arXivSub Start free trial

AAAI 2023 Papers with Code β€” Page 3

AAAI Conference on Artificial Intelligence Β· 696 papers

EMEF: Ensemble Multi-Exposure Image Fusion

Renshuai Liu (Xiamen University), Xuan Cheng (Xiamen University)

CodeImage TranslationOptimizationConvolutional Neural NetworkGenerative Adversarial NetworkImage

🎯 What it does: An integrated multi-exposure image fusion method EMEF is proposed, which mimics and optimizes the fusion results of various existing MEF methods to obtain a better fused image than a single method.

Emergence of Punishment in Social Dilemma with Environmental Feedback

Zhen Wang (Northwestern Polytechnical University), Shuyue Hu (Kyushu University)

Code

🎯 What it does: A third-party punishment public goods game evolutionary model incorporating environmental feedback is proposed and analyzed, studying the conditions and dynamics of the co-evolution of punishment and cooperation.

Enabling Knowledge Refinement upon New Concepts in Abductive Learning

Yu-Xuan Huang (Nanjing University), Zhi-Hua Zhou (Nanjing University)

CodeTabular

🎯 What it does: The ABL nc framework is proposed, which combines new concept detection, rule learning, and conflict resolution to achieve adaptive updates of the knowledge base, and enhances the performance of the perception model through inductive logic programming and reasoning.

End-to-End Deep Reinforcement Learning for Conversation Disentanglement

Karan Bhukar (IBM Research), Ajay Gupta (IBM Research)

CodeRecurrent Neural NetworkGraph Neural NetworkTransformerReinforcement LearningText

🎯 What it does: An end-to-end reinforcement learning framework is proposed to directly optimize the multi-party dialogue segmentation task using global metrics;

End-to-End Entity Linking with Hierarchical Reinforcement Learning

Lihan Chen (Fudan University), Yanghua Xiao (Fudan University)

CodeRecurrent Neural NetworkTransformerLarge Language ModelReinforcement LearningText

🎯 What it does: Modeling the entity linking task as a hierarchical reinforcement learning framework, first detecting mentions through high-level decisions, and then completing entity disambiguation with a low-level generative strategy;

End-to-End Learning for Optimization via Constraint-Enforcing Approximators

Rares Cristian (Massachusetts Institute of Technology), Ioannis Spantidakis (Massachusetts Institute of Technology)

CodeOptimization

🎯 What it does: A ProjectNet neural network architecture is designed to achieve end-to-end learning by approximating projections to solve linear optimization problems, embedding it into a prediction-optimization pipeline.

End-to-End Zero-Shot HOI Detection via Vision and Language Knowledge Distillation

Mingrui Wu (Xiamen University), Xiaoshuai Sun (Xiamen University)

CodeObject DetectionKnowledge DistillationTransformerVision Language ModelContrastive LearningImageMultimodality

🎯 What it does: This paper proposes an end-to-end zero-shot human-object interaction detection framework EoID, which utilizes CLIP's visual-language knowledge for distillation.

Energy-Motivated Equivariant Pretraining for 3D Molecular Graphs

Rui Jiao (Tsinghua University), Yang Liu (Tencent)

CodeRepresentation LearningDrug DiscoveryGraph Neural NetworkGraph

🎯 What it does: A novel energy-driven equivariant pre-training framework (3D-EMGP) is proposed, achieving unsupervised pre-training of 3D molecular structures through the construction of equivariant force prediction and layer noise scale classification tasks for 3D molecular graphs.

Entity-Agnostic Representation Learning for Parameter-Efficient Knowledge Graph Embedding

Mingyang Chen (Zhejiang University), Huajun Chen (University of Edinburgh)

CodeRepresentation LearningGraph Neural NetworkGraph

🎯 What it does: This paper proposes an entity-agnostic representation learning framework (EARL) that addresses the problem of linear growth in parameter count with the number of entities in knowledge graph embedding by encoding distinguishable information of entities (edge relationships, k-nearest reserved entities, multi-hop neighbors) instead of storing vectors for each entity individually.

Entropy Regularization for Population Estimation

Ben Chugg (Carnegie Mellon University), Daniel E. Ho (Stanford University)

CodeOptimizationReinforcement LearningTabular

🎯 What it does: A sampling strategy based on entropy regularization and KL divergence is proposed to simultaneously maximize rewards and accurately estimate the overall mean in the optimization-estimation structured multi-armed bandit problem.

ESPT: A Self-Supervised Episodic Spatial Pretext Task for Improving Few-Shot Learning

Yi Rong (Wuhan University of Technology), Shengwu Xiong (Wuhan University of Technology)

CodeClassificationMeta LearningConvolutional Neural NetworkContrastive LearningImage

🎯 What it does: This paper proposes a self-supervised episodic spatial pretext task (ESPT) that enhances few-shot learning performance in image classification by utilizing the consistency of local spatial relationships between original images and images transformed by random geometric transformations in few-shot episodes.

Estimating Average Causal Effects from Patient Trajectories

Dennis Frauen (LMU Munich), Stefan Feuerriegel (LMU Munich)

CodeRecurrent Neural NetworkTime SeriesBiomedical DataElectronic Health Records

🎯 What it does: An end-to-end deep learning model, DeepACE, was designed and implemented to estimate the average causal effect (ACE) over time from follow-up electronic medical records (temporal patient trajectories). This was achieved through joint learning of iterative G-computation and sequential targeting to obtain doubly robust and asymptotically efficient estimates.

Explaining Random Forests Using Bipolar Argumentation and Markov Networks

Nico Potyka (Imperial College London), Francesca Toni (Imperial College London)

CodeExplainability and InterpretabilityTabular

🎯 What it does: This paper proposes the use of a bipolar argumentation framework and Markov networks to explain random forests, constructing interpretable models.

Explicit Invariant Feature Induced Cross-Domain Crowd Counting

Yiqing Cai (East China Normal University), Gaoqi He (East China Normal University)

CodeDomain AdaptationGraph Neural NetworkImage

🎯 What it does: The IF-CKT framework is proposed for cross-domain crowd counting, explicitly separating and aligning domain-invariant and domain-specific features, and enhancing target domain adaptability through graph neural networks and pseudo-labeling.

Exploit Domain-Robust Optical Flow in Domain Adaptive Video Semantic Segmentation

Yuan Gao (University of Science and Technology of China), Junjie Li (University of Science and Technology of China)

CodeSegmentationDomain AdaptationOptical FlowVideo

🎯 What it does: This paper studies domain adaptive video semantic segmentation and proposes to improve the model using domain-robust optical flow through segmentation-flow consistency supervision.

Exploiting Multiple Abstractions in Episodic RL via Reward Shaping

Roberto Cipollone (Universita degli Studi di Roma La Sapienza), Fabio Patrizi (Banca d'Italia)

CodeReinforcement Learning

🎯 What it does: This paper proposes a reward shaping method using multi-layer abstract MDPs, which significantly improves sample efficiency by guiding finer-grained RL learning through the optimal value function learned at the abstract level.

Exploring Non-target Knowledge for Improving Ensemble Universal Adversarial Attacks

Juanjuan Weng (Xiamen University), Shaozi Li (Xiamen University)

CodeKnowledge DistillationAdversarial AttackConvolutional Neural NetworkImage

🎯 What it does: A min-max training framework is proposed that only considers KL loss of non-target classes and adaptive weights, enhancing the transferability of Universal Adversarial Perturbations (UAP).

Exploring Self-Distillation Based Relational Reasoning Training for Document-Level Relation Extraction

Liang Zhang (Xiamen University), Yidong Chen (Xiamen University)

CodeKnowledge DistillationRepresentation LearningTransformerSupervised Fine-TuningTextBiomedical Data

🎯 What it does: In document-level relation extraction, a reasoning module that combines four reasoning patterns and a self-distillation training framework is proposed to explicitly model the relationship reasoning process.

Exploring Stochastic Autoregressive Image Modeling for Visual Representation

Yu Qi (Tsinghua University), Wei Li (SenseTime Research)

CodeObject DetectionSegmentationRepresentation LearningTransformerImage

🎯 What it does: This paper proposes a self-regressive image modeling method based on random sequences, called SAIM, which achieves visual self-supervised pre-training by randomly permuting patch sequences and using a parallel encoder-decoder architecture.

Exploring Stroke-Level Modifications for Scene Text Editing

Yadong Qu (University of Science and Technology of China), Yongdong Zhang (University of Science and Technology of China)

CodeImage TranslationDomain AdaptationGenerative Adversarial NetworkImage

🎯 What it does: This paper proposes the MOSTEL framework, which can replace and edit scene text while keeping the background texture unchanged, and trains using both synthetic pairs and unlabelled real images through semi-supervised mixed learning.

Exploring Temporal Information Dynamics in Spiking Neural Networks

Youngeun Kim (Yale University), Priyadarshini Panda (Technology Innovation Institute)

CodeClassificationOptimizationComputational EfficiencySpiking Neural NetworkImage

🎯 What it does: This paper analyzes the phenomenon of Temporal Information Concentration in SNN training by measuring the dynamic Fisher information of temporal information.

Exploring the Interaction between Local and Global Latent Configurations for Clustering Single-Cell RNA-Seq: A Unified Perspective

Nairouz Mrabah (University of Quebec at Montreal), Abdoulaye Banire Diallo (University of Quebec at Montreal)

CodeClassificationRepresentation LearningGraph Neural NetworkAuto EncoderBiomedical Data

🎯 What it does: A single-cell RNA-seq clustering method called scTPF based on graph autoencoders is proposed, which gradually adjusts pseudo-supervised and self-supervised tasks by utilizing the interaction between local and global latent spaces to achieve high-quality cell clustering.

Exposing the Self-Supervised Space-Time Correspondence Learning via Graph Kernels

Zheyun Qin (Shandong University), Jianbing Shen (University of Macau)

CodeRecognitionSegmentationGraph Neural NetworkContrastive LearningVideo

🎯 What it does: Proposes the VideoHiGraph framework, which utilizes self-supervised graph kernel learning to generate hidden graphs and perform subgraph matching and node-level temporal consistency by capturing the spatial-temporal correspondence in videos.

Extracting Low-/High- Frequency Knowledge from Graph Neural Networks and Injecting It into MLPs: An Effective GNN-to-MLP Distillation Framework

Lirong Wu (Westlake University), Stan Z. Li (Zhejiang University)

CodeKnowledge DistillationGraph Neural NetworkGraph

🎯 What it does: This paper studies GNN-to-MLP knowledge distillation, proposing to separate the low-frequency and high-frequency knowledge learned by the pre-trained GNN and inject them into the MLP respectively to enhance model performance.

Factual and Informative Review Generation for Explainable Recommendation

Zhouhang Xie (University of California), Bodhisattwa Prasad Majumder (University of California)

CodeRecommendation SystemExplainability and InterpretabilityTransformerTextRetrieval-Augmented Generation

🎯 What it does: The PRAG model is proposed, which combines a personalized retriever with a question-answering reader, utilizing historical reviews to generate factual and diverse recommendation explanations.

Fairness and Explainability: Bridging the Gap towards Fair Model Explanations

Yuying Zhao (Vanderbilt University), Tyler Derr (Vanderbilt University)

CodeOptimizationExplainability and InterpretabilityGraph Neural NetworkGraphTabular

🎯 What it does: This paper proposes a program-oriented fairness metric based on explanation quality and designs a Comprehensive Fairness Algorithm (CFA) that takes into account prediction accuracy, traditional fairness, and explanation fairness.

FanoutNet: A Neuralized PCB Fanout Automation Method Using Deep Reinforcement Learning

Haiyun Li (Wuhan University of Technology), Mingyu Liu (Huawei Device Co., Ltd.)

CodeOptimizationConvolutional Neural NetworkTransformerReinforcement LearningGraph

🎯 What it does: This paper presents FanoutNet, an automated PCB fanout method based on deep reinforcement learning, which significantly improves the routability of PCBs by pre-allocating layers and routing resources.

Farsighted Probabilistic Sampling: A General Strategy for Boosting Local Search MaxSAT Solvers

Jiongzhi Zheng (Huazhong University of Science and Technology), Jianrong Zhou (Huazhong University of Science and Technology)

CodeOptimizationBenchmark

🎯 What it does: A general Future Probability Sampling (FPS) strategy is proposed to replace the single-variable flip mechanism in MaxSAT local search, thereby improving the performance of (W)PMS solvers.

Fast and Accurate Binary Neural Networks Based on Depth-Width Reshaping

Ping Xue (Hefei University of Technology), Zhen Wei (Hefei University of Technology)

CodeOptimizationNeural Architecture SearchConvolutional Neural NetworkImage

🎯 What it does: This paper proposes the Depth-Width Reshaping (DWR) method, which adjusts the depth and width of existing full-precision network backbones and combines pruning techniques to construct a backbone network more suitable for binary neural networks.

Fast Converging Anytime Model Counting

Yong Lai (Jilin University), Roland H.C. Yap

CodeBenchmark

🎯 What it does: A new approximate model counting algorithm called PartialKC is proposed, which can be terminated midway and converge to exact counts. It utilizes partial knowledge compilation (partial CCDD) to generate random partial CCDD for unbiased estimation.

Fast Offline Policy Optimization for Large Scale Recommendation

Otmane Sakhi (Criteo AI Lab), Alexandre Gilotte (Criteo AI Lab)

CodeRecommendation SystemOptimizationReinforcement LearningTabular

🎯 What it does: In the offline context of the multi-armed bandit framework, self-normalized importance sampling and approximate maximum inner product search (MIPS) are utilized to achieve offline policy optimization for large-scale recommendation systems, significantly reducing dependence on the size of the product catalog.

Fast Online Hashing with Multi-Label Projection

Wenzhe Jia (Ocean University of China), Jie Gui (Southeast University)

CodeRetrievalOptimizationComputational EfficiencyImage

🎯 What it does: A Fast Online Hashing (FOH) method is proposed, which constructs a query pool, employs a neighbor maintenance algorithm, and utilizes layer sampling. By leveraging a multi-label similarity matrix and label projection loss, it achieves online hashing updates that only modify a small number of database entries, significantly reducing query time.

Fast Saturating Gate for Learning Long Time Scales with Recurrent Neural Networks

Kentaro Ohno (NTT), Yasutoshi Ida (NTT)

CodeRecurrent Neural NetworkSequential

🎯 What it does: A new fast gate function is proposed, which achieves faster saturation by embedding the hyperbolic sine function into the sigmoid, thereby enhancing the ability of RNNs to learn extremely long time scales.

FastAMI – a Monte Carlo Approach to the Adjustment for Chance in Clustering Comparison Metrics

Kai Klede (Friedrich-Alexander Universitat Erlangen-Nrnberg), BjΓΆrn Eskofier (Friedrich-Alexander Universitat Erlangen-Nrnberg)

CodeOptimizationComputational EfficiencyBenchmark

🎯 What it does: This paper proposes FastAMI, an algorithm based on Monte Carlo sampling for the rapid approximation of Adjusted Mutual Information (AMI) and Standardized Mutual Information (SMI).

FASTDIAGP: An Algorithm for Parallelized Direct Diagnosis

Viet-Man Le (Graz University of Technology), Thi Ngoc Trang Tran (University of Sevilla)

CodeOptimizationComputational EfficiencyTabular

🎯 What it does: This paper presents FASTDIAGPβ€”a diagnostic method that parallelizes the traditional FASTDIAG direct diagnosis algorithm, significantly improving execution speed through speculative programming and early parallel computation of consistency checks.

Feature-Level Debiased Natural Language Understanding

Yougang Lyu (Shandong University), Zhaochun Ren (Shandong University)

CodeClassificationContrastive LearningText

🎯 What it does: A debiasing method based on contrastive learning (DCT) is proposed, which reduces the model's dependence on biased features by dynamically sampling positive and negative biased samples.

FedALA: Adaptive Local Aggregation for Personalized Federated Learning

Jianqing Zhang (Shanghai Jiao Tong University), Haibing Guan (Shanghai Jiao Tong University)

CodeFederated LearningConvolutional Neural NetworkImageText

🎯 What it does: The FedALA method is proposed, which enhances local training performance in personalized federated learning by initializing the local model through an Adaptive Local Aggregation (ALA) module that integrates the global model and local model using element-level weighting.

Federated Learning on Non-IID Graphs via Structural Knowledge Sharing

Yue Tan (Australian Artificial Intelligence Institute), Chengqi Zhang (Australian Artificial Intelligence Institute)

CodeFederated LearningGraph Neural NetworkGraph

🎯 What it does: A federated graph learning framework named FedStar is proposed, specifically designed to enhance model performance on non-IID graph data by sharing structural knowledge.

Federated Robustness Propagation: Sharing Adversarial Robustness in Heterogeneous Federated Learning

Junyuan Hong (Michigan State University), Jiayu Zhou (University of Texas at Austin)

CodeFederated LearningAdversarial AttackImage

🎯 What it does: This study investigates how to achieve the propagation of adversarial robustness in a heterogeneous federated learning environment through batch normalization techniques, proposing the FedRBN algorithm.

FedGS: Federated Graph-Based Sampling with Arbitrary Client Availability

Zheng Wang (Xiamen University), Cheng Wang (Xiamen University)

CodeFederated LearningGraph Neural NetworkGraph

🎯 What it does: This paper proposes the FEDGS (Federated Graph-based Sampling) framework, which achieves stable global model updates and reduces long-term bias by constructing a data distribution-dependent graph (3DG) under arbitrary client availability.

FEditNet: Few-Shot Editing of Latent Semantics in GAN Spaces

Mengfei Xia (Tsinghua University), Yong-Jin Liu (Tsinghua University)

CodeGenerationData SynthesisGenerative Adversarial NetworkContrastive LearningImage

🎯 What it does: Learn interpretable latent semantics and achieve attribute editing using a small number of labeled samples in a pre-trained GAN.

FedMDFG: Federated Learning with Multi-Gradient Descent and Fair Guidance

Zibin Pan (Chinese University of Hong Kong), Junhua Zhao (Chinese University of Hong Kong)

CodeOptimizationFederated LearningImage

🎯 What it does: This paper proposes FedMDFG, which utilizes multi-gradient descent and fairness guidance to simultaneously solve for a fair descent direction and an appropriate step size in federated learning, thereby enhancing the fairness and convergence speed of the global model.

FedNP: Towards Non-IID Federated Learning via Federated Neural Propagation

Xueyang Wu (Hong Kong University of Science and Technology), Qian Xu (Hong Kong University of Science and Technology)

CodeFederated LearningImageAudio

🎯 What it does: The FedNP method is proposed, which incorporates a neural propagation task into local training in federated learning to explicitly estimate the global data distribution, thereby alleviating the performance degradation caused by non-IID data.

FeedFormer: Revisiting Transformer Decoder for Efficient Semantic Segmentation

Jae-hun Shim (Sogang University), Suk-Ju Kang (Sogang University)

CodeSegmentationTransformerImage

🎯 What it does: This paper proposes FeedFormer, a model that enhances high-level structural information and completes semantic segmentation by utilizing a Transformer decoder, using low-level features as keys/values and high-level features as queries.

Few-Shot 3D Point Cloud Semantic Segmentation via Stratified Class-Specific Attention Based Transformer Network

Canyu Zhang (University of South Carolina), Song Wang (University of South Carolina)

CodeSegmentationTransformerPoint Cloud

🎯 What it does: A hierarchical class-specific attention Transformer network is proposed for few-shot 3D point cloud semantic segmentation, directly utilizing multi-scale point cloud features without relying on pooling or graph construction, improving matching accuracy and inference speed.

Few-Shot Defect Image Generation via Defect-Aware Feature Manipulation

Yuxuan Duan (Shanghai Jiao Tong University), Liqing Zhang (Shanghai Jiao Tong University)

CodeGenerationData SynthesisAnomaly DetectionGenerative Adversarial NetworkImage

🎯 What it does: This paper proposes a few-shot defect image generation method called DFMGAN, which addresses the scarcity of industrial defect images. It first trains StyleGAN2 on defect-free images as a base, and then applies defect-aware residual blocks to manipulate features in specific defect areas, generating realistic and diverse defect images while simultaneously producing corresponding defect masks, achieving automated augmentation of defect images.

FinalMLP: An Enhanced Two-Stream MLP Model for CTR Prediction

Kelong Mao (Renmin University of China), Zhenhua Dong (Tsinghua University)

CodeRecommendation SystemTabularBenchmark

🎯 What it does: A two-stream MLP structure called FinalMLP is proposed, with the addition of stream-specific feature selection and multi-head bilinear aggregation layers to enhance CTR prediction performance.

Finding Good Partial Assignments during Restart-Based Branch and Bound Search

Hongbo Li (Northeast Normal University), Jimmy H.M. Lee (Chinese University of Hong Kong)

CodeOptimizationTabularBenchmark

🎯 What it does: This paper proposes an algorithm that dynamically generates and utilizes good partial assignments in restart-based branch-and-bound search to accelerate the solution of general constraint optimization problems.

Fine-Grained Position Helps Memorizing More, a Novel Music Compound Transformer Model with Feature Interaction Fusion

Zuchao Li (Wuhan University), Kehua Su (Wuhan University)

CodeClassificationRecognitionTransformerSequentialAudio

🎯 What it does: In the music notation understanding task, the authors proposed the Feature Interaction Fusion (FiF) module and the Rotational Absolute-Relative Position Encoding (RoAR), improving the Composite Word Transformer (CP+Transformer) model to better capture the interrelationships among multiple attributes in music events and obtain finer-grained positional information.

FiTs: Fine-Grained Two-Stage Training for Knowledge-Aware Question Answering

Qichen Ye (Peking University), Yuexian Zou (Peking University)

CodeRetrievalKnowledge DistillationGraph Neural NetworkTransformerContrastive LearningText

🎯 What it does: A Fine-grained Two-stage Training (FiTs) framework is proposed for Knowledge-aware Question Answering (KAQA), which first aligns the PLM and KG representations through knowledge adaptation and then fine-tunes the model reasoning through self-supervised objectives (KSD, KBR).

Fixed-Weight Difference Target Propagation

Tatsukichi Shibuya (Tokyo Institute of Technology), Ikuro Sato (Denso IT Laboratory)

CodeClassificationConvolutional Neural NetworkImage

🎯 What it does: A differential target propagation algorithm with fixed feedback weights is proposed, eliminating the need for feedback network training in traditional target propagation.

Flexible 3D Lane Detection by Hierarchical Shape Matching

Zhihao Guan (Xi'an Jiaotong University), Shuqi Mei (Tencent)

CodeObject DetectionAutonomous DrivingConvolutional Neural NetworkTransformerPoint Cloud

🎯 What it does: A hierarchical end-to-end 3D lane detection framework (FHLD) has been designed and implemented, capable of simultaneously predicting global parameter curves and local segment shapes on the bird's-eye view (BEV) of point clouds, thereby outputting a flexible and accurate set of lane points.

Flora: Dual-Frequency LOss-Compensated ReAl-Time Monocular 3D Video Reconstruction

Likang Wang (Hong Kong University of Science and Technology), Lei Chen (Hong Kong University of Science and Technology)

CodeSegmentationData SynthesisDepth EstimationRecurrent Neural NetworkSimultaneous Localization and MappingVideo

🎯 What it does: A real-time monocular 3D video reconstruction method called Flora is proposed, which achieves high-quality and complete 3D reconstruction through dual-frequency distortion compensation aggregation and loss compensation.

Flow-Based Robust Watermarking with Invertible Noise Layer for Black-Box Distortions

Han Fang (National University of Singapore), Ee-Chien Chang (National University of Singapore)

CodeRestorationFlow-based ModelImage

🎯 What it does: A digital watermarking framework based on reversible networks has been designed and implemented, integrating reversible noise layers to achieve high robustness against both white-box and black-box distortions.

fmLRE: A Low-Resource Relation Extraction Model Based on Feature Mapping Similarity Calculation

Peng Wang (Southeast University), Wenjun Ke (Southeast University)

CodeReinforcement LearningText

🎯 What it does: In the low-resource relation extraction task, fmLRE is proposed, which generates pseudo-labels through self-training and calculates the similarity between pseudo-labels and true labels in the feature mapping space, using reinforcement learning iterative feedback to filter high-precision pseudo-labels, thereby reducing the bias in pseudo-label selection.

FoPro: Few-Shot Guided Robust Webly-Supervised Prototypical Learning

Yulei Qin (Tencent), Chunhua Shen (Shanghai Jiao Tong University)

CodeClassificationDomain AdaptationContrastive LearningImage

🎯 What it does: This paper proposes a prototype learning framework called FoPro, guided by a small number of real samples, to jointly train on noisy images crawled from the web and a limited amount of real labeled data, enhancing classification performance in real-world domains.

Forecasting with Sparse but Informative Variables: A Case Study in Predicting Blood Glucose

Harry Rubin-Falcone (University of Michigan), Jenna Wiens (University of Michigan)

CodeRecurrent Neural NetworkTime SeriesBiomedical Data

🎯 What it does: A link encoder/decoder architecture is proposed to handle sparse but informative auxiliary variables (SIV) in time series tasks such as blood glucose prediction.

Fourier-Net: Fast Image Registration with Band-Limited Deformation

Xi Jia (University of Birmingham), Jinming Duan (Alan Turing Institute)

CodeSegmentationOptimizationComputational EfficiencyConvolutional Neural NetworkOptical FlowImageBiomedical DataMagnetic Resonance Imaging

🎯 What it does: An unsupervised Fourier-Net is proposed, which represents the displacement/velocity field in a low-dimensional frequency domain (band-limited Fourier), trains a simplified U-Net structure, and decodes the complete displacement field through zero-padding + inverse discrete Fourier transform (iDFT), achieving fast medical image registration.

From Coarse to Fine: Hierarchical Pixel Integration for Lightweight Image Super-resolution

Jie Liu (Nanjing University), Gangshan Wu (Nanjing University)

CodeRestorationSuper ResolutionTransformerImage

🎯 What it does: A lightweight single-image super-resolution network HPINet has been designed and implemented, capable of achieving high-quality super-resolution results while maintaining a low parameter count.

FSR: A General Frequency-Oriented Framework to Accelerate Image Super-resolution Networks

Jinmin Li (Tsinghua University), Shu-Tao Xia (Tsinghua University)

CodeRestorationSuper ResolutionImage

🎯 What it does: This paper proposes a framework for a general frequency domain accelerated super-resolution network called FSR.

FTM: A Frame-Level Timeline Modeling Method for Temporal Graph Representation Learning

Bowen Cao (Peking University), Yuexian Zou (Peking University)

CodeRepresentation LearningGraph Neural NetworkGraphTime SeriesFinance Related

🎯 What it does: This paper proposes a temporal graph representation learning method called FTM, which is based on frame-level timeline modeling. It uses framing techniques and a timeline aggregator to capture both short-term and long-term features, thereby enhancing the quality of temporal graph representations.

Fully-Dynamic Decision Trees

Marco Bressan (University of Milan), Mauro Sozio (Institut Polytechnique de Paris)

CodeClassificationOptimizationTabular

🎯 What it does: Designed and implemented the first fully dynamic decision tree algorithm (FUDYADT) that supports arbitrary insert/delete operations, capable of keeping the split point of each node in the tree within a preset Ρ of the optimal Gini gain at any moment, and achieving nearly optimal time-space complexity under real-time updates.

Function Approximation for Solving Stackelberg Equilibrium in Large Perfect Information Games

Chun Kai Ling (Carnegie Mellon University), Fei Fang (Carnegie Mellon University)

CodeOptimizationNeural Architecture SearchReinforcement LearningTabular

🎯 What it does: A function approximation-based learning framework is proposed, using neural networks to approximate the enforceable payoff frontier (EPF), thereby solving the Stackelberg extensive form correlated equilibrium (SEFCE) in large complete information two-player games.

GAN Prior Based Null-Space Learning for Consistent Super-resolution

Yinhuai Wang (Peking University), Jian Zhang (Peking University)

CodeRestorationSuper ResolutionGenerative Adversarial NetworkImage

🎯 What it does: A Pooling-Based Decomposition (PD) method is proposed based on GAN prior and null-space learning, utilizing the pseudo-inverse of average pooling to achieve parameter-free, no additional computation range-zero space decomposition, significantly eliminating low-frequency inconsistencies in super-resolution results.

Generalized Category Discovery with Decoupled Prototypical Network

Wenbin An (Xi'an Jiaotong University), Ping Chen (University of Massachusetts Boston)

CodeClassificationTransformerText

🎯 What it does: By decoupling known categories from unknown categories in unlabeled data and utilizing prototype matching and semantic-aware prototype learning, a Decoupled Prototypical Network (DPN) is proposed for Generalized Category Discovery (GCD).

Generalized Semantic Segmentation by Self-Supervised Source Domain Projection and Multi-Level Contrastive Learning

Liwei Yang (Xi'an Jiaotong University), Jian Sun (Pazhou Laboratory)

CodeSegmentationDomain AdaptationAutonomous DrivingContrastive LearningImage

🎯 What it does: Enhancing the domain generalization performance of semantic segmentation through Self-Supervised Source Domain Projection (SSDP) and Multi-Layer Contrastive Learning (MLCL).

Generalizing Math Word Problem Solvers via Solution Diversification

Zhenwen Liang (University of Notre Dame), Xiangliang Zhang (University of Notre Dame)

CodeContrastive LearningText

🎯 What it does: A general training framework is proposed, which dynamically generates and filters diverse, quality-controllable answers during the training process of the MWP Solver by introducing an answer buffer and a discriminator, thereby enhancing the model's generalization ability.

Generalizing Multiple Object Tracking to Unseen Domains by Introducing Natural Language Representation

En Yu (Huazhong University of Science and Technology), Wenbing Tao (Huazhong University of Science and Technology)

CodeObject TrackingDomain AdaptationAutonomous DrivingTransformerVision Language ModelVideoText

🎯 What it does: This paper proposes LTrack, a multi-object tracking model that combines natural language descriptions with visual context. It utilizes Visual Context Prompts (VCP) and a Visual-Language Mixing (VLM) module to automatically generate Pseudo Text Descriptions (PTD), thereby enhancing cross-domain generalization performance.

Generating Coherent Narratives by Learning Dynamic and Discrete Entity States with a Contrastive Framework

Jian Guan (Tsinghua University), Minlie Huang (Tsinghua University)

CodeGenerationTransformerContrastive LearningText

🎯 What it does: A discrete entity state representation framework based on contrastive learning (ERIC) is proposed, integrating dynamic updates of entity states and a state attention layer into the Transformer decoder to achieve coherent text generation for stories and news.

Generating Transferable 3D Adversarial Point Cloud via Random Perturbation Factorization

Bangyan He (University of Chinese Academy of Sciences), Xiaochun Cao (Sun Yat-sen University)

CodeAdversarial AttackGraph Neural NetworkPoint Cloud

🎯 What it does: This paper proposes an attack method based on Randomized Perturbation Factorization (PF-Attack), which generates more transferable 3D point cloud adversarial samples by simultaneously optimizing the perturbation and its sub-perturbations.

Generic and Dynamic Graph Representation Learning for Crowd Flow Modeling

Liangzhe Han (Beihang University), Tongyu Zhu (Beihang University)

CodeRepresentation LearningGraph Neural NetworkGraphTime Series

🎯 What it does: A general dynamic graph representation learning framework GDCF is constructed for various crowd flow prediction tasks.

GENNAPE: Towards Generalized Neural Architecture Performance Estimators

Keith G. Mills (University of Alberta), Di Niu (University of Alberta)

CodeNeural Architecture SearchGraph Neural NetworkContrastive LearningGraphBenchmark

🎯 What it does: A general neural network architecture performance estimator, GENNAPE, is proposed, which can make accurate predictions on unseen network structures.

Global-Local Characteristic Excited Cross-Modal Attacks from Images to Videos

Ruikui Wang (Beihang University), Yunhong Wang (Beihang University)

CodeAdversarial AttackConvolutional Neural NetworkImageVideo

🎯 What it does: A cross-modal adversarial attack method from images to videos is proposed, enhancing the cross-modal transferability of adversarial samples through global interaction enhancement and local relevant perturbation techniques.

GLT-T: Global-Local Transformer Voting for 3D Single Object Tracking in Point Clouds

Jiahao Nie (Hangzhou Dianzi University), Jing Zhang (University of Sydney)

CodeObject TrackingAutonomous DrivingTransformerPoint Cloud

🎯 What it does: A global-local Transformer voting framework GLT-T is proposed to improve the seed point voting of VoteNet for higher quality 3D single-object tracking.

GMDNet: A Graph-Based Mixture Density Network for Estimating Packages’ Multimodal Travel Time Distribution

Xiaowei Mao (Beijing Jiaotong University), Youfang Lin (Cainiao Network)

CodeGraph Neural NetworkMixture of ExpertsGraphTime Series

🎯 What it does: This paper proposes a Graph Mixture Density Network (GMDNet) for accurately predicting the multimodal travel time distribution of packages in logistics networks based on routes and graph structures.

Goal-Conditioned Generators of Deep Policies

Francesco Faccio (AI Initiative), JΓΌrgen Schmidhuber

CodeGenerationReinforcement LearningSequential

🎯 What it does: This paper proposes a goal-conditioned policy generator GoGePo, which directly generates deep neural network policies based on the 'expected return' command using Fast Weight Programmer (FWP) and hypernetworks;

Goal-Conditioned Q-learning as Knowledge Distillation

Alexander Levine (University of Maryland), Soheil Feizi (University of Maryland)

CodeKnowledge DistillationReinforcement Learning

🎯 What it does: Two target-conditioned offline reinforcement learning algorithms based on gradient attention transfer, ReenGAGE and Multi-ReenGAGE, are proposed, and their performance is validated in high-dimensional target spaces and multi-target sparse reward environments.

Gradient-Based Graph Attention for Scene Text Image Super-resolution

Xiangyuan Zhu (Central South University), Gerald Schaefer (Loughborough University)

CodeRestorationSuper ResolutionGraph Neural NetworkImageBenchmark

🎯 What it does: A gradient map-based graph attention network is proposed, combining group attention and channel attention to achieve text image super-resolution.

Graph Anomaly Detection via Multi-Scale Contrastive Learning Networks with Augmented View

Jingcan Duan (National University of Defense Technology), Zhibin Dong (National University of Defense Technology)

CodeAnomaly DetectionGraph Neural NetworkContrastive LearningGraph

🎯 What it does: This paper proposes the GRADATE framework for the task of graph anomaly detection, employing multi-scale contrastive learning (node-subgraph, node-node, subgraph-subgraph) and achieving multi-view enhancement of graphs through edge modification to improve the robustness of subgraph embeddings and detection performance.

Graph Component Contrastive Learning for Concept Relatedness Estimation

Yueen Ma (Chinese University of Hong Kong), Irwin King (Chinese University of Hong Kong)

CodeRepresentation LearningGraph Neural NetworkTransformerContrastive LearningGraph

🎯 What it does: Construct a concept-level graph called ConcreteGraph, and use it for data augmentation and graph component contrastive learning to address the high-order relationships in concept relevance estimation and the issue of data scarcity.

Graph Ordering Attention Networks

Michail Chatzianastasis (Ecole Polytechnique), Michalis Vazirgiannis (Ecole Polytechnique)

CodeRecurrent Neural NetworkGraph Neural NetworkGraph

🎯 What it does: A new graph neural network layer (GOAT) is proposed, which captures the collaborative information between neighbors by learning neighbor ranking and using recurrent networks for aggregation.

Graph Structure Learning on User Mobility Data for Social Relationship Inference

Guangming Qin (Ocean University of China), Junyu Dong (Ocean University of China)

CodeRecommendation SystemGraph Neural NetworkGraph

🎯 What it does: Proposes the SRINet framework, which utilizes graph structure learning to remove noise edges from user mobility data and applies GCN for social relationship inference on multi-semantic meeting graphs.

Graphix-T5: Mixing Pre-trained Transformers with Graph-Aware Layers for Text-to-SQL Parsing

Jinyang Li (University of Hong Kong), Yongbin Li (Alibaba Group)

CodeGraph Neural NetworkTransformerLarge Language ModelText

🎯 What it does: The GRAPHIX-T5 model is proposed, which integrates the pre-trained T5 with graph neural networks to enhance the reasoning and structured generation capabilities from cross-domain text to SQL.

GraphPrompt: Graph-Based Prompt Templates for Biomedical Synonym Prediction

Hanwen Xu (University of Washington), Sheng Wang (University of Washington)

CodeDrug DiscoveryGraph Neural NetworkPrompt EngineeringGraphBiomedical Data

🎯 What it does: By transforming the ontology graph structure into prompt templates, a masked language model is utilized to complete the biomedical synonym prediction task.

Graphs, Constraints, and Search for the Abstraction and Reasoning Corpus

Yudong Xu (University of Toronto), Scott Sanner (University of Toronto)

CodeGraph Neural NetworkGraph

🎯 What it does: This paper proposes an object-centric framework called ARGA, based on graph abstraction and constraint-guided search, to solve ARC tasks.

GRASMOS: Graph Signage Model Selection for Gene Regulatory Networks

Angelina Brilliantova (Rochester Institute of Technology), Ivona BezΓ‘kovΓ‘ (Rochester Institute of Technology)

CodeGraph Neural NetworkGraph

🎯 What it does: This paper proposes a maximum likelihood-based graph identification model selection method (GRASMOS) for inferring the distribution of positive and negative edges in a gene regulatory network with a given topology.

GRLSTM: Trajectory Similarity Computation with Graph-Based Residual LSTM

Silin Zhou (University of Electronic Science and Technology of China), Peng Han (University of Electronic Science and Technology of China)

CodeRecurrent Neural NetworkGraph Neural NetworkGraphTime SeriesSequential

🎯 What it does: This paper proposes the GRLSTM framework, which implements trajectory similarity calculation based on road networks using knowledge graph embedding, graph attention networks, and residual multi-layer LSTM.

H-TSP: Hierarchically Solving the Large-Scale Traveling Salesman Problem

Xuanhao Pan (Huazhong University of Science and Technology), Jiang Bian (University of Science and Technology of China)

CodeOptimizationConvolutional Neural NetworkTransformerReinforcement LearningGraph

🎯 What it does: An end-to-end framework H-TSP based on hierarchical reinforcement learning is proposed to solve large-scale (up to 10,000 nodes) traveling salesman problems;

Hard Sample Aware Network for Contrastive Deep Graph Clustering

Yue Liu (National University of Defense Technology), Cancan Chen (National University of Defense Technology)

CodeGraph Neural NetworkContrastive LearningGraph

🎯 What it does: A Hard Sample Aware Network (HSAN) is designed for deep graph clustering, improving clustering performance through dynamic weighting of hard positive and negative samples.

Head-Free Lightweight Semantic Segmentation with Linear Transformer

Bo Dong (Alibaba Group), Fan Wang (Amazon Prime Video)

CodeSegmentationTransformerImage

🎯 What it does: A headless lightweight semantic segmentation framework AFFormer is proposed, which directly utilizes prototype learning and adaptive frequency filtering on high-resolution features to achieve semantic recovery, completely eliminating the traditional decoder.

Heterogeneous Graph Masked Autoencoders

Yijun Tian (University of Notre Dame), Nitesh V. Chawla (Brandeis University)

CodeGenerationRepresentation LearningGraph Neural NetworkAuto EncoderGraph

🎯 What it does: This paper proposes HGMAE, a generative self-supervised masked autoencoder for heterogeneous graphs, aimed at learning node representations in the absence of labels.

Heterogeneous Region Embedding with Prompt Learning

Silin Zhou (University of Electronic Science and Technology of China), Peng Han (University of Electronic Science and Technology of China)

CodeGraph Neural NetworkPrompt EngineeringGraph

🎯 What it does: This paper proposes the HREP framework, which constructs a heterogeneous graph containing human flow, POI, and geographical adjacency relationships. It utilizes a relation-aware GCN and self-attention fusion to obtain high-quality regional embeddings, and enhances performance in downstream tasks through prefix-tuning for prompt learning.

Heuristic Search for Multi-Objective Probabilistic Planning

Dillon Z. Chen (Australian National University), Sylvie ThiΓ©baux (Australian National University)

CodeOptimizationReinforcement Learning from Human FeedbackTabularBenchmark

🎯 What it does: This paper defines the Multi-Objective Stochastic Shortest Path (MOSSP) problem and proposes a heuristic search-based solution algorithm.

Hierarchical Contrast for Unsupervised Skeleton-Based Action Representation Learning

Jianfeng Dong (Zhejiang Gongshang University), Xun Wang (Zhejiang Gongshang University)

CodeRecognitionRetrievalRepresentation LearningContrastive LearningVideo

🎯 What it does: A hierarchical contrastive learning framework named HiCo is proposed for unsupervised skeleton action representation learning.

Hierarchical ConViT with Attention-Based Relational Reasoner for Visual Analogical Reasoning

Wentao He (University of Nottingham), Xudong Jiang (Nanyang Technological University)

CodeRecognitionConvolutional Neural NetworkTransformerImage

🎯 What it does: A model based on hierarchical ConViT and an attention-based relational reasoner is proposed to address the visual analogy reasoning task in Raven Progressive Matrices (RPM).

Hierarchical Event Grounding

Jiefu Ou (Carnegie Mellon University), Teruko Mitamura (Carnegie Mellon University)

CodeRetrievalTransformerSupervised Fine-TuningText

🎯 What it does: Designed and implemented a hierarchical event alignment task, constructed a multilingual event hierarchy dataset, and proposed a retrieval method based on hierarchical loss.

Hierarchical Text Classification as Sub-hierarchy Sequence Generation

SangHun Im (Korea University of Technology and Education), Dong Hwan Kim (Korea University of Technology and Education)

CodeClassificationTransformerText

🎯 What it does: This paper proposes transforming hierarchical text classification into sub-hierarchical sequence generation and designs the HiDEC recursive sub-hierarchical decoder, which utilizes hierarchical-aware masked self-attention and text-hierarchical attention to achieve unstructured encoding of hierarchical information modeling.

High-Dimensional Dueling Optimization with Preference Embedding

Yangwenhui Zhang (East China Normal University), Aimin Zhou (East China Normal University)

CodeOptimizationDrug DiscoveryTabular

🎯 What it does: This paper proposes Preference-Embedded Dual Decision Bayesian Optimization (PE-DBO), which addresses the scalability issue of high-dimensional dual decision black-box optimization by optimizing in a low-dimensional subspace and mapping the pairwise comparisons back to the original high-dimensional space.

High-Resolution GAN Inversion for Degraded Images in Large Diverse Datasets

Yanbo Wang (East China Normal University), Yuan Xie (East China Normal University)

CodeImage TranslationRestorationGenerative Adversarial NetworkImage

🎯 What it does: A GAN inverse framework CRi is proposed for degraded image restoration using StyleGAN-XL.

HRDoc: Dataset and Baseline Method toward Hierarchical Reconstruction of Document Structures

Jiefeng Ma (University of Science and Technology of China), Cong Liu (iFLYTEK Research)

CodeClassificationSegmentationRecurrent Neural NetworkTransformerSupervised Fine-TuningTextMultimodality

🎯 What it does: Proposes a hierarchical document structure reconstruction task and implements an end-to-end parsing system (DSPS).