IJCAI 2024 Papers with AI Summaries
International Joint Conference on Artificial Intelligence · 790 papers
→ IJCAI 2024 papers with code (296)
Each paper below shows an AI-generated one-line summary. Get the full 6-part summary (innovation, method, data, results, limitations) and search all 790 IJCAI 2024 papers by keyword, author or institution —
free trial on arXivSub.
1DFormer: A Transformer Architecture Learning 1D Landmark Representations for Facial Landmark Tracking
Shi Yin (iFLYTEK Research), Cong Liu (iFLYTEK Research)
Object TrackingTransformerVideo
🎯 What it does: This paper proposes a Transformer architecture named 1DFormer for learning one-dimensional token representations, thereby achieving facial landmark tracking.
3D Vision and Language Pretraining with Large-Scale Synthetic Data
Dejie Yang (Peking University), Yang Liu (Peking University)
Data SynthesisDomain AdaptationRepresentation LearningTransformerVision Language ModelTextPoint Cloud
🎯 What it does: Constructed a large-scale synthetic 3D scene and text-aligned dataset named SynVL3D, pre-trained the SynFormer3D model using this dataset, and introduced simulation-to-real domain adaptation in downstream 3D vision-language tasks;
3DBench: A Scalable 3D Benchmark and Instruction-Tuning Dataset
Junjie Zhang (Shanghai University), Dan Zeng (Shanghai University)
Large Language ModelPrompt EngineeringTextPoint CloudBenchmark
🎯 What it does: Proposes the 3DBench evaluation framework and an automatically constructed 0.23M instruction fine-tuning dataset to comprehensively evaluate the perception, reasoning, and expression capabilities of 3D-LLMs.
A Behavior-Aware Approach for Deep Reinforcement Learning in Non-stationary Environments without Known Change Points
Zihe Liu (University of Technology Sydney), Junyu Xuan (University of Technology Sydney)
Reinforcement LearningImage
🎯 What it does: Proposes the BADA framework, which utilizes behavior embeddings and Wasserstein distance for threshold-free environment change detection, and achieves rapid policy adaptation through regularization after detecting changes;
A Better Approximation for Bipartite Traveling Tournament in Inter-League Sports Scheduling
Jingyang Zhao (University of Electronic Science and Technology of China), Mingyu Xiao (University of Electronic Science and Technology of China)
OptimizationGraphTabular
🎯 What it does: Proposed a randomized (3/2+ε) approximation algorithm for any scale BTTP (inter-league double round-robin tournament) problem
A Bias-Free Revenue-Maximizing Bidding Strategy for Data Consumers in Auction-based Federated Learning
Xiaoli Tang (Nanyang Technological University), Xiaoxiao Li (University of British Columbia)
OptimizationFederated LearningImageBenchmark
🎯 What it does: Propose a bias-free, revenue-maximizing bidding strategy called BR-FEDBIDDER, specifically designed for data consumers in federated learning;
A Coarse-to-Fine Fusion Network for Event-Based Image Deblurring
Huan Li (Institute of Microelectronics, Chinese Academy of Sciences), Xingyu Gao (Institute of Microelectronics, Chinese Academy of Sciences)
RestorationConvolutional Neural NetworkImageMultimodality
🎯 What it does: Propose a coarse-to-fine fusion network called CFFNet based on event cameras, which is used to recover a clear image from a single blurry image.
A Cognitive-Driven Trajectory Prediction Model for Autonomous Driving in Mixed Autonomy Environments
Haicheng Liao (University of Macau), Zhiyong Cui (Beihang University)
Autonomous DrivingConvolutional Neural NetworkRecurrent Neural NetworkGraph Neural NetworkTransformerGraphSequential
🎯 What it does: The study proposes and implements a cognition-driven trajectory prediction model that integrates perceptual safety assessment and driving behavior analysis for vehicle trajectory prediction in mixed autonomous environments.
A Complete Landscape of EFX Allocations on Graphs: Goods, Chores and Mixed Manna
Yu Zhou (City University of Hong Kong), Bo Li (Hong Kong Polytechnic University)
OptimizationGraph
🎯 What it does: This paper investigates the fair allocation of mixed goods (containing both good and bad items) in graph structures, systematically studying the existence and computational complexity of four EFX variants ("envy-free up to any item revocation") on graphs. For directed allocation (each edge can be assigned to only one of its endpoints), it proves that all four variants may not exist and the decision problem is NP-complete; however, in general allocation (edges can be assigned to any agent), except for the weakest EFX 0 0 variant, the other three variants always exist and can be constructed in polynomial time. Additionally, polynomial-time decision algorithms are provided for simple graphs such as trees, stars, and paths.
A Conservative Approach for Few-Shot Transfer in Off-Dynamics Reinforcement Learning
Paul Daoudi (Huawei Noah's Ark Lab), Ludovic Dos Santos (Criteo AI Lab)
Reinforcement Learning
🎯 What it does: Proposed the FOOD (Few-shOt Off Dynamics) algorithm, combining conservative RL regularization with simulation learning, leveraging the distribution gap between transition trajectories in source and target environments to achieve few-shot offline dynamic transfer.
A Consistency and Integration Model with Adaptive Thresholds for Weakly Supervised Object Localization
Hao Su (Sun Yat-sen University), Meng Yang (Sun Yat-sen University)
Object DetectionConvolutional Neural NetworkTransformerImage
🎯 What it does: Proposed a depth and shallow feature consistency and fusion model (CIAT) based on an adaptable threshold for weakly supervised object localization.
A Context-Enhanced Framework for Sequential Graph Reasoning
Shuo Shi (East China Normal University), Zhengfeng Yang (East China Normal University)
Representation LearningGraph Neural NetworkTransformerGraphSequentialBenchmark
🎯 What it does: Propose a context-enhanced framework (CEF), inserting a preprocessing module between the encoder and processor during serialized graph reasoning, utilizing historical hidden states to enhance features at each step and update context;
A Dataset and Model for Realistic License Plate Deblurring
Haoyan Gong (Xi'an Jiaotong-Liverpool University), Hongbin Liu (Xi'an Jiaotong-Liverpool University)
RestorationTransformerGenerative Adversarial NetworkImage
🎯 What it does: Proposed a large-scale real-world license plate motion blur dataset, LPBlur, and designed a generative adversarial network (GAN) called LPDGAN specifically for license plate deblurring.
A De-singularity Subgradient Approach for the Extended Weber Location Problem
Zhao-Rong Lai (Jinan University), Ziliang Chen (Peng Cheng Laboratory)
OptimizationTime SeriesFinance Related
🎯 What it does: Propose a subgradient method for removing singularities and the corresponding Weiszfeld algorithm (q-PWAWS) to address the singularity trap in the extended Weber location problem;
A Deep Probabilistic Spatiotemporal Framework for Dynamic Graph Representation Learning with Application to Brain Disorder Identification
Sin-Yee Yap (Monash University), David L. Dowe (Monash University)
ClassificationRepresentation LearningRecurrent Neural NetworkGraph Neural NetworkGraphTime SeriesMagnetic Resonance Imaging
🎯 What it does: Propose a variational Bayesian-based deep spatiotemporal graph learning framework, DSVB, for learning spatiotemporal embeddings of dynamic functional connectivity networks and identifying autism spectrum disorder (ASD)
A Deep Reinforcement Learning Approach to Balance Viewport Prediction and Video Transmission in 360° Video Streaming
Guanghui Zhang (Shandong University), Jing Guo (Hong Kong Polytechnic University)
OptimizationConvolutional Neural NetworkReinforcement LearningVideoTime Series
🎯 What it does: Proposed the QUTA system, which optimizes the QoE of 360° video streams by coordinating viewport prediction with video transmission parameters;
A Density-driven Iterative Prototype Optimization for Transductive Few-shot Learning
Jingcong Li (South China Normal University), Jiahui Pan (South China Normal University)
ClassificationMeta LearningImage
🎯 What it does: Proposed a density-based iterative prototype optimization method (DIPO) aimed at improving prototype quality in few-shot learning (FSL) to enhance model generalization.
A Fast Algorithm for MaxSAT above Half Number of Clauses
Junqiang Peng (University of Electronic Science and Technology of China), Mingyu Xiao (University of Electronic Science and Technology of China)
OptimizationComputational Efficiency
🎯 What it does: Studied a parameterized version of the MaxSAT problem, proposing a new algorithm with a runtime of O*(2^1479·µ), significantly improving the previous upper bound of O*(5^4064·µ).
A Fourier Perspective of Feature Extraction and Adversarial Robustness
Liangqi Zhang (Huazhong University of Science and Technology), Tianjiang Wang (Huazhong University of Science and Technology)
Adversarial AttackConvolutional Neural NetworkTransformerImage
🎯 What it does: The paper models adversarial attacks in the frequency domain, proposing a frequency-limited PGD attack method called F-PGD, and systematically studies feature extraction and robustness of deep networks across different frequency bands.
A General Black-box Adversarial Attack on Graph-based Fake News Detectors
Peican Zhu (Northwestern Polytechnical University), Zhen Wang (Northwestern Polytechnical University)
OptimizationAdversarial AttackGraph Neural NetworkTextGraph
🎯 What it does: Propose a general black-box attack framework called GAFSI, which can disrupt the predictions of graph neural network (GNN) fake news detectors under various graph structures by simulating the social interactions of malicious actors (posting, forwarding);
A Graph-based Representation Framework for Trajectory Recovery via Spatiotemporal Interval-Informed Seq2Seq
Yaya Zhao (Renmin University of China), Xiaoling Lu (Renmin University of China)
RestorationRepresentation LearningGraph Neural NetworkTransformerGraphSequential
🎯 What it does: In low-sampling-rate urban trajectory data, a graph-based trajectory recovery framework named GRFTrajRec is constructed, leveraging trajectory-road network interaction and spatiotemporal interval information to achieve precise recovery of trajectory points.
A Grassmannian Manifold Self-Attention Network for Signal Classification
Rui Wang (Jiangnan University), Xiaoning Song (Jiangnan University)
ClassificationConvolutional Neural NetworkTransformerTime SeriesBiomedical Data
🎯 What it does: This paper proposes a lightweight deep network called GDLNet based on a Grassmannian manifold self-attention mechanism (GMSA) for efficient feature extraction and classification of sequential signals (e.g., radar, EEG).
A Lightweight U-like Network Utilizing Neural Memory Ordinary Differential Equations for Slimming the Decoder
Quansong He (Sichuan University), Tao He (Sichuan University)
SegmentationConvolutional Neural NetworkBiomedical DataOrdinary Differential Equation
🎯 What it does: A lightweight modification was made to the decoder of the U-shaped network, proposing three discrete decoders based on neural memory ordinary differential equations (nmODE).
A Logic for Reasoning about Aggregate-Combine Graph Neural Networks
Pierre Nunn (University of Rennes), Nicolas Troquard (Gran Sasso Science Institute)
Explainability and InterpretabilityGraph Neural NetworkGraph
🎯 What it does: Proposed a modal logic K# with counting modalities and linear inequalities, and proved that it can fully represent aggregation-combination graph neural networks (AC-GNN).
A Meta-Game Evaluation Framework for Deep Multiagent Reinforcement Learning
Zun Li (Google DeepMind), Michael P. Wellman (University of Michigan)
Reinforcement LearningTabularBenchmark
🎯 What it does: Proposed a meta-game evaluation framework to statistically quantify the performance of deep multi-agent reinforcement learning (MARL) algorithms in general and sum games, and evaluated the performance of 17 mainstream MARL algorithms on negotiation games using this framework.
A Multi-Valued Decision Diagram-Based Approach to Constrained Optimal Path Problems over Directed Acyclic Graphs
Mingwei Zhang (Jinan University), Yong Lai (Jilin University)
OptimizationGraphBenchmark
🎯 What it does: Propose a constraint optimal path search algorithm (MCS) based on multi-valued decision diagrams (MDD), which solves the shortest/longest path problem on directed acyclic graphs (DAGs) with logical constraints by modeling path constraints as multi-valued functions and efficiently maintaining local constraints using MDD.
A Neural Column Generation Approach to the Vehicle Routing Problem with Two-Dimensional Loading and Last-In-First-Out Constraints
Yifan Xia (State Key Laboratory for Novel Software Technology, Nanjing University), Xiangyi Zhang (1QB Information Technologies, Inc.)
OptimizationRecurrent Neural NetworkTransformerTabular
🎯 What it does: Proposed a neural network-based column generation (NCG) algorithm that uses a machine learning model to predict the feasibility of two-dimensional loading under LIFO constraints, thereby improving traditional column generation for solving 2L-CVRP.
A New Guaranteed Outlier Removal Method Based on Plane Constraints for Large-Scale LiDAR Point Cloud Registration
Gang Ma (Fudan University), Jialiang Wu (Fudan University)
Pose EstimationPoint Cloud
🎯 What it does: This paper proposes a separable point cloud registration framework based on plane constraints, first using plane normals for robust rotation estimation, and then solving translation component-wise through plane translation constraints.
A New Paradigm for Counterfactual Reasoning in Fairness and Recourse
Lucius E.J. Bynum (New York University), Julia Stoyanovich (New York University)
Explainability and InterpretabilityTabular
🎯 What it does: This paper proposes a fairness and explainability evaluation framework based on backtracking counterfactuals, defining individual/group quantitative metrics for opportunity and effort, and providing corresponding fairness criteria.
A Prior-information-guided Residual Diffusion Model for Multi-modal PET Synthesis from MRI
Zaixin Ou (ShanghaiTech University), Dinggang Shen (ShanghaiTech University)
Data SynthesisVision Language ModelDiffusion modelMultimodalityBiomedical DataMagnetic Resonance ImagingPositron Emission TomographyAlzheimer's Disease
🎯 What it does: Propose a unified residual diffusion model for synthesizing multi-modal PET images from MRI, enabling low-cost, rapid joint diagnosis of Alzheimer's disease with multiple indicators.
A SAT Solver + Computer Algebra Attack on the Minimum Kochen–Specker Problem
Zhengyu Li (Georgia Institute of Technology), Vijay Ganesh (Georgia Institute of Technology)
OptimizationGraphPhysics Related
🎯 What it does: Proposed a verifiable proof generation method combining a SAT solver with a computer algebra system (CAS) to solve the three-dimensional Kochen-Specker (KS) problem and prove that the minimal KS system contains at least 24 vectors.
A Self-explaining Neural Architecture for Generalizable Concept Learning
Sanchit Sinha (University of Virginia), Aidong Zhang (University of Virginia)
Domain AdaptationExplainability and InterpretabilityRepresentation LearningAuto EncoderContrastive LearningImage
🎯 What it does: Propose a self-explaining neural network architecture for cross-domain generalizable concept learning.
A Semi-supervised Molecular Learning Framework for Activity Cliff Estimation
Fang Wu (Stanford University)
Drug DiscoveryGraph Neural NetworkBiomedical Data
🎯 What it does: Propose the SemiMol semi-supervised framework, which uses a teacher model to evaluate pseudo-label confidence and dynamically introduces pseudo-labels through adaptive curriculum learning to address the active cliff estimation problem under low-data scenarios.
A Strategic Analysis of Prepayments in Financial Credit Networks
Hao Zhou (University of Oxford), Michael Wooldridge (University of Oxford)
GraphFinance Related
🎯 What it does: Study prepayment behavior in financial credit networks, explore the impact of prepayment on network structure and corporate financial welfare, and analyze the prepayment game from both theoretical and experimental perspectives;
A Successful Strategy for Multichannel Iterated Prisoner’s Dilemma
Zhen Wang (Northwestern Polytechnical University), Chen Chu (Northwestern Polytechnical University)
Optimization
🎯 What it does: Proposed and analyzed a new multi-channel iterative prisoner's dilemma strategy called MCSUC based on cumulative reciprocity threshold decision-making
A Swap Relaxation-Based Local Search for the Latin Square Completion Problem
Zhenxuan Xie (Huazhong University of Science and Technology), Yuxuan Wang (Huazhong University of Science and Technology)
OptimizationBenchmark
🎯 What it does: This paper proposes a local search algorithm called SRLS based on exchange relaxation for solving the partial Latin square completion problem.
A Tensor-Based Formalization of the Event Calculus
Efthimis Tsilionis (National and Kapodistrian University of Athens), Georgios Paliouras (NCSR 'Demokritos')
RecognitionTabularTime Series
🎯 What it does: Tensorize the Event Calculus (EC) and solve its perfect model using linear algebra to achieve composite event recognition in the maritime domain.
A Top-Down Tree Model Counter for Quantified Boolean Formulas
Florent Capelli (University of Artois), Martina Seidl (Johannes Kepler University Linz)
Benchmark
🎯 What it does: Proposed a top-down counter called d4-QBF for tree model counting on any quantified prefix QBF, and implemented the complete algorithm and tool.
A Transformer-Based Adaptive Prototype Matching Network for Few-Shot Semantic Segmentation
Sihan Chen (Nanjing University of Information Science and Technology), Enhua Wu (Chinese Academy of Sciences)
SegmentationTransformerImage
🎯 What it does: Proposed a Transformer-based adaptive prototype matching network (TAMPN), which effectively mitigates inherent bias, attention bias, and spatial perception bias in few-shot semantic segmentation (FSS) through three modules (TEM, DCAM, DCM) during feature extraction, matching, and classification stages, achieving more robust few-shot semantic segmentation.
ABM: Attention before Manipulation
Fan Zhuo (Guangdong Laboratory of Artificial Intelligence and Digital Economy), Xilong Sun (Guangdong Laboratory of Artificial Intelligence and Digital Economy)
Robotic IntelligenceTransformerVision-Language-Action ModelImageTextMultimodality
🎯 What it does: Proposed an 'attention-first then operation' framework based on CLIP knowledge, using Object Mask Field to decouple visual localization and action prediction, achieving zero-shot and compositional generalization for robots with new objects and instructions.
Accelerating Diffusion Models for Inverse Problems through Shortcut Sampling
Gongye Liu (Tsinghua University), Yujiu Yang (Tsinghua University)
RestorationSuper ResolutionDiffusion modelImage
🎯 What it does: In the unsupervised zero-shot scenario, solving inverse problems using diffusion models by proposing a shortcut sampling path from 'input-transition state-target' to avoid starting from noise.
Active Deep Multi-view Clustering
Helin Zhao (Anhui University), Peng Zhou (Anhui University)
Representation LearningData-Centric LearningAuto EncoderContrastive LearningMultimodalityBenchmark
🎯 What it does: Propose an active deep multi-view clustering method called ADMC, which can guide the clustering process by actively selecting important samples for manual annotation.
Adaptive Order Q-learning
Tao Tan (Chongqing University), Defu Lian (University of Science and Technology of China)
Reinforcement Learning
🎯 What it does: Propose Order Q-learning and AdaOrder Q-learning, achieving fine-grained estimation bias control through order statistics from multiple Q-tables, and generalize to DQN.
ADELT: Transpilation between Deep Learning Frameworks
Linyuan Gong (University of California, Berkeley), Alvin Cheung (University of California, Berkeley)
Domain AdaptationAI Code AssistantTransformerLarge Language ModelGenerative Adversarial NetworkText
🎯 What it does: This study proposes ADELT, a source-to-source deep learning framework translator that leverages LLMs to translate code skeletons, then employs domain adversarial learning to generate API keyword mapping tables for cross-framework translation;
ADMN: Agent-Driven Modular Network for Dynamic Parameter Sharing in Cooperative Multi-Agent Reinforcement Learning
Yang Yu (School of Artificial Intelligence, University of Chinese Academy of Sciences), Kaiqi Huang (School of Artificial Intelligence, University of Chinese Academy of Sciences)
Reinforcement Learning
🎯 What it does: Propose Agent-Driven Modular Network (ADMN) to achieve dynamic parameter sharing and balance diversity
Advancing Generalized Transfer Attack with Initialization Derived Bilevel Optimization and Dynamic Sequence Truncation
Yaohua Liu (Dalian University of Technology), Risheng Liu (Dalian University of Technology)
Adversarial AttackConvolutional Neural NetworkImage
🎯 What it does: Proposed a two-level optimization framework called BETAK based on initialization to enhance the transferability of adversarial attacks against unknown target models.
Advancing Medical Image Segmentation via Self-supervised Instance-adaptive Prototype Learning
Guoyan Liang (Zhejiang University), Chang Yao (Zhejiang University)
SegmentationTransformerBiomedical DataMagnetic Resonance ImagingComputed Tomography
🎯 What it does: Proposed a self-supervised instance-adaptive prototype learning framework (SIPL) that utilizes shared prototypes and instance-specific prototypes for medical image segmentation.
Agentive Permissions in Multiagent Systems
Qi Shi (University of Southampton)
Agentic AI
🎯 What it does: Proposed and formalized four types of agent permissions in multi-agent systems (strong/weak + ensure/permit)
Aggregation and Purification: Dual Enhancement Network for Point Cloud Few-shot Segmentation
Guoxin Xiong (University of Science and Technology of China), Yongdong Zhang (University of Science and Technology of China)
SegmentationMeta LearningPoint Cloud
🎯 What it does: Proposed the Dual Enhancement Network (DENet) for few-shot semantic segmentation of point clouds.
Aggregation of Continuous Preferences in One Dimension
Alberto Del Pia (University of Wisconsin-Madison), Nimrod Talmon (Ben-Gurion University of Negev)
Optimization
🎯 What it does: This paper proposes a unified continuous preference one-dimensional aggregation framework, defines a cost function based on Lp and Lq norms, and formalizes the problem of solving it;
AK4Prompts: Aesthetics-driven Automatically Keywords-Ranking for Prompts in Text-To-Image Models
Haiyang Zhang (Beijing University of Posts and Telecommunications), Anlong Ming (Beijing University of Posts and Telecommunications)
GenerationTransformerPrompt EngineeringDiffusion modelImageText
🎯 What it does: Propose an automated keyword evaluation model, AK4Prompts, which can quantify the multidimensional impact of keywords on the image quality of text-to-image synthesis (TIS) models, and automatically select the optimal keywords for given prompts;
Alleviating Imbalanced Pseudo-label Distribution: Self-Supervised Multi-Source Domain Adaptation with Label-specific Confidence
Shuai Lü, Ximing Li (Jilin University)
ClassificationDomain AdaptationConvolutional Neural NetworkAuto EncoderContrastive LearningImage
🎯 What it does: Proposed a self-supervised multi-source domain adaptation method called S3DA-LC, which alleviates the problem of pseudo-label distribution imbalance by dynamically generating pseudo-labels using label-specific confidence to improve target domain performance.
AllMatch: Exploiting All Unlabeled Data for Semi-Supervised Learning
Zhiyu Wu (Peking University), Jinshi Cui (Peking University)
ClassificationData-Centric LearningImage
🎯 What it does: Propose a semi-supervised learning framework named AllMatch, which maximizes the utilization of unlabeled data through adaptive thresholds and binary classification consistency mechanisms.
Allocating Mixed Goods with Customized Fairness and Indivisibility Ratio
Bo Li (Hong Kong Polytechnic University), Zekai Wu (Harbin Institute of Technology)
Optimization
🎯 What it does: This paper studies fair division of mixed divisible and indivisible goods, proposing a fairness concept called 'proportionally tolerable' (EFα, PROPα) parameterized by individual 'indivisibility ratio,' and provides theoretical results on existence, computability, and efficiency compatibility.
AMO-aware Aggregates in Answer Set Programming
Mario Alviano (University of Calabria), Marco Maratea (University of Calabria)
OptimizationBenchmark
🎯 What it does: Designed and implemented a new ASP constraint called AMOSUM, which combines SUM aggregation with at-most-one (AMO) constraints and provides dedicated propagators to enhance the search space pruning capability of conflict-driven clause learning solvers.
An Archive Can Bring Provable Speed-ups in Multi-Objective Evolutionary Algorithms
Chao Bian (Nanjing University), Chao Qian (Nanjing University)
OptimizationComputational EfficiencyBenchmark
🎯 What it does: This paper theoretically analyzes the expected runtime of NSGA-II and SMS-EMOA when using an archive, and conducts experimental validation on two bi-objective Boolean problems, OneMinMax and LeadingOnesTrailingZeroes.
An Efficient Prototype-Based Clustering Approach for Edge Pruning in Graph Neural Networks to Battle Over-Smoothing
Yuyang Huang (Shanghai Jiao Tong University), Yang Yang (Shanghai Jiao Tong University)
Computational EfficiencyRepresentation LearningGraph Neural NetworkGraph
🎯 What it does: Propose a clustering method called ClusterDrop based on learnable prototypes for targeted edge removal during GNN training to alleviate the over-smoothing problem.
An Image-enhanced Molecular Graph Representation Learning Framework
Hongxin Xiang (Hunan University), Xiangxiang Zeng (Hunan University)
Knowledge DistillationRepresentation LearningDrug DiscoveryConvolutional Neural NetworkGraph Neural NetworkContrastive LearningImageGraphBiomedical DataBenchmark
🎯 What it does: Built a cross-modal knowledge distillation-based molecular graph representation learning framework called IEM, which enhances graph neural networks using multi-view 3D molecular images.
An NCDE-based Framework for Universal Representation Learning of Time Series
Zihan Liu (Beihang University), Leilei Sun (Beihang University)
ClassificationRepresentation LearningContrastive LearningTime SeriesStochastic Differential Equation
🎯 What it does: Propose CTRL, a self-supervised temporal representation learning framework based on neural controlled differential equations (NCDE), achieving general representations through dual tasks (reconstruction + contrastive) and masking augmentation.
AnchorGT: Efficient and Flexible Attention Architecture for Scalable Graph Transformers
Wenhao Zhu (Peking University), Shaoguo Liu (Alibaba Group)
Computational EfficiencyRepresentation LearningGraph Neural NetworkTransformerGraph
🎯 What it does: Propose AnchorGT, a Graph Transformer attention mechanism that utilizes k-dominating sets as anchors, reducing the original quadratic complexity of self-attention to near-linear while retaining global perception capabilities.
Angluin-Style Learning of Deterministic Büchi and Co-Büchi Automata
Yong Li (University of Liverpool), Qiyi Tang (University of Liverpool)
🎯 What it does: Proposed an Angluin-style active learning framework-based algorithm for deterministic Büchi automata (DBA) and deterministic co-Büchi automata (DCA), and reduced the complexity of equivalence queries by converting finite deterministic finite automata (FDFA) into DBA.
Anomaly Subgraph Detection through High-Order Sampling Contrastive Learning
Ying Sun (Tianjin University), Chunlong Bao (Hainan Tropical Ocean University)
Anomaly DetectionGraph Neural NetworkContrastive LearningGraph
🎯 What it does: Proposed an end-to-end anomaly subgraph detection framework ASD-HC based on GNN, which constructs node-subgraph instance pairs using high-order neighbor sampling, trains self-supervised contrastive learning to extract anomaly features, and combines non-parametric graph scan statistics with random walk restart algorithms to detect maximum connected anomaly subgraphs.
Ansatz-Agnostic Exponential Resource Saving in Variational Quantum Algorithms Using Shallow Shadows
Afrad Basheer (University of Technology Sydney), Sanjiang Li (University of Technology Sydney)
Physics Related
🎯 What it does: Proposed a variational quantum algorithm training method based on shallow shadows (AISO), achieving exponential quantum resource savings.
Anytime Sorting Algorithms
Emma Caizergues (Nokia Bell Labs), Fabien Mathieu (Swapcard Lab)
🎯 What it does: This paper proposes a general 'on-the-fly sorting' framework, and based on this framework, designs two new algorithms—Multizip Sort and Corsort—as well as a set of estimators based on partial information;
Apprenticeship-Inspired Elegance: Synergistic Knowledge Distillation Empowers Spiking Neural Networks for Efficient Single-Eye Emotion Recognition
Yang Wang (Dalian University of Technology), Xin Yang (Dalian University of Technology)
RecognitionKnowledge DistillationSpiking Neural NetworkMultimodality
🎯 What it does: Train a lightweight spiking neural network (SNN) for monocular emotion recognition using only intensity frames through multi-modal collaborative knowledge distillation.
Approximate Algorithms for k-Sparse Wasserstein Barycenter with Outliers
Qingyuan Yang (University of Science and Technology of China), Hu Ding (University of Science and Technology of China)
Anomaly DetectionOptimizationComputational EfficiencyImageTabular
🎯 What it does: Studied the k-sparse Wasserstein barycenter problem with outliers, and proposed a clustering-based LP algorithm as well as a (1+ε)-approximation scheme implemented in low-dimensional space.
Approximate Dec-POMDP Solving Using Multi-Agent A*
Wietze Koops (Radboud University), Nils Jansen (Ruhr University Bochum)
OptimizationReinforcement LearningBenchmark
🎯 What it does: This paper proposes an approximate solution algorithm based on multi-agent A* (small-step MAA*), aiming to solve Dec-POMDP over longer time horizons while providing feasible lower and upper bounds.
Are Logistic Models Really Interpretable?
Danial Dervovic (JP Morgan AI Research), Daniele Magazzeni (JP Morgan AI Research)
ClassificationExplainability and InterpretabilityTabularFinance Related
🎯 What it does: This paper points out the problems of traditional Logistic regression in explainability through user studies and experiments, and proposes a Linear Additive Model (LAM) to improve interpretability.
Are Watermarks Bugs for Deepfake Detectors? Rethinking Proactive Forensics
Xiaoshuai Wu (Hunan University), Zheng Qin (Hunan University)
Anomaly DetectionAdversarial AttackConvolutional Neural NetworkSupervised Fine-TuningGenerative Adversarial NetworkImage
🎯 What it does: Proposed a program called AdvMark that transforms robust watermarks into adversarial watermarks, achieving both image source tracking and improving the detection accuracy of deepfake detectors.
Atomic Recovery Property for Multi-view Subspace-Preserving Recovery
Yulong Wang (Huazhong Agricultural University)
ClassificationOptimizationRepresentation LearningMultimodality
🎯 What it does: This paper proposes the necessary and sufficient conditions for multi-view subspace preservation recovery—the atomic recovery property (ARP)—and generalizes the atomic norm to the multi-view atomic norm (MAN). Subsequently, ARP is applied to multi-view subspace clustering and classification, with geometric interpretability and invariance conditions provided.
ATTA: Adaptive Test-Time Adaptation for Multi-Modal Sleep Stage Classification
Ziyu Jia (Chinese Academy of Sciences), Tianzi Jiang (Chinese Academy of Sciences)
ClassificationDomain AdaptationMultimodalityBiomedical Data
🎯 What it does: Designed an adaptive test-time adaptation method for multi-modal sleep staging, ATTA, combining retaining-adapting modules, modal contribution evaluation, and adaptive learning rate to improve sleep staging performance.
Attention Based Document-level Relation Extraction with None Class Ranking Loss
Xiaolong Xu (Nanjing University of Information Science and Technology), Wanchun Dou (Nanjing University)
Graph Neural NetworkTransformerText
🎯 What it does: This paper proposes a document-level relation extraction model based on attention mechanisms, explicitly considering the no-relation scenario through None Class Ranking Loss to enhance multi-label prediction performance.
Attention Shifting to Pursue Optimal Representation for Adapting Multi-granularity Tasks
Gairui Bai (Xi'an Jiaotong University), Jizhong Zhao (Xi'an Jiaotong University)
ClassificationRecognitionObject DetectionSegmentationRepresentation LearningConvolutional Neural NetworkContrastive LearningImage
🎯 What it does: Proposes the SegAS method, which learns discriminative representations for multi-grained tasks through self-supervised learning via dynamic attention transfer and prototype consistency regularization.
Attribution Quality Metrics with Magnitude Alignment
Chase Walker (University of Central Florida), Rickard Ewetz (University of Central Florida)
ClassificationExplainability and InterpretabilityConvolutional Neural NetworkTransformerImage
🎯 What it does: Proposed the Magnitude Aligned Scoring (MAS) evaluation metric, addressing the issue of existing perturbation tests neglecting the magnitude of feature importance by introducing an alignment penalty between model responses and normalized density responses.
AutoAgents: A Framework for Automatic Agent Generation
Guangyao Chen (Peking University), Yemin Shi (Peking University)
TransformerLarge Language ModelAgentic AIText
🎯 What it does: Propose the AutoAgents framework, which can automatically generate specialized agents based on tasks and enable collaborative execution.
Automated CPU Design by Learning from Input-Output Examples
Shuyao Cheng (State Key Lab of Processors Institute of Computing Technology Chinese Academy of Sciences), Yunji Chen (State Key Lab of Processors Institute of Computing Technology Chinese Academy of Sciences)
OptimizationComputational EfficiencyGraph Neural NetworkTabular
🎯 What it does: Automatically design a 32-bit RISC-V CPU (Enlightenment-1) using only input-output examples, achieving a complete closed-loop from IO to RTL, ultimately generating a processor capable of running Linux on FPGA and silicon chips.
Automatic De-Biased Temporal-Relational Modeling for Stock Investment Recommendation
Weijun Chen (Peking University), Tengjiao Wang (Peking University)
Recommendation SystemMeta LearningConvolutional Neural NetworkRecurrent Neural NetworkGraph Neural NetworkTime SeriesSequentialFinance Related
🎯 What it does: Proposed an automatic bias-removal temporal relational model (ADB-TRM) for stock investment recommendations.
BADFSS: Backdoor Attacks on Federated Self-Supervised Learning
Jiale Zhang (Yangzhou University), Guodong Long (University of Technology Sydney)
Federated LearningRepresentation LearningAdversarial AttackImage
🎯 What it does: This paper studies backdoor attacks in the federated self-supervised learning (FSSL) environment and proposes for the first time a backdoor strategy that can be implanted during the FL aggregation process and remain effective.
BadFusion: 2D-Oriented Backdoor Attacks against 3D Object Detection
Saket S. Chaturvedi (Clemson University), Xiaoyong Yuan (Clemson University)
Object DetectionAutonomous DrivingAdversarial AttackConvolutional Neural NetworkImageMultimodalityPoint Cloud
🎯 What it does: Propose BadFusion, a 2D camera-guided backdoor attack targeting LiDAR-camera fusion based 3D object detection;
Balancing Multimodal Learning via Online Logit Modulation
Daoming Zong (SenseTime Research), Ken Zheng (SenseTime Research)
ClassificationRecognitionOptimizationTransformerImageVideoTextMultimodalityAudio
🎯 What it does: This paper proposes an intermediate representation fusion module (IRFB) that decouples unimodal feature learning from cross-modal interaction, as well as an online logit modulation (OLM) technique that dynamically adjusts the logit amplitude of each modality in real-time, to achieve balanced and enhanced multimodal learning.
Bandits with Concave Aggregated Reward
Yingqi Yu (University of Science and Technology of China), Xiang-Yang Li (University of Science and Technology of China)
Reinforcement Learning
🎯 What it does: Studied the multi-armed bandit problem with concave aggregate rewards and proposed two algorithms, SW-BCAR and SWUCB-BCAR, to address this issue.
BATON: Aligning Text-to-Audio Model Using Human Preference Feedback
Huan Liao (Tsinghua University), Xiu Li (Tsinghua University)
GenerationSupervised Fine-TuningPrompt EngineeringDiffusion modelTextAudio
🎯 What it does: Introduces the BATON framework, which aligns text-audio generation models using human preference feedback through three steps: dataset construction, reward model training, and model fine-tuning.
Best Arm Identification with Retroactively Increased Sampling Budget for More Resource-Efficient HPO
Jasmin Brandt (Paderborn University), Eyke Hüllermeier
OptimizationComputational EfficiencyHyperparameter SearchBenchmark
🎯 What it does: An incremental Successive Halving (iSHA) algorithm is proposed for non-random multi-armed bandit (Best Arm Identification) and hyperparameter optimization (HPO), which can increase the maximum budget after the original SHA runs as needed, while providing corresponding theoretical guarantees and experimental validation.
Beyond Alignment: Blind Video Face Restoration via Parsing-Guided Temporal-Coherent Transformer
Kepeng Xu (Xidian University), Yunsong Li (Xidian University)
RestorationTransformerAuto EncoderVideo
🎯 What it does: Proposes an end-to-end alignment-free blind video face restoration framework named PGTFormer, which utilizes a semantically guided spatiotemporal vector quantized autoencoder and a spatiotemporally parsed codebook predictor to achieve high-quality and temporally coherent video face reconstruction.
Beyond What If: Advancing Counterfactual Text Generation with Structural Causal Modeling
Ziao Wang (Harbin Institute of Technology), Hongwei Du (Harbin Institute of Technology)
GenerationTransformerLarge Language ModelTextFinance Related
🎯 What it does: This paper proposes a counterfactual text generation method based on Structural Causal Models (SCM), aiming to explore complex multi-layer causal relationships and surpass traditional single-causal studies.
BeyondVision: An EMG-driven Micro Hand Gesture Recognition Based on Dynamic Segmentation
Nana Wang (Beihang University), Hao Su (Zhengzhou University)
RecognitionComputational EfficiencyConvolutional Neural NetworkBiomedical Data
🎯 What it does: This paper proposes a micro-gesture recognition system called BeyondVision, which includes a wearable wristband EMG acquisition device, a lightweight CNN network BV-Net, and a post-processing algorithm based on weight segmentation, achieving real-time mapping from micro-gestures to control commands.
BlockEcho: Retaining Long-Range Dependencies for Imputing Block-Wise Missing Data
Qiao Han (Zhejiang Lab), Yiteng Zhai (Zhejiang Lab)
Data SynthesisGenerative Adversarial NetworkTabular
🎯 What it does: Proposed a method called BlockEcho that integrates matrix factorization with generative adversarial networks for missing block data imputation, and validated its effectiveness on multi-domain datasets.
BoostDream: Efficient Refining for High-Quality Text-to-3D Generation from Multi-View Diffusion
Yonghao Yu (Waseda University), Haorui Li (Southeast University)
GenerationData SynthesisComputational EfficiencyVision Language ModelDiffusion modelScore-based ModelNeural Radiance FieldGaussian SplattingImageTextMesh
🎯 What it does: Propose BoostDream, a three-stage efficient refinement framework that converts rough 3D assets generated by feed-forward processes into high-quality, controllable 3D assets.
Boosting Diffusion Models with an Adaptive Momentum Sampler
Xiyu Wang (University of Sydney), Chang Xu (University of Sydney)
GenerationDiffusion modelImage
🎯 What it does: Proposed an adaptive momentum-based untrained backward sampler to enhance the quality of images generated by diffusion models
Boosting Efficiency in Task-Agnostic Exploration through Causal Knowledge
Yupei Yang (Shanghai Jiao Tong University), Lei Xu (Shanghai Jiao Tong University)
Autonomous DrivingComputational EfficiencyReinforcement LearningWorld ModelPhysics Related
🎯 What it does: Proposes a task-agnostic reinforcement learning framework based on causal exploration, leveraging online causal discovery and structural constraints to enhance the sample efficiency and reliability of world model learning.
Boosting Model Resilience via Implicit Adversarial Data Augmentation
Xiaoling Zhou (Peking University), Shikun Zhang (Peking University)
Data-Centric LearningMeta LearningImage
🎯 What it does: Proposes an implicit data augmentation method that samples adversarial and non-adversarial perturbation distributions within the sample depth feature space, and achieves model optimization without explicit augmentation through a meta-learning framework;
Boosting Single Positive Multi-label Classification with Generalized Robust Loss
Yanxi Chen (University of International Business and Economics), Bo Wang (University of International Business and Economics)
ClassificationImage
🎯 What it does: Proposed a general robust loss framework (GR Loss) for single positive multi-label learning (SPML), addressing missing labels and class imbalance through soft pseudo labels and weight adjustment.
Boundary-aware Decoupled Flow Networks for Realistic Extreme Rescaling
Jinmin Li (Tsinghua University), Rizen Guo (Tencent)
RestorationConvolutional Neural NetworkFlow-based ModelImage
🎯 What it does: Designed a boundary-aware separated flow network (BDFlow), which achieves more realistic image resizing by splitting high-frequency information into semantic boundary distribution and non-semantic Gaussian distribution.
Breaking Barriers of System Heterogeneity: Straggler-Tolerant Multimodal Federated Learning via Knowledge Distillation
Jinqian Chen (Shandong University), Liqiang Nie (Harbin Institute of Technology)
Federated LearningKnowledge DistillationTransformerVision Language ModelMultimodality
🎯 What it does: This paper proposes a robust framework called MFL-AKD for multimodal federated learning, which addresses the straggler problem caused by system heterogeneity by leveraging knowledge distillation and fast knowledge transfer mechanisms;
Bridge to Non-Barrier Communication: Gloss-Prompted Fine-Grained Cued Speech Gesture Generation with Diffusion Model
Wentao Lei (Hong Kong University of Science and Technology (Guangzhou)), Jun Wang (Tencent AI Lab)
GenerationLarge Language ModelPrompt EngineeringDiffusion modelContrastive LearningVideoTextMultimodalityAudio
🎯 What it does: A diffusion model based on Gloss is studied to generate fine-grained lip and gesture movements for visual expression of Chinese Cued Speech.
Bridging Generative and Discriminative Models for Unified Visual Perception with Diffusion Priors
Shiyin Dong (Xidian University), Xinbo Gao (Chongqing University of Posts and Telecommunications)
ClassificationSegmentationRetrievalConvolutional Neural NetworkTransformerDiffusion modelImageText
🎯 What it does: Propose a unified framework Vermouth, which uses the pre-trained Stable Diffusion as a prior, and migrates the latent representations of generative models to visual perception tasks through U-head and Adapted-Expert.
Bridging LiDAR Gaps: A Multi-LiDARs Domain Adaptation Dataset for 3D Semantic Segmentation
Shaoyang Chen (Xiamen University), Cheng Wang (Xiamen University)
SegmentationDomain AdaptationAutonomous DrivingContrastive LearningPoint CloudBenchmark
🎯 What it does: This study proposes a multi-LiDAR synchronized calibration 3D semantic segmentation domain adaptation dataset (MLDAS), and based on this dataset, introduces a Hierarchical Spatial Consistency Network (HSSC) to achieve unsupervised domain adaptation across LiDARs and scenes.
Bridging Stereo Geometry and BEV Representation with Reliable Mutual Interaction for Semantic Scene Completion
Bohan Li (Shanghai Jiao Tong University), Wenjun Zeng (Shanghai Jiao Tong University)
SegmentationDepth EstimationAutonomous DrivingConvolutional Neural NetworkTransformerOptical FlowImage
🎯 What it does: Proposed a framework named BRGScene, which integrates stereo matching with bird's-eye view (BEV) 3D volumes to achieve monocular camera-based 3D semantic scene completion (SSC)
Bridging the Gap between General and Down-Closed Convex Sets in Submodular Maximization
Loay Mualem (University of Haifa), Moran Feldman (University of Haifa)
OptimizationGraph
🎯 What it does: Proposed offline and online algorithms for maximizing DR-submodular functions under non-lower closed convex constraints.
Bridging the Gap: Learning Pace Synchronization for Open-World Semi-Supervised Learning
Bo Ye (Southeast University), Min-Ling Zhang (Southeast University)
ClassificationContrastive LearningImage
🎯 What it does: Propose a Learning Rhythm Synchronization (LPS) framework that maintains consistent learning rates between known and unknown classes through adaptive margin loss and pseudo-label contrastive clustering, achieving better new class discovery and known class classification in open semi-supervised learning.