideo
🎯 What it does: To address the host hierarchy imbalance problem in e-commerce live streaming recommendation scenarios, we propose a curriculum learning-based multi-level learning framework called CMLIR, which separates and collaboratively trains shared parameters and level-specific parameters within a unified model.
CVTP3D: Cross-view Trajectory Prediction Using Shared 3D Queries for Autonomous Driving
Zijian Song (Chinese Academy of Sciences), Zhaoqi Wang (Chinese Academy of Sciences)
Autonomous DrivingTransformerImageMultimodalityPoint Cloud
🎯 What it does: Propose a cross-perspective trajectory prediction model (XVTP3D) based on shared 3D queries, which can generate multi-modal predictions while maintaining 3D perspective consistency.
DAMO-StreamNet: Optimizing Streaming Perception in Autonomous Driving
Jun-Yan He (DAMO Academy, Alibaba Group), Xuansong Xie (DAMO Academy, Alibaba Group)
Autonomous DrivingKnowledge DistillationConvolutional Neural NetworkVideo
🎯 What it does: Developed a real-time streaming perception framework called DAMO-StreamNet for autonomous driving, capable of efficiently detecting and predicting objects in video streams;
Data Level Lottery Ticket Hypothesis for Vision Transformers
Xuan Shen (Northeastern University), Yanzhi Wang (Northeastern University)
ClassificationTransformerImage
🎯 What it does: This paper investigates the lottery hypothesis of visual Transformers (ViT) at the data level, demonstrating that models comparable to those trained with the full dataset can be trained from scratch by selecting a subset of input image patches (winning tickets);
Decentralized Anomaly Detection in Cooperative Multi-Agent Reinforcement Learning
Kiarash Kazari (KTH Royal Institute of Technology), Gyorgy Dan (KTH Royal Institute of Technology)
Anomaly DetectionRecurrent Neural NetworkReinforcement LearningBenchmark
🎯 What it does: Propose a distributed anomaly detection framework that uses RNN to predict the action distribution of peer agents and calculates normality scores to identify attacked agents;
Decoupling with Entropy-based Equalization for Semi-Supervised Semantic Segmentation
Chuanghao Ding (Jilin University), Runbo Hu (Didi Chuxing)
SegmentationKnowledge DistillationData-Centric LearningConvolutional Neural NetworkImage
🎯 What it does: Propose a semi-supervised semantic segmentation method called DeS4 based on a teacher-student framework, primarily addressing the class imbalance problem through decoupling encoder and decoder training, sharing a non-learnable prototype classification head, and employing a multi-entropy sampling (MES) strategy.
Deep Hashing-based Dynamic Stock Correlation Estimation via Normalizing Flow
Xiaolin Zheng (Zhejiang University), Mengying Zhu (Zhejiang University)
OptimizationRecurrent Neural NetworkFlow-based ModelTime SeriesFinance Related
🎯 What it does: Proposed the HDCF model, which dynamically estimates the stock correlation matrix using deep hashing and regularized flow, and constructs a risk-averse portfolio based on this estimation.
Deep Hierarchical Communication Graph in Multi-Agent Reinforcement Learning
Zeyang Liu (Xi'an Jiaotong University), Xuguang Lan (Xi'an Jiaotong University)
Graph Neural NetworkReinforcement LearningGraphTabularBenchmark
🎯 What it does: Propose Deep Hierarchical Communication Graph (DHCG), a method in multi-agent reinforcement learning that achieves dynamic, dependency-driven intent sharing and communication by learning a directed acyclic graph (DAG).
Deep Multi-view Subspace Clustering with Anchor Graph
Chenhang Cui (University Of Electronic Science And Technology Of China), Lifang He (Lehigh University)
Auto EncoderContrastive LearningImage
🎯 What it does: This paper proposes the DMCAG method, which combines deep autoencoders, anchor graphs, and spectral self-supervised learning to achieve multi-view subspace clustering.
Deep Partial Multi-Label Learning with Graph Disambiguation
Haobo Wang (Zhejiang University), Gang Chen (Zhejiang University)
ClassificationGraph Neural NetworkImageGraph
🎯 What it does: A deep graph-based confusion elimination model called PLAIN is proposed for partial multi-label learning, which propagates pseudo-labels through similarity graphs at both instance and label levels to train a deep multi-label classifier.
Deep Symbolic Learning: Discovering Symbols and Rules from Perceptions
Alessandro Daniele (Fondazione Bruno Kessler), Luciano Serafini (Fondazione Bruno Kessler)
Explainability and InterpretabilityRepresentation LearningConvolutional Neural NetworkReinforcement LearningImage
🎯 What it does: Proposed an end-to-end differentiable neuro-symbolic learning framework called DSL, which can simultaneously learn perceptual functions and symbolic rules, and automatically generate symbols and map perceptual inputs under supervised NeSy function settings.
Deep Unfolding Convolutional Dictionary Model for Multi-Contrast MRI Super-resolution and Reconstruction
Pengcheng Lei (East China Normal University), Ming Xu (East China Normal University)
RestorationSuper ResolutionConvolutional Neural NetworkMultimodalityBiomedical DataMagnetic Resonance Imaging
🎯 What it does: Proposed a multi-contrast MRI super-resolution and reconstruction model (MC-CDic) based on deep unfolded convolutional dictionaries, achieving precise reference image transfer by explicitly separating common and unique features;
DeepPSL: End-to-End Perception and Reasoning
Sridhar Dasaratha (EY Global Delivery Services India LLP), Nigel P. Duffy (Ernst & Young LLP USA)
ClassificationRepresentation LearningConvolutional Neural NetworkGraph Neural NetworkImageTextGraph
🎯 What it does: Propose DeepPSL, an end-to-end trainable framework that combines deep learning with Probabilistic Soft Logic (PSL), achieving a unified system that integrates perception (deep network predicate prediction) and reasoning (HL-MRF).
DEIR: Efficient and Robust Exploration through Discriminative-Model-Based Episodic Intrinsic Rewards
Shanchuan Wan (University of Tokyo), Tomoyuki Kaneko (University of Tokyo)
Computational EfficiencyConvolutional Neural NetworkRecurrent Neural NetworkReinforcement LearningImage
🎯 What it does: Propose a DEIR intrinsic reward mechanism based on conditional mutual information scaling, which improves exploration efficiency in sparse reward environments by leveraging a discriminative forward model.
Delegated Online Search
Pirmin Braun (Goethe University Frankfurt), Conrad Schecker (Goethe University Frankfurt)
Optimization
🎯 What it does: Studies the delegated online search problem, analyzes decision-making between principals and agents, and provides theoretical lower and upper bounds for the approximation ratio achievable by the principal compared to the optimal search.
DeLELSTM: Decomposition-based Linear Explainable LSTM to Capture Instantaneous and Long-term Effects in Time Series
Chaoqun Wang (City University of Hong Kong), Zhixiang Huang (JD Digits)
Explainability and InterpretabilityRecurrent Neural NetworkTime Series
🎯 What it does: Proposed a DeLELSTM model that decomposes LSTM hidden states into a linear combination of past and current information, capturing the long-term and immediate effects of each variable to achieve interpretability in time series prediction.
Deliberation and Voting in Approval-Based Multi-Winner Elections
Kanav Mehra (University of Waterloo), Kate Larson (University of Waterloo)
Tabular
🎯 What it does: Studied the impact of deliberation in multi-winner elections on voting outcomes, and experimentally evaluated the interaction between different deliberation mechanisms and voting rules.
Deliberation as Evidence Disclosure: A Tale of Two Protocol Types
Julian Chingoma (University of Amsterdam), Adrian Haret (University of Amsterdam)
Tabular
🎯 What it does: A rational decision-making model based on evidence disclosure was constructed, and three deliberation protocols (simultaneous, sequential constant, sequential abstention) were proposed. The study investigates their convergence and decision accuracy under different agent behaviors (lazy, active) and evidence distribution conditions.
Denial-of-Service or Fine-Grained Control: Towards Flexible Model Poisoning Attacks on Federated Learning
Hangtao Zhang (Xiangtan University), Zhetao Li (Xiangtan University)
Federated LearningAdversarial AttackConvolutional Neural NetworkImage
🎯 What it does: Propose a flexible model poisoning attack (FMPA) in federated learning to achieve DoS attacks on the global model and finely controllable performance degradation.
Denoised Self-Augmented Learning for Social Recommendation
Tianle Wang (University of Hong Kong), Chao Huang (University of Hong Kong)
Recommendation SystemGraph Neural NetworkSupervised Fine-TuningContrastive LearningGraph
🎯 What it does: Propose a denoising self-incremental learning framework DSL for social recommendation, which can perform adaptive cross-view alignment between user-item interactions and social graphs, thereby improving recommendation quality under sparse data.
DenseDINO: Boosting Dense Self-Supervised Learning with Token-Based Point-Level Consistency
Yike Yuan (Zhejiang University), Xi Li (Zhejiang University)
ClassificationSegmentationRepresentation LearningTransformerContrastive LearningImage
🎯 What it does: Propose the DenseDINO transformer framework, which introduces reference tokens to perform self-supervised learning for point-level consistency, thereby enhancing performance on dense prediction tasks (e.g., semantic segmentation) while maintaining competitiveness on image-level tasks (e.g., classification).
Depth-Relative Self Attention for Monocular Depth Estimation
Kyuhong Shim (Seoul National University), Byonghyo Shim (Seoul National University)
Depth EstimationAutonomous DrivingTransformerImage
🎯 What it does: Propose the RED-T model, which incorporates relative depth as a bias in self-attention to guide the model to focus on pixels with similar depths in monocular depth estimation, thereby reducing reliance on RGB visual pitfalls (e.g., misleading information such as color, texture, and reflections).
Description Logics with Pointwise Circumscription
Federica Di Stefano (TU Wien), Mantas Šimkus (Umeå University)
Computational Efficiency
🎯 What it does: Propose the definition and reasoning method of pointwise circumscription in description logic (DL), addressing the complexity and undecidability issues caused by traditional global circumscription;
Detecting Adversarial Faces Using Only Real Face Self-Perturbations
Qian Wang (Huazhong University of Science and Technology), Ning Yu (Salesforce Research)
RecognitionAnomaly DetectionAdversarial AttackConvolutional Neural NetworkImage
🎯 What it does: Construct a detector that does not require pre-generated adversarial samples or access to the target face recognition system, using training data containing only real human faces and their self-perturbed samples. The detector cascades self-perturbation generation, decision boundary regularization, and max pooling classifier on input images, ultimately achieving detection of adversarial faces generated by unknown attack methods.
DFVSR: Directional Frequency Video Super-Resolution via Asymmetric and Enhancement Alignment Network
Shuting Dong (Tsinghua Shenzhen International Graduate School, Tsinghua University), Chun Yuan (Tsinghua Shenzhen International Graduate School, Tsinghua University)
Super ResolutionConvolutional Neural NetworkVideo
🎯 What it does: Designed and proposed the DFVSR network, which employs directional frequency representation, DFEA alignment, and an asymmetric U structure for video super-resolution.
Diagnose Like a Pathologist: Transformer-Enabled Hierarchical Attention-Guided Multiple Instance Learning for Whole Slide Image Classification
Conghao Xiong (Chinese University of Hong Kong), Irwin King (Chinese University of Hong Kong)
ClassificationTransformerImageBiomedical Data
🎯 What it does: Proposed a hierarchical attention-guided multi-instance learning (HAG-MIL) framework that mimics pathologists' layer-by-layer focus and classification on multi-resolution whole slide images (WSI).
Diagram Visual Grounding: Learning to See with Gestalt-Perceptual Attention
Xin Hu (Xi'an Jiaotong University), Qianying Wang (Lenovo Research)
Object DetectionConvolutional Neural NetworkTransformerLarge Language ModelVision Language ModelImageTextMultimodality
🎯 What it does: This paper proposes a 'Gestalt-Perceptual Attention' model (GPA) for diagram visual grounding, which enhances low-level visual features of diagrams by simulating human visual Gestalt laws and combines multi-modal context attention to improve high-level semantic representations, ultimately achieving precise localization of objects referred to by expressions.
Dichotomous Image Segmentation with Frequency Priors
Yan Zhou (Zhejiang University), Yanning Zhang (Northwestern Polytechnical University)
SegmentationConvolutional Neural NetworkTransformerImage
🎯 What it does: Studied methods to improve binary image segmentation (DIS) using frequency prior.
DiffAR: Adaptive Conditional Diffusion Model for Temporal-augmented Human Activity Recognition
Shuokang Huang (Imperial College London), Julie McCann (Imperial College London)
RecognitionConvolutional Neural NetworkTransformerDiffusion modelTime Series
🎯 What it does: Propose the DiffAR method, which utilizes an Adaptive Conditional Diffusion Model (ACDM) to perform time augmentation on WiFi CSI, including predicting future time periods and imputing missing values. The enhanced CSI is combined with the original CSI to train an ensemble classifier, achieving more robust human activity recognition.
Differentiable Economics for Randomized Affine Maximizer Auctions
Michael Curry (University of Zurich), John Dickerson (University of Maryland)
OptimizationFinance Related
🎯 What it does: Propose a differentiable randomized affine maximization auction (Lottery AMA) architecture, using gradient descent to learn weights, uplifts, and allocation parameters, thereby achieving multi-buyer, multi-item, and perfectly strategy-provable auctions.
Differentiable Model Selection for Ensemble Learning
James Kotary (University of Virginia), Ferdinando Fioretto (University of Virginia)
ClassificationOptimizationImage
🎯 What it does: Proposes an end-to-end differentiable combinatorial learning framework called e2e-CEL, which improves ensemble classifier performance by learning to select subset models.
Differentially Private Partial Set Cover with Applications to Facility Location
George Z. Li (University of Maryland), Anil Vullikanti (University of Virginia)
OptimizationSafty and PrivacyTabular
🎯 What it does: In this paper, the authors propose an algorithm for solving the Partial Set Cover problem under differential privacy constraints, and apply it to facility location planning in vaccine distribution scenarios.
Discounting in Strategy Logic
Munyque Mittelmann (University of Naples Federico II), Laurent Perrussel (University of Toulouse IRIT)
🎯 What it does: This paper introduces a discounting mechanism into Strategy Logic, defining a new logic SL_disc[D], and theoretically analyzes its model checking complexity under memoryless and memory-full strategies.
Discovering Sounding Objects by Audio Queries for Audio Visual Segmentation
Shaofei Huang (Chinese Academy of Sciences), Si Liu (Beihang University)
SegmentationTransformerVideoMultimodalityAudio
🎯 What it does: This work proposes the Audio-Queried Transformer (AQFormer), achieving frame-wise segmentation of sound-emitting objects in videos through audio-conditioned object queries;
Discrepancy-Guided Reconstruction Learning for Image Forgery Detection
Zenan Shi (Jilin University), Dong Zhang (Hong Kong University of Science and Technology)
ClassificationAnomaly DetectionConvolutional Neural NetworkGraph Neural NetworkImage
🎯 What it does: Propose an image forgery detection framework based on difference-guided reconstruction learning called DisGRL
Discrete Two Player All-Pay Auction with Complete Information
Marcin Dziubiński (University of Warsaw), Krzysztof Jahn (Warsaw University of Technology)
Finance Related
🎯 What it does: Analyzed discrete all-pay auctions with two players and complete information, providing a complete characterization of mixed Nash equilibrium.
Discriminative-Invariant Representation Learning for Unbiased Recommendation
Hang Pan (University of Science and Technology of China), Xiangnan He (University of Science and Technology of China)
Recommendation SystemRepresentation LearningContrastive Learning
🎯 What it does: This paper proposes a Discriminative-Invariant Representation Learning (DIRL) framework to address the selection bias problem in recommendation systems.
Disentanglement of Latent Representations via Causal Interventions
Gaël Gendron (University of Auckland), Gillian Dobbie (University of Auckland)
GenerationExplainability and InterpretabilityRepresentation LearningGraph Neural NetworkAuto EncoderImage
🎯 What it does: Propose a causality-intervention-based quantized variational autoencoder (CT-VAE) to achieve interpretable disentanglement of image causal factors and single-factor interventions.
DiSProD: Differentiable Symbolic Propagation of Distributions for Planning
Palash Chatterjee (Indiana University), Roni Khardon (Indiana University)
OptimizationRobotic IntelligenceBenchmarkStochastic Differential EquationOrdinary Differential Equation
🎯 What it does: Proposed and implemented DiSProD, a differentiable symbolic propagation distribution online planner for continuous random transition environments.
Distilling Universal and Joint Knowledge for Cross-Domain Model Compression on Time Series Data
Qing Xu (Institute for Infocomm Research, A*STAR), Zhenghua Chen (Institute for Infocomm Research, A*STAR)
CompressionDomain AdaptationKnowledge DistillationGenerative Adversarial NetworkTime Series
🎯 What it does: Propose an end-to-end cross-domain knowledge distillation framework called UNI-KD for compressing deep models in time series tasks, enabling knowledge transfer between source and target domains.
Distributional Multi-Objective Decision Making
Willem Röpke, Diederik M. Roijers (Vrije Universiteit Brussel)
OptimizationReinforcement LearningTabular
🎯 What it does: This paper proposes two novel multi-objective decision scheme sets, namely Distributed Undominated Set (DUS) and Convex Distributed Undominated Set (CDUS), and provides a learning and pruning algorithm based on distributed advantage criteria;
Diverse Approximations for Monotone Submodular Maximization Problems with a Matroid Constraint
Anh Viet Do (University of Adelaide), Frank Neumann (University of Adelaide)
OptimizationGraphBenchmark
🎯 What it does: This paper proposes two simple greedy algorithms for finding diverse solutions to the problem of maximizing a monotonic submodular function under the constraint of a supergraph, and provides corresponding approximation guarantees;
Diversity, Agreement, and Polarization in Elections
Piotr Faliszewski (AGH University), Tomasz Wąs (Pennsylvania State University)
OptimizationTabularBenchmark
🎯 What it does: Designed and implemented diversity and polarization metrics based on k-Kemeny distance, and validated their effectiveness across various statistical cultures through experiments.
Divide Rows and Conquer Cells: Towards Structure Recognition for Large Tables
Huawen Shen (Institute of Information Engineering, Chinese Academy of Sciences), Zhanzhan Cheng (Hikvision Research Institute)
RecognitionConvolutional Neural NetworkTransformerVision Language ModelImageBenchmark
🎯 What it does: This paper proposes a two-step Transformer architecture named DRCC for table structure recognition, directly converting table images into HTML text while predicting cell borders.
Do We Need an Encoder-Decoder to Model Dynamical Systems on Networks?
Bing Liu (Jilin University), Bo Yang (Jilin University)
Graph Neural NetworkGraphTime SeriesPhysics RelatedOrdinary Differential Equation
🎯 What it does: This paper provides a critical analysis of the commonly used encoder-decoder structure in network dynamics, pointing out that the embedding dimension leads to learning-induced fidelity loss, and proposes an embedding-free Dy-Network Neural Dynamics (DNND) model that directly learns self-dynamics and coupling terms in the original state space.
Domain-Adaptive Self-Supervised Face & Body Detection in Drawings
Barış Batuhan Topal (Koc University), Tevfik Metin Sezgin (Koc University)
Object DetectionDomain AdaptationKnowledge DistillationConvolutional Neural NetworkGenerative Adversarial NetworkImage
🎯 What it does: This paper proposes a facial and body detection framework based on self-supervised teacher-student networks and multi-style transfer pre-training, aiming to efficiently identify characters in comics and cartoon paintings.
Don't Ignore Alienation and Marginalization: Correlating Fraud Detection
Yilong Zang (Wuhan University), Lingfei Ren (Wuhan University)
Anomaly DetectionGraph Neural NetworkGraphFinance Related
🎯 What it does: This paper proposes a model called COFRAUD for detecting fraud by leveraging the interaction relevance of different relationships in multi-relational graphs.
Doubly Stochastic Graph-based Non-autoregressive Reaction Prediction
Ziqiao Meng (Chinese University of Hong Kong), Irwin King (Chinese University of Hong Kong)
Drug DiscoveryGraph Neural NetworkAuto EncoderGraph
🎯 What it does: Propose the ReactionSink framework, which uses the Sinkhorn algorithm to iteratively approximate the self-attention matrix to a doubly stochastic matrix, thereby simultaneously satisfying the electron counting rules and symmetry rules in non-autoregressive reaction prediction.
DPMAC: Differentially Private Communication for Cooperative Multi-Agent Reinforcement Learning
Canzhe Zhao (Shanghai Jiao Tong University), Shuai Li (Shanghai Jiao Tong University)
Safty and PrivacyReinforcement Learning
🎯 What it does: Proposes the DPMAC framework to achieve differential privacy communication in multi-agent reinforcement learning, ensuring that each agent can collaborate while protecting its own privacy during CTDE execution.
Dual Personalization on Federated Recommendation
Chunxu Zhang (Key Laboratory of Symbolic Computation and Knowledge Engineering of Ministry of Education), Bo Yang (Key Laboratory of Symbolic Computation and Knowledge Engineering of Ministry of Education)
Recommendation SystemFederated LearningMeta LearningTabular
🎯 What it does: Proposed the PFedRec framework to learn lightweight, edge-deployable personalized models in federated recommendation environments, achieving fine-grained personalization for scoring functions and item embeddings through a dual personalization mechanism.
Dual Relation Knowledge Distillation for Object Detection
Zhen-Liang Ni (Institute of Automation, Chinese Academy of Sciences), Gang Zhang (Department of Computer Vision Technology, Baidu Inc.)
Object DetectionKnowledge DistillationGraph Neural NetworkImage
🎯 What it does: Propose a dual-relationship knowledge distillation method (DRKD) that employs pixel-level and instance-level relation distillation to enable the student detector to learn global pixel relationships and instance relationships.
Dual Video Summarization: From Frames to Captions
Zhenzhen Hu (Hefei University of Technology), Richang Hong (Hefei University of Technology)
GenerationRetrievalRepresentation LearningRecurrent Neural NetworkTransformerVision Language ModelVideoTextRetrieval-Augmented Generation
🎯 What it does: This paper proposes a dual-modal video summarization framework. It first obtains visual features by encoding video frames with ViT. Then, it learns frame scores using self-attention combined with pseudo labels (ClipScore), selecting a small number of semantically consistent key frames (≤3% of video length) as video representations. Finally, it generates video captions using a lightweight LSTM decoder.
Dual-view Correlation Hybrid Attention Network for Robust Holistic Mammogram Classification
Zhiwei Wang (Huazhong University of Science and Technology), Xin Yang (Huazhong University of Science and Technology)
ClassificationConvolutional Neural NetworkImageBiomedical Data
🎯 What it does: Propose a dual-view correlation hybrid attention network (DCHA-Net) for overall classification of whole breast images.
Dynamic Belief for Decentralized Multi-Agent Cooperative Learning
Yunpeng Zhai (Peking University), Yonghong Tian (Peking University)
Reinforcement LearningAuto Encoder
🎯 What it does: Proposed a dynamic belief learning framework that adjusts its own strategy by predicting the dynamic strategies of other agents in decentralized multi-agent collaborative learning;
Dynamic Flows on Curved Space Generated by Labeled Data
Xinru Hua (Stanford University), Viet Anh Nguyen (Chinese University of Hong Kong)
Data SynthesisDomain AdaptationFlow-based ModelImage
🎯 What it does: This paper proposes a generative method based on gradient flow, which maps source domain and target domain data into a feature-Gaussian space (characterized by feature mean and covariance). A Riemannian gradient flow is constructed in this space, and the target distribution is progressively approximated using the MMD distance, thereby generating labeled samples in the target domain.
Dynamic Group Link Prediction in Continuous-Time Interaction Network
Shijie Luo (Xidian University), Jianbin Huang (Xidian University)
Recommendation SystemGraph Neural NetworkGraphSequential
🎯 What it does: Proposed an individual-group link prediction method called CTGLP in continuous-time dynamic networks, and designed core modules including CTGNN and importance-based module modeling.
Efficient and Equitable Deployment of Mobile Vaccine Distribution Centers
Da Qi Chen (University of Virginia), Anil Vullikanti (University of Virginia)
OptimizationTabularTime Series
🎯 What it does: Propose and solve the Dynamic k-Supplier problem for mobile vaccine distribution centers to minimize the travel distance of the population;
Efficient Computation of General Modules for ALC Ontologies
Hui Yang (Universite Paris-Saclay), Nicole Bidoit (Vrije Universiteit Amsterdam)
TextBenchmark
🎯 What it does: This paper proposes an algorithm for computing general modules of ALC ontologies, which maintains all implications under a specified signature while allowing the use of axioms not originally present, thereby achieving smaller and more concise sub-ontologies; simultaneously, this method can also generate deductive modules and uniform interpolants.
Efficient Multi-View Inverse Rendering Using a Hybrid Differentiable Rendering Method
Xiangyang Zhu (Tsinghua University), Bin Wang (Tsinghua University)
OptimizationComputational EfficiencyNeural Radiance FieldImageMesh
🎯 What it does: Propose a hybrid differentiable rendering method to simultaneously reconstruct triangle mesh geometry and physically-based PBR materials from real images captured with handheld cameras from multiple perspectives under uncontrolled lighting conditions.
Efficient NLP Model Finetuning via Multistage Data Filtering
Xu Ouyang (University of Virginia), Yangfeng Ji (University of Virginia)
Computational EfficiencyKnowledge DistillationData-Centric LearningTransformerSupervised Fine-TuningText
🎯 What it does: Propose a multi-stage data filtering method for NLP model fine-tuning, which dynamically skips unimportant training samples during training based on real-time training loss, significantly reducing forward and backward computational costs.
Efficient Object Search in Game Maps
Jinchun Du (Monash University), Adel Nadjaran Toosi (Monash University)
RetrievalComputational EfficiencyTextBenchmark
🎯 What it does: This paper proposes a lightweight index structure called Grid Tree for efficiently performing nearest neighbor search of text keywords in dynamic game maps.
Efficient Online Decision Tree Learning with Active Feature Acquisition
Arman Rahbar (Chalmers University of Technology), Morteza Haghir Chehreghani (Chalmers University of Technology)
ClassificationTabularFinance Related
🎯 What it does: Proposes an online decision tree learning framework (UFODT) based on posterior sampling and adaptive feature querying, achieving decision tree construction under fully online, feature-cost-sensitive settings.
Efficient Sign Language Translation with a Curriculum-based Non-autoregressive Decoder
Pei Yu (Xiamen University), Yidong Chen (Xiamen University)
Image TranslationTransformerVideoText
🎯 What it does: Proposed a completely non-autoregressive decoder (CND) based on curriculum learning, achieving generation of sign language translation results in a single decoding step, significantly reducing inference latency.
Eliminating the Computation of Strongly Connected Components in Generalized Arc Consistency Algorithm for AllDifferent Constraint
Luhan Zhen (Jilin University), Hongbo Li (Northeast Normal University)
OptimizationComputational EfficiencyGraphBenchmark
🎯 What it does: Proposed a new general arc consistency (GAC) algorithm for all-different constraints that completely eliminates the reliance on strong connected component (SCC) computation in traditional algorithms;