IJCAI 2023 Papers — Page 3
International Joint Conference on Artificial Intelligence · 639 papers
Enabling Abductive Learning to Exploit Knowledge Graph
Yu-Xuan Huang (Nanjing University), Zhi-Hua Zhou (Nanjing University)
ClassificationImageTextGraphTabular
🎯 What it does: Propose the ABL-KG framework, which can automatically mine and memorize logical rules from large-scale knowledge graphs without manual annotation of knowledge bases, and then use these rules in abductive learning (ABL) to improve machine learning models.
Engineering an Efficient Approximate DNF-Counter
Mate Soos (National University of Singapore), Maciej Obremski (Centre for Quantum Technologies)
OptimizationComputational EfficiencyBenchmark
🎯 What it does: Proposes a new efficient approximate DNF counter called pepin, which can quickly estimate the number of satisfying solutions for DNF formulas under a given error tolerance.
Enhancing Datalog Reasoning with Hypertree Decompositions
Xinyue Zhang (University of Oxford), Ian Horrocks (University of Oxford)
Computational EfficiencyGraph
🎯 What it does: This paper proposes an algorithm that utilizes hypertree decomposition to materialize and perform incremental reasoning for recursive Datalog programs, combining it with traditional semi-naïve methods and the DRed maintenance framework to form a pluggable modular reasoning system.
Enhancing Efficient Continual Learning with Dynamic Structure Development of Spiking Neural Networks
Bing Han (Chinese Academy of Sciences), Guobin Shen (Chinese Academy of Sciences)
Computational EfficiencySpiking Neural NetworkImage
🎯 What it does: Propose a dynamic structure development spiking neural network (DSD-SNN) that achieves continual learning for multi-task scenarios through random growth of new neurons, adaptive pruning, and freezing mechanisms.
Enhancing Network by Reinforcement Learning and Neural Confined Local Search
Qifu Hu (Inspur Electronic Information Industry Co., Ltd), Rengang Li (Inspur Electronic Information Industry Co., Ltd)
OptimizationGraph Neural NetworkTransformerReinforcement LearningGraph
🎯 What it does: Propose two reinforcement learning models for the Network Enhancement Problem (NEP): NEP-AM (domain knowledge-enhanced attention model) and NEP-HAM (hierarchical attention model), and design Neural Confined Local Search (NCLS) to efficiently search for improvement solutions within a large neighborhood.
Enriching Phrases with Coupled Pixel and Object Contexts for Panoptic Narrative Grounding
Tianrui Hui (Chinese Academy of Sciences), Si Liu (Beihang University)
SegmentationRetrievalConvolutional Neural NetworkTransformerLarge Language ModelContrastive LearningMultimodality
🎯 What it does: Propose a PNG method that enriches phrase features by combining pixel-level and object-level contexts.
Ensemble Reinforcement Learning in Continuous Spaces -- A Hierarchical Multi-Step Approach for Policy Training
Gang Chen (Victoria University of Wellington), Victoria Huang (National Institute of Water and Atmospheric Research)
Reinforcement Learning
🎯 What it does: This paper proposes a hierarchical multi-step ensemble deep deterministic policy gradient algorithm (HED) for reinforcement learning in continuous action spaces;
Error in the Euclidean Preference Model
Luke Thorburn (King's College London), Carmine Ventre (King's College London)
Optimization
🎯 What it does: This paper theoretically investigates the expressiveness of Euclidean space preference models, providing a lower bound on the proportion of non-Euclidean preference configurations and their approximation error.
Expanding the Hyperbolic Kernels: A Curvature-aware Isometric Embedding View
Meimei Yang (Southeast University), Hui Xue (Southeast University)
ClassificationRecognitionImageGraph
🎯 What it does: Propose a curvature-aware isometric embedding that maps the hyperbolic space of the Poincaré model to a reproducing kernel Hilbert space (RKHS), thereby constructing a family of new hyperbolic kernels (including positive definite and indefinite kernels) in RKHS, and verifying their performance in graph learning and zero-shot learning tasks.
Explainable Multi-Agent Reinforcement Learning for Temporal Queries
Kayla Boggess (University of Virginia), Lu Feng (University of Virginia)
Explainability and InterpretabilityReinforcement LearningTime SeriesSequential
🎯 What it does: This paper proposes a method that generates strategy-level contrastive explanations for time series queries using an abstract model of multi-agent reinforcement learning (MARL) policies, and determines whether the query is feasible; if not feasible, it generates complete and correct explanations to clarify the reasons for failure.
Explainable Reinforcement Learning via a Causal World Model
Zhongwei Yu (Institute of Automation, Chinese Academy of Sciences), Dengpeng Xing (Institute of Automation, Chinese Academy of Sciences)
Explainability and InterpretabilityRecurrent Neural NetworkReinforcement LearningWorld Model
🎯 What it does: Propose an interpretable world model in reinforcement learning without prior causal structure, which learns environment dynamics through causal discovery and an attention inference network, and generates interpretable explanations for action decisions via causal chains.
Explainable Text Classification via Attentive and Targeted Mixing Data Augmentation
Songhao Jiang (Institute of Information Engineering, Chinese Academy of Sciences), Bo Wang (CNCERT/CC)
ClassificationExplainability and InterpretabilityConvolutional Neural NetworkTransformerText
🎯 What it does: Proposes a clause-level targeted mixing data augmentation method called ATMIX based on self-attention, aiming to improve the performance and interpretability of text classification models.
Explaining Answer-Set Programs with Abstract Constraint Atoms
Thomas Eiter (TU Wien), Tobias Geibinger (TU Wien)
Explainability and InterpretabilityComputational Efficiency
🎯 What it does: This paper proposes an explanation method for answer set programs containing abstract constraint atoms (c-atoms), defining model-based m-justification and rule-based r-justification, and combining both approaches.
Explanation-Guided Reward Alignment
Saaduddin Mahmud (University of Massachusetts Amherst), Shlomo Zilberstein (University of Massachusetts Amherst)
Explainability and InterpretabilityReinforcement Learning from Human FeedbackReinforcement Learning
🎯 What it does: This paper proposes REVEALE, an explanation-based reward alignment and verification framework, which learns reward functions from ranked trajectories using inverse reinforcement learning, and gradually eliminates reward ambiguity through human approval of explanations or ranking feedback, ultimately obtaining rewards that better align with true intentions.
Exploiting Non-Interactive Exercises in Cognitive Diagnosis
Fangzhou Yao (University of Science and Technology of China), Shijin Wang (University of Science and Technology of China)
Data-Centric LearningTabularSequential
🎯 What it does: In cognitive diagnosis, the authors propose an EIRS (Exercise-aware Informative Response Sampling) framework, which alleviates the long-tail problem caused by sparse student interactions by leveraging the potential sorting information of uninteracted exercises.
Exploration via Joint Policy Diversity for Sparse-Reward Multi-Agent Tasks
Pei Xu (University of Chinese Academy of Sciences), Kaiqi Huang (University of Chinese Academy of Sciences)
Reinforcement LearningBenchmark
🎯 What it does: In multi-agent reinforcement learning tasks with sparse rewards, an exploration method based on Constrained Joint Policy Diversity (JPD) is proposed, combined with traditional count rewards based on state uncertainty, improving exploration efficiency.
Exploring Effective Inter-Encoder Semantic Interaction for Document-Level Relation Extraction
Liang Zhang (Xiamen University), Yidong Chen (Xiamen University)
Knowledge DistillationRepresentation LearningGraph Neural NetworkTransformerLarge Language ModelTextGraph
🎯 What it does: Propose a document-level relation extraction model based on a graph-Transformer network (GTN), which can simultaneously capture local and global semantics in the document graph (HDG), and bidirectionally couple the document encoder and graph encoder through cross-attention sublayers.
Exploring Leximin Principle for Fair Core-Selecting Combinatorial Auctions: Payment Rule Design and Implementation
Hao Cheng (Nanjing University), Chongjun Wang (Nanjing University)
OptimizationTabularBenchmark
🎯 What it does: Studied core selection in combinatorial auctions, proposed a fair payment rule based on the Leximin principle (BLO), and provided an implementation algorithm.
Exploring Safety Supervision for Continual Test-time Domain Adaptation
Xu Yang (Xidian University), Cheng Deng (Xidian University)
Domain AdaptationContrastive LearningImage
🎯 What it does: Propose a safe supervision framework for continuous test time domain adaptation, utilizing adaptive thresholds, soft weighted contrastive learning, and soft weight alignment to improve pseudo-label quality and alleviate knowledge forgetting.
Exploring Structural Similarity in Fitness Landscapes via Graph Data Mining: A Case Study on Number Partitioning Problems
Mingyu Huang (University of Electronic Science and Technology of China), Ke Li (University of Exeter)
OptimizationGraph
🎯 What it does: Construct and combine the Local Optima Network (LON) with graph data mining techniques to conduct qualitative and quantitative analysis of the fitness landscapes of the Number Partitioning Problem (NPP) across different dimensions, exploring their structural similarity.
Fair and Efficient Allocation of Indivisible Chores with Surplus
Hannaneh Akrami (Max Planck Institute for Informatics), Ruta Mehta (University of Illinois at Urbana-Champaign)
Optimization
🎯 What it does: Investigated chore allocation with additive utility functions, constructing an allocation that satisfies both EF1 and fPO, and introduced the concept of up to n-1 surplus.
Fair Division of a Graph into Compact Bundles
Jayakrishnan Madathil (University of Glasgow)
OptimizationGraph
🎯 What it does: The study replaces traditional connectivity constraints with 'compactness' constraints when allocating indivisible items on a graph, exploring feasible allocations under fairness concepts such as proportionality, fairness, and maximin share, and analyzing their computational complexity.
Fair Division with Two-Sided Preferences
Ayumi Igarashi (University of Tokyo), Hanna Sumita (Tokyo Institute of Technology)
Optimization
🎯 What it does: Study the fairness and stability issues of allocating indivisible players to teams under two-sided preference conditions, propose and prove the existence of allocations satisfying EF1, exchange stability, and individual stability, provide an EF1+Pareto optimal algorithm for non-negative value players, and explore the compatibility between rationality and justified envy;
Fairly Allocating Goods and (Terrible) Chores
Hadi Hosseini (Pennsylvania State University), Tomasz Wąs (Pennsylvania State University)
Optimization
🎯 What it does: This paper studies the issues of fairness and efficiency in allocating mixed items (containing both goods and bads) under lexicographic preferences, focusing on fairness concepts such as EFX, EF1, and MMS, and provides corresponding existence and computational complexity results.
Fairness via Group Contribution Matching
Tianlin Li (Nanyang Technological University), Yang Liu (Nanyang Technological University)
OptimizationExplainability and InterpretabilityImageTabular
🎯 What it does: This paper studies how fairness develops progressively during the training of deep learning models and proposes mitigating bias by matching contributions across different subgroups.
Fast Algorithms for SAT with Bounded Occurrences of Variables
Junqiang Peng (University of Electronic Science and Technology of China), Mingyu Xiao (University of Electronic Science and Technology of China)
Computational Efficiency
🎯 What it does: This paper proposes a deterministic algorithm for CNF SAT where each variable occurs a limited number of times (d ≤ 5), and provides the corresponding upper bound on time complexity.
Fast-StrucTexT: An Efficient Hourglass Transformer with Modality-guided Dynamic Token Merge for Document Understanding
Mingliang Zhai (Beijing Institute of Technology), Yunde Jia (Beijing Institute of Technology)
RecognitionComputational EfficiencyRepresentation LearningTransformerVision Language ModelMultimodality
🎯 What it does: Propose Fast-StrucTexT, a Transformer that accelerates visual document understanding through an hourglass structure and modality-guided dynamic token merging.
Faster Exact MPE and Constrained Optimization with Deterministic Finite State Automata
Filippo Bistaffa (Artificial Intelligence Institute, Spanish National Research Council)
OptimizationComputational EfficiencyBenchmark
🎯 What it does: Propose a function representation based on Deterministic Acyclic Finite State Automaton (DAFSA) for performing exact Most Probable Explanation (MPE) and constraint optimization tasks in Bucket Elimination (BE), forming the FABE algorithm.
FEAMOE: Fair, Explainable and Adaptive Mixture of Experts
Shubham Sharma (University of Texas at Austin), Joydeep Ghosh (University of Texas at Austin)
ClassificationDomain AdaptationExplainability and InterpretabilityMixture of ExpertsTabular
🎯 What it does: Propose FEAMOE, a hybrid expert model that can adapt online to data drift while simultaneously considering fairness and interpretability.
Feature Staleness Aware Incremental Learning for CTR Prediction
Zhikai Wang (Shanghai Jiao Tong University), Kangyi Lin (Tencent)
Recommendation SystemTabular
🎯 What it does: Propose a solution called FeSAIL for the feature obsolescence problem in incremental learning of CTR prediction models, which can adaptively replay samples with obsolete features and adjust feature embedding updates.
FedBFPT: An Efficient Federated Learning Framework for Bert Further Pre-training
Xin'ao Wang (Zhejiang University), Lidan Shou (Zhejiang University)
Federated LearningComputational EfficiencyRepresentation LearningTransformerLarge Language ModelText
🎯 What it does: Propose the FEDBFPT framework, which utilizes federated learning to further pre-train only partial layers of the BERT model on clients, and progressively advances layer by layer through the PL-SDL strategy, ultimately generating a global BERT model applicable to downstream tasks.
FedDWA: Personalized Federated Learning with Dynamic Weight Adjustment
Jiahao Liu (Sun Yat-sen University), Di Wu (Sun Yat-sen University)
Federated LearningImage
🎯 What it does: Propose the FedDWA algorithm, achieving personalized federated learning through dynamic weight adjustment
Federated Graph Semantic and Structural Learning
Wenke Huang (Wuhan University), Bo Du (Wuhan University)
Federated LearningKnowledge DistillationGraph Neural NetworkContrastive LearningGraph
🎯 What it does: Proposes the Federated Graph Semantic and Structural Learning (FGSSL) framework, which simultaneously calibrates node semantic and graph structural biases in federated graph learning scenarios.
Federated Probabilistic Preference Distribution Modelling with Compactness Co-Clustering for Privacy-Preserving Multi-Domain Recommendation
Weiming Liu (Zhejiang University), Longfei Zheng (Ant Financial)
Recommendation SystemFederated LearningSafty and PrivacyGraph Neural NetworkTabular
🎯 What it does: Propose a federated probabilistic preference distribution modeling framework (FPPDM) to share Gaussian distribution representations of users and items in privacy-preserving multi-domain recommendation scenarios, and further introduce compactness co-clustering (FPPDM++) to aggregate similar users.
FedET: A Communication-Efficient Federated Class-Incremental Learning Framework Based on Enhanced Transformer
Chenghao Liu (Ping An Technology (Shenzhen) Co., Ltd.), Jing Xiao (Ping An Technology (Shenzhen) Co., Ltd.)
ClassificationFederated LearningComputational EfficiencyKnowledge DistillationRepresentation LearningTransformerImageText
🎯 What it does: Propose the FedET framework, which inserts an Enhancer module into a frozen Transformer backbone, updating only a minimal number of parameters to achieve both high accuracy and low communication cost in federated continual learning (FCIL).
FedHGN: A Federated Framework for Heterogeneous Graph Neural Networks
Xinyu Fu (Chinese University of Hong Kong), Irwin King (Chinese University of Hong Kong)
Federated LearningGraph Neural NetworkGraph
🎯 What it does: Propose the FedHGN framework to enable collaborative training of heterogeneous graph neural networks in a federated environment.
FedNoRo: Towards Noise-Robust Federated Learning by Addressing Class Imbalance and Label Noise Heterogeneity
Nannan Wu (Huazhong University of Science and Technology), Zengqiang Yan (Huazhong University of Science and Technology)
Federated LearningKnowledge DistillationImageBiomedical DataComputed Tomography
🎯 What it does: This paper proposes the FedNoRo two-phase federated learning framework to address class imbalance and label noise heterogeneity in medical image data.
FedOBD: Opportunistic Block Dropout for Efficiently Training Large-scale Neural Networks through Federated Learning
Yuanyuan Chen (Nanyang Technological University), Han Yu (Nanyang Technological University)
ClassificationFederated LearningComputational EfficiencyImageText
🎯 What it does: Propose FEDOBD, which utilizes semantic block-level importance for adaptive block dropping combined with NNADQ quantization to efficiently train large neural networks in federated learning.
FedPass: Privacy-Preserving Vertical Federated Deep Learning with Adaptive Obfuscation
Hanlin Gu (Webank), Qiang Yang (Webank)
Federated LearningSafty and PrivacyImagePoint Cloud
🎯 What it does: Proposes the FedPass framework, which protects features and labels in vertical federated learning through adaptive confusion.
FedSampling: A Better Sampling Strategy for Federated Learning
Tao Qi (Tsinghua University), Xing Xie (Microsoft Research Asia)
Federated LearningImageText
🎯 What it does: Propose FedSampling, which performs uniform sampling of each local sample in federated learning and estimates the global total sample count through a local differential privacy mechanism, thereby achieving data utilization at the data level comparable to centralized training.
Few-shot Classification via Ensemble Learning with Multi-Order Statistics
Sai Yang (Nantong University), Jun Zhou (Griffith University)
ClassificationMeta LearningConvolutional Neural NetworkContrastive LearningImage
🎯 What it does: For few-shot classification, this paper proposes an ensemble learning framework called EL-MOS based on multi-order statistics. During the pre-training phase, multiple branches are added after the same backbone, and diverse sub-models are generated by calculating first-order, second-order, and third-order statistics separately. During the few-shot evaluation phase, the features from each branch are concatenated for classification.
FGNet: Towards Filling the Intra-class and Inter-class Gaps for Few-shot Segmentation
Yuxuan Zhang (University of Science and Technology of China), Shaowei Wang (Guangzhou University)
SegmentationConvolutional Neural NetworkImage
🎯 What it does: FGNet addresses the intra-class and inter-class gaps in few-shot segmentation through adaptive modules and inter-class feature separation modules, thereby improving the quality of mask prediction.
Finding an ϵ-Close Minimal Variation of Parameters in Bayesian Networks
Bahare Salmani (RWTH Aachen University), Joost-Pieter Katoen (RWTH Aachen University)
OptimizationHyperparameter SearchGraph
🎯 What it does: Propose an ε-close minimum change parameter tuning algorithm based on regional verification for multi-parameter, multi-CPT Bayesian networks (pBN), minimizing parameter changes while satisfying given quantitative constraints.
Finding Mixed-Strategy Equilibria of Continuous-Action Games without Gradients Using Randomized Policy Networks
Carlos Martin (Carnegie Mellon University), Tuomas Sandholm (Carnegie Mellon University)
OptimizationReinforcement LearningBenchmark
🎯 What it does: Studied a method that combines randomized policy networks with zeroth-order optimization to solve approximate mixed Nash equilibria in continuous action games without gradient information.
Fine-tuned vs. Prompt-tuned Supervised Representations: Which Better Account for Brain Language Representations?
Jingyuan Sun (KU Leuven), Marie-Francine Moens (KU Leuven)
Representation LearningTransformerSupervised Fine-TuningPrompt EngineeringTextBiomedical DataMagnetic Resonance Imaging
🎯 What it does: This paper generates supervised sentence representations by fully fine-tuning BERT and performing Prompt-tuning on 10 NLU tasks, then maps them to fMRI brain activation data using a regression neural decoding method to evaluate their decoding performance in the brain language network.
First-Choice Maximality Meets Ex-ante and Ex-post Fairness
Xiaoxi Guo (Peking University), Hanpin Wang (Guangzhou University)
Optimization
🎯 What it does: Designed two random mechanisms (GEBM and GPBM) to achieve first-choice maximization (FCM), Pareto efficiency (PE), and combinations of ex-post fairness (EF1) with ex-ante and ex-post fairness (sd-WEF or sd-E) in the assignment problem of allocating indivisible items.
Flaws of Termination and Optimality in ADOPT-based Algorithms
Koji Noshiro (University of Tsukuba), Koji Hasebe (University of Tsukuba)
Optimization
🎯 What it does: Systematic error analysis of ADOPT and its seven mainstream variants regarding termination and optimality is conducted, along with an improved version of ADOPT that eliminates these defects and proves its termination and optimality.
Fluid Dynamics-Inspired Network for Infrared Small Target Detection
Tianxiang Chen (University of Science and Technology of China), Nenghai Yu (University of Science and Technology of China)
Object DetectionConvolutional Neural NetworkTransformerImagePhysics Related
🎯 What it does: Propose an infrared small target detection network called FDI-Net based on fluid dynamics, simulating pixel motion as fluid flow.
Formal Explanations of Neural Network Policies for Planning
Renee Selvey (Australian National University), Sylvie Thiébaux (University of Toulouse)
OptimizationExplainability and InterpretabilitySequentialBenchmark
🎯 What it does: Studied how to generate formal, interpretable decision sequence explanations for deep learning-based planning strategies, proposing a linearly decomposable algorithm.
From Association to Generation: Text-only Captioning by Unsupervised Cross-modal Mapping
Junyang Wang (Beijing Jiaotong University), Jitao Sang (Beijing Jiaotong University)
GenerationTransformerLarge Language ModelVision Language ModelContrastive LearningImageVideoTextRetrieval-Augmented Generation
🎯 What it does: Propose a fully unsupervised text generation method for image/video captioning called Knight, which maps input images/videos to the k most similar text captions through CLIP's cross-modal mapping, and then generates captions using GPT-2 decoding;
From Generation to Suppression: Towards Effective Irregular Glow Removal for Nighttime Visibility Enhancement
Wanyu Wu (Wuhan University of Science and Technology), Xin Xu (Wuhan University of Science and Technology)
RestorationConvolutional Neural NetworkImagePhysics Related
🎯 What it does: A physics-based model for suppressing night halos was studied, utilizing APSF to estimate multiply scattered light and employing a light perception blind deconvolution network (LBDN) with a Retinex brightness enhancement module (REM) to remove irregular halos and enhance low-light images.
Front-to-End Bidirectional Heuristic Search with Consistent Heuristics: Enumerating and Evaluating Algorithms and Bounds
Lior Siag (Ben Gurion University of Negev), Nathan Sturtevant (University of Alberta)
Optimization
🎯 What it does: This paper studies the unified perspective of MEP theory and search bounds under bidirectional search with consistency heuristics, deriving 17 new search bounds from this perspective; based on these bounds, multiple bidirectional search algorithms targeting individual bounds are proposed and implemented; experiments validate the effectiveness of these bounds and the performance of different algorithms across various problem domains.
G2Pxy: Generative Open-Set Node Classification on Graphs with Proxy Unknowns
Qin Zhang (Shenzhen University), Shirui Pan (Griffith University)
ClassificationGraph Neural NetworkGraph
🎯 What it does: Proposes a generative open-set node classification method called G2Pxy, which achieves identification of unknown classes and classification of known classes by generating proxy unknown nodes in graph neural networks.
Game Theory with Simulation of Other Players
Vojtěch Kovařík (Carnegie Mellon University), Vincent Conitzer (Carnegie Mellon University)
🎯 What it does: This paper studies scenarios in two-player games where one party can simulate the other at a cost, and systematically analyzes the Nash equilibrium structure and utility impact of this 'simulation game'.
Gapformer: Graph Transformer with Graph Pooling for Node Classification
Chuang Liu (Wuhan University), Wenbin Hu (Wuhan University)
ClassificationTransformerGraph
🎯 What it does: Propose Gapformer, a node classification model that combines graph Transformer with graph pooling, reducing computational complexity and irrelevant information by first pooling the graph before computing attention.
GeNAS: Neural Architecture Search with Better Generalization
Joonhyun Jeong (NAVER Cloud, ImageVision), YoungJoon Yoo (NAVER Cloud, ImageVision)
Neural Architecture SearchImageBenchmark
🎯 What it does: This paper proposes using the flatness of the local loss surface as a proxy metric for NAS search, constructing the GeNAS framework;
Generalization Bounds for Adversarial Metric Learning
Wen Wen (Huazhong Agricultural University), Liangxuan Zhu (Huazhong Agricultural University)
Representation LearningImageTabular
🎯 What it does: Studied the generalization properties of adversarial metric learning, provided high-probability generalization upper bounds for adversarial pair learning, and derived fast convergence rates under smooth loss.
Generalization Guarantees of Self-Training of Halfspaces under Label Noise Corruption
Lies Hadjadj (Universit e Grenoble Alpes), Sana Louhichi (Universit e Grenoble Alpes)
ClassificationImageText
🎯 What it does: Proposed a self-training algorithm that iteratively learns from labeled and unlabeled data using half-spaces (linear threshold functions), combining two steps: exploration (searching for half-spaces with large cosine distances and pseudo-labeling) and pruning (removing low-confidence samples), ultimately obtaining a series of half-space predictors.
Generalization through Diversity: Improving Unsupervised Environment Design
Wenjun Li (Singapore Management University), Dexun Li (Singapore Management University)
Data SynthesisReinforcement Learning
🎯 What it does: Proposed the DIPLR method, which uses a diversity metric (occupancy distribution differences based on Wasserstein distance) to guide unsupervised environment design, enabling more efficient exploration of diverse and challenging environments during training.
Generalized Discriminative Deep Non-Negative Matrix Factorization Based on Latent Feature and Basis Learning
Zijian Yang (ShanghaiTech University), Lu Sun (ShanghaiTech University)
ClassificationRepresentation LearningImage
🎯 What it does: Proposed a Generalized Deep Non-Negative Matrix Factorization (GDNMF) that simultaneously performs deep decomposition on features and bases, integrates shallow linear and deep nonlinear models, and implements a semi-supervised version.
Generative Flow Networks for Precise Reward-Oriented Active Learning on Graphs
Yinchuan Li (Huawei Noah's Ark Lab), Jianye Hao (Huawei Noah's Ark Lab)
Graph Neural NetworkFlow-based ModelGraph
🎯 What it does: Proposes GFlowGNN, transforming the active learning problem on graphs into a generative process, and using generative flow networks to progressively select a set of labeled nodes.
Genetic Prompt Search via Exploiting Language Model Probabilities
Jiangjiang Zhao (Beijing University of Posts and Telecommunications), Fangchun Yang (Beijing University of Posts and Telecommunications)
OptimizationTransformerLarge Language ModelPrompt EngineeringText
🎯 What it does: This paper proposes a genetic algorithm called GAP3, guided by language model probabilities, for automatically searching discrete prompts on black-box pre-trained language models.
GIDnets: Generative Neural Networks for Solving Inverse Design Problems via Latent Space Exploration
Carlo Adornetto (University of Calabria), Gianluigi Greco (University of Calabria)
GenerationOptimizationConvolutional Neural NetworkFlow-based ModelAuto EncoderImageTabularPhysics Related
🎯 What it does: Propose GIDNET, a generative inverse design network, which utilizes latent space exploration to address inverse design problems; an autoencoder is employed to ensure the feasibility of solutions, while seed selection and conditional generators are used for guided search.
Globally Consistent Federated Graph Autoencoder for Non-IID Graphs
Kun Guo (Fuzhou University), Wenzhong Guo (Fuzhou University)
Federated LearningRepresentation LearningGraph Neural NetworkAuto EncoderGraph
🎯 What it does: Proposes GCFGAE, a graph autoencoder combining federated learning and split learning for unsupervised federated graph learning, addressing accuracy degradation caused by non-IID graphs.
GLPocket: A Multi-Scale Representation Learning Approach for Protein Binding Site Prediction
Peiying Li (Shanghai Jiao Tong University), Lei Xu (Shanghai Jiao Tong University)
Representation LearningDrug DiscoveryConvolutional Neural NetworkTransformerBiomedical Data
🎯 What it does: Designed and implemented a multi-scale network called GLPocket based on Lmser for precise prediction of protein binding sites;
GPLight: Grouped Multi-agent Reinforcement Learning for Large-scale Traffic Signal Control
Yilin Liu (Beijing University of Posts and Telecommunications), Rui Pan (Beijing University of Posts and Telecommunications)
OptimizationGraph Neural NetworkReinforcement LearningGraph
🎯 What it does: Designed GPLight, a dynamic clustering-based multi-agent reinforcement learning framework for large-scale traffic signal control;
GPMO: Gradient Perturbation-Based Contrastive Learning for Molecule Optimization
Xixi Yang (Hunan University), Xiangxiang Zeng (Hunan University)
OptimizationDrug DiscoveryTransformerContrastive LearningGraph
🎯 What it does: Propose a contrastive learning framework called GPMO based on gradient perturbation to address the exposure bias issue in translation models for molecular optimization.
Graph Neural Convection-Diffusion with Heterophily
Kai Zhao (Nanyang Technological University), Wee Peng Tay (Nanyang Technological University)
ClassificationGraph Neural NetworkDiffusion modelGraphOrdinary Differential Equation
🎯 What it does: Proposed a Graph Neural Convection-Diffusion (GNC-D) model based on the convection-diffusion equation (CDE), which enhances traditional diffusion models by introducing a learnable convection term to better model information flow in heterophilic graphs.
Graph Propagation Transformer for Graph Representation Learning
Zhe Chen (Nanjing University), Yue Qi (OPPO Research Institute)
Representation LearningGraph Neural NetworkTransformerGraph
🎯 What it does: Propose Graph Propagation Transformer (GPTrans), which explicitly realizes three types of information propagation between nodes and edges (node-to-node, node-to-edge, edge-to-node) within Transformer modules through Graph Propagation Attention (GPA), and constructs an efficient framework for graph representation learning.
Graph Sampling-based Meta-Learning for Molecular Property Prediction
Xiang Zhuang (Zhejiang University), Huajun Chen (Zhejiang University)
Meta LearningDrug DiscoveryGraph Neural NetworkContrastive LearningGraph
🎯 What it does: Proposes a meta-learning framework GS-Meta based on graph sampling, which reconstructs meta-learning tasks as subgraphs using a molecular-attribute relationship graph (MPG), and improves few-shot performance in molecular property prediction through contrastive learning-based subgraph sampling scheduling.
Graph-based Semi-supervised Local Clustering with Few Labeled Nodes
Zhaiming Shen, Sheng Li (University of Virginia)
Graph Neural NetworkGraph
🎯 What it does: Propose a semi-supervised local clustering method CS-LCE based on compressed sensing, which can extract target clusters using only a small number of seed nodes.
Group Fairness in Set Packing Problems
Sharmila Duppala (University of Maryland), Aravind Srinivasan (University of Maryland)
OptimizationTabularBiomedical Data
🎯 What it does: Proposed a fair k-set packing (FAIRSP) model for the kidney exchange (Kep) problem, and presented two randomized algorithms (FAIRSAMPLE and FAIRELIMINATE) to solve it.
GTR: A Grafting-Then-Reassembling Framework for Dynamic Scene Graph Generation
Jiafeng Liang (Harbin Institute of Technology), Bing Qin (Harbin Institute of Technology)
RecognitionObject DetectionGenerationConvolutional Neural NetworkTransformerVideo
🎯 What it does: Proposes the GTR framework, which first generates static relationships for each frame using a static scene graph generation model, and then reassembles them into dynamic scene graphs through a temporal dependency model.
Guide to Control: Offline Hierarchical Reinforcement Learning Using Subgoal Generation for Long-Horizon and Sparse-Reward Tasks
Wonchul Shin (Sungkyunkwan University), Yusung Kim (Sungkyunkwan University)
Robotic IntelligenceReinforcement LearningAuto EncoderTabularSequential
🎯 What it does: Proposed an offline hierarchical reinforcement learning framework called Guider, which generates reachable subgoals through a high-level policy and learns a low-level policy to achieve these subgoals, thereby solving long-horizon sparse reward tasks without online interaction.
Guided Patch-Grouping Wavelet Transformer with Spatial Congruence for Ultra-High Resolution Segmentation
Deyi Ji (University of Science and Technology of China), Hongtao Lu (Shanghai Jiao Tong University)
SegmentationConvolutional Neural NetworkTransformerImage
🎯 What it does: Propose a dual-branch network named GPWFormer for ultra-high-resolution image segmentation; the Transformer branch efficiently extracts local details and global long-range dependencies through a dynamically grouped Patch-Grouping Wavelet Transformer (WFormer); the CNN branch obtains deep class context via downsampling + wavelet transform, guiding Patch-Grouping and spatial consistency constraints.
Handling Learnwares Developed from Heterogeneous Feature Spaces without Auxiliary Data
Peng Tan (Nanjing University), Zhi-Hua Zhou (Nanjing University)
Domain AdaptationRecommendation SystemRepresentation LearningTabular
🎯 What it does: Construct a learnware market that can accommodate and reuse models from different feature spaces, achieving feature space alignment between models via the RKME specification without relying on any auxiliary data.
Hawkes Process Based on Controlled Differential Equations
Minju Jo (Yonsei University), Noseong Park (Yonsei University)
Time SeriesElectronic Health RecordsStochastic Differential EquationOrdinary Differential Equation
🎯 What it does: This paper proposes a Hawkes process model HP-CDE based on neural controlled differential equations (CDE) to continuously model irregular time series events and precisely compute the log-likelihood.
HDFormer: High-order Directed Transformer for 3D Human Pose Estimation
Hanyuan Chen (DAMO Academy, Alibaba Group), Xuansong Xie (DAMO Academy, Alibaba Group)
Pose EstimationConvolutional Neural NetworkTransformerGraphSequential
🎯 What it does: Propose a high-order directed Transformer (HDFormer), which achieves direct mapping from 2D keypoints to 3D pose by modeling high-order attention (bone↔joint, super-bone↔joint) in skeletal structures.
Helpful Information Sharing for Partially Informed Planning Agents
Sarah Keren (Technion Israel Institute of Technology), Sara Bernardini (Royal Holloway University of London)
OptimizationReinforcement LearningAgentic AIBenchmark
🎯 What it does: This paper proposes the Helpful Information Sharing (HIS) problem, where in a two-agent collaboration scenario, the auxiliary agent must provide the minimal critical information to the partially informed execution agent to ensure the goal can be achieved; two solution methods are presented: Lazy-BFS based on breadth-first search, and Tka translation, which compiles HIS into a single-agent classical planning problem solvable directly by existing classical planners;
Hierarchical Apprenticeship Learning for Disease Progression Modeling
Xi Yang (IBM Research), Min Chi (North Carolina State University)
Reinforcement Learning from Human FeedbackRecurrent Neural NetworkReinforcement LearningTabularBiomedical DataElectronic Health Records
🎯 What it does: This study integrates apprenticeship learning into disease progression modeling, constructing a hierarchical framework named THEMES to extract progression stages and intervention strategies, and validates its effectiveness on sepsis early prediction tasks.
Hierarchical Prompt Learning for Compositional Zero-Shot Recognition
Henan Wang (Xidian University), Cheng Deng (Xidian University)
ClassificationRecognitionPrompt EngineeringVision Language ModelContrastive LearningMultimodality
🎯 What it does: Proposed a hierarchical prompt learning method that learns an embedding space on a pre-trained vision-language model (CLIP) using three distinct prompts (state, object, combination), and fuses them during inference to achieve compositional zero-shot recognition.
Hierarchical Semantic Contrast for Weakly Supervised Semantic Segmentation
Yuanchen Wu (Shanghai University), Shaorong Xie (Shanghai University)
SegmentationConvolutional Neural NetworkContrastive LearningImage
🎯 What it does: Propose Hierarchical Semantic Contrast (HSC), which utilizes ROI, class, and pixel-level three-tier contrast in weakly supervised semantic segmentation, combined with cross-supervision and momentum prototype learning to improve CAM quality and final segmentation accuracy.
Hierarchical State Abstraction based on Structural Information Principles
Xianghua Zeng (Beihang University), Philip S. Yu (Didi Chuxing)
Computational EfficiencyRepresentation LearningReinforcement LearningAuto EncoderImage
🎯 What it does: Proposes the SISA framework, an unsupervised adaptive hierarchical state abstraction method based on structural information principles, which enhances RL decision-making efficiency in environments with rich observations.
Hierarchical Transformer for Scalable Graph Learning
Wenhao Zhu (Peking University), Liang Wang (Alibaba Group)
Computational EfficiencyRepresentation LearningGraph Neural NetworkTransformerGraph
🎯 What it does: Proposed a hierarchical Transformer model, HSGT, in large-scale graph learning, achieving efficient learning on graphs with millions of nodes.
HireVAE: An Online and Adaptive Factor Model Based on Hierarchical and Regime-Switch VAE
Zikai Wei (Chinese University of Hong Kong), Dahua Lin (Chinese University of Hong Kong)
Auto EncoderTabularTime SeriesFinance Related
🎯 What it does: Proposed HireVAE, an end-to-end online adaptive factor model that utilizes hierarchical latent spaces and online market state identification to achieve dynamic factor combination and real-time prediction of stock returns.
HOI-aware Adaptive Network for Weakly-supervised Action Segmentation
Runzhong Zhang (Nanyang Technological University), Yap-Peng Tan (Nanyang Technological University)
SegmentationRecurrent Neural NetworkTransformerVideo
🎯 What it does: Propose AdaAct, which leverages video-level human-object interaction (HOI) sequences as priors to adaptively adjust temporal encoder parameters for weakly supervised action segmentation.
HOUDINI: Escaping from Moderately Constrained Saddles
Dmitrii Avdiukhin (Indiana University), Grigory Yaroslavtsev (George Mason University)
Optimization
🎯 What it does: This paper proposes a polynomial-time stochastic gradient descent (SGD) and deterministic gradient descent (GD) algorithm that can escape saddle points of high-dimensional non-convex functions under linear inequality constraints with the number of constraints being logarithmic (k = O(log^d)) and converge to an approximate second-order feasible stationary point (δ-SOSP).
HyperFed: Hyperbolic Prototypes Exploration with Consistent Aggregation for Non-IID Data in Federated Learning
Xinting Liao (Zhejiang University), Yue Qi (OPPO Research Institute)
Federated LearningRepresentation LearningImage
🎯 What it does: The study proposes a three-module framework called HyperFed for non-IID federated learning, utilizing hyperbolic space prototypes for initialization, learning, and consistent aggregation.
Hyperspectral Image Denoising Using Uncertainty-Aware Adjustor
Jiahua Xiao (Xi'an Jiaotong University), Xing Wei (Xi'an Jiaotong University)
RestorationConvolutional Neural NetworkImage
🎯 What it does: This paper proposes a plug-and-play UA-Adjustor module to dynamically enhance the denoising performance of spectral images in existing spectral-aided denoising networks.
ICDA: Illumination-Coupled Domain Adaptation Framework for Unsupervised Nighttime Semantic Segmentation
Chenghao Dong (Beijing University of Posts and Telecommunications), Anlong Ming (Beijing University of Posts and Telecommunications)
SegmentationDomain AdaptationTransformerGenerative Adversarial NetworkContrastive LearningImage
🎯 What it does: This paper proposes a Lighting-Coupled Domain Adaptation framework (ICDA), which addresses the challenges of illumination differences and dataset discrepancies in nighttime semantic segmentation by constructing day-night image pairs and leveraging semantic relevance.
IID-GAN: an IID Sampling Perspective for Regularizing Mode Collapse
Yang Li (Shanghai Jiao Tong University), Junchi Yan (Shanghai Jiao Tong University)
GenerationGenerative Adversarial NetworkImage
🎯 What it does: Propose IID-GAN, which alleviates mode collapse by regularizing the sampling distribution of the inverse mapping to maintain generated samples as independent and identically distributed (IID) within the target distribution.
Image Composition with Depth Registration
Zan Li (State Key Laboratory of Computer Science, Institute of Software, Chinese Academy of Sciences), Fei Hou (State Key Laboratory of Computer Science, Institute of Software, Chinese Academy of Sciences)
Image HarmonizationSegmentationGenerationDepth EstimationConvolutional Neural NetworkImage
🎯 What it does: Propose a depth registration method that places the target object from the source image directly into the three-dimensional space represented by the target image, thereby automatically determining occlusion relationships through pixel-level depth comparison to achieve seamless image synthesis.
IMF: Integrating Matched Features Using Attentive Logit in Knowledge Distillation
Jeongho Kim (Korea Advanced Institute of Science and Technology), Simon S. Woo (Sungkyunkwan University)
ClassificationKnowledge DistillationConvolutional Neural NetworkImage
🎯 What it does: Proposed a flexibly configurable knowledge distillation framework named IMF, which utilizes an Intermediate Feature Distiller (IFD) and attention logit for distillation, directly training in student network branches and using only branch outputs during inference.
Improve Video Representation with Temporal Adversarial Augmentation
Jinhao Duan (Drexel University), Kaidi Xu (Drexel University)
Representation LearningGenerative Adversarial NetworkVideo
🎯 What it does: Propose Temporal Adversarial Augmentation (TA) and the TAF framework, which utilize temporal loss generated by CAM during video model fine-tuning for adversarial augmentation. This balances the model's attention distribution across different temporal segments of the video, thereby enhancing video representation and generalization capabilities.
Improved Algorithms for Allen's Interval Algebra by Dynamic Programming with Sublinear Partitioning
Leif Eriksson (Linkoping University), Victor Lagerkvist (Linkoping University)
OptimizationComputational Efficiency
🎯 What it does: Proposed a new dynamic programming framework (sublinear partitioning) and applied it to Allen's temporal interval algebra and Cardinal Direction Point Algebra, reducing their time complexities from O*((1.0615 n)^n) to O*((c n log n)^n) and O*((c n log n)^{2n/3}) respectively.
Improving Heterogeneous Model Reuse by Density Estimation
Anke Tang (Wuhan University), Dacheng Tao (University of Sydney)
Federated LearningSafty and PrivacyFlow-based ModelContrastive LearningImageBenchmark
🎯 What it does: Studied model reuse in federated learning using local models and local data density estimation, constructing a global model and proposing a cross-party cross-entropy calibration method;
Improving LaCAM for Scalable Eventually Optimal Multi-Agent Pathfinding
Keisuke Okumura (National Institute of Advanced Industrial Science and Technology)
OptimizationGraphBenchmark
🎯 What it does: Improved the LaCAM algorithm to become a multi-agent path planning algorithm (LaCAM*) that can provide suboptimal solutions within limited time and eventually converge to the optimal solution, and designed an improved PIBT configuration generator to enhance search efficiency.
In Which Graph Structures Can We Efficiently Find Temporally Disjoint Paths and Walks?
Pascal Kunz (Humboldt University of Berlin), Meirav Zehavi (Ben-Gurion University of Negev)
Computational EfficiencyGraph
🎯 What it does: Analyzes the parameterized complexity of non-crossing paths and walks in temporal graphs, and provides boundaries for hardware feasibility and solvability.
Incentive-Compatible Selection for One or Two Influentials
Yuxin Zhao (ShanghaiTech University), Dengji Zhao (ShanghaiTech University)
OptimizationGraph
🎯 What it does: Designed a mechanism in directed acyclic graphs (DAGs) to prevent agents from hiding edges, achieving near-optimal selection of one or two most influential agents;
Incentivizing Recourse through Auditing in Strategic Classification
Andrew Estornell (Washington University in Saint Louis), Yevgeniy Vorobeychik (Washington University in Saint Louis)
ClassificationTabular
🎯 What it does: This paper proposes an audit method to incentivize subjects to take real feasible recourse actions rather than merely manipulating features, and provides an optimal audit strategy;