AAAI 2024 Papers — Page 18
AAAI Conference on Artificial Intelligence · 2331 papers
Refining Latent Homophilic Structures over Heterophilic Graphs for Robust Graph Convolution Networks
Chenyang Qiu (Beijing University of Posts and Telecommunications), Xiaofeng Tao (Beijing University of Posts and Telecommunications)
ClassificationAdversarial AttackGraph Neural NetworkContrastive LearningGraph
🎯 What it does: This paper studies the robustness of Graph Convolutional Networks (GCN) on heterophilic graphs and proposes a framework named LHS, which enhances the performance of GCN under structural attacks by learning potential homophilic structures.
ReGCL: Rethinking Message Passing in Graph Contrastive Learning
Cheng Ji (Beihang University), Jianxin Li (Beihang University)
Representation LearningGraph Neural NetworkContrastive LearningGraph
🎯 What it does: This paper studies the conflict between the graph neural network (GNN) encoder and the InfoNCE loss in graph contrastive learning (GCL), and proposes the ReGCL framework to alleviate this conflict through gradient-guided structural learning and gradient-weighted InfoNCE, thereby improving the quality of node representations.
Region-Aware Exposure Consistency Network for Mixed Exposure Correction
Jin Liu (Beijing University of Posts and Telecommunications), Huadong Ma (Beijing University of Posts and Telecommunications)
RestorationConvolutional Neural NetworkContrastive LearningImage
🎯 What it does: Designed the Region-aware Exposure Correction Network (RECNet), which simultaneously corrects mixed exposure images through a region-aware de-exposure module and a multi-scale recovery unit.
Region-Disentangled Diffusion Model for High-Fidelity PPG-to-ECG Translation
Debaditya Shome (Queen's University), Ali Etemad (Queen's University)
GenerationData SynthesisDiffusion modelTime SeriesBiomedical DataElectronic Health RecordsElectrocardiogramBenchmark
🎯 What it does: A regional decoupled diffusion model (RDDM) is proposed to convert photoplethysmography (PPG) signals into high-fidelity electrocardiogram (ECG) signals.
REGLO: Provable Neural Network Repair for Global Robustness Properties
Feisi Fu (Boston University), Wenchao Li (Boston University)
OptimizationAdversarial AttackTabularFinance Related
🎯 What it does: A neural network repair framework called REGLO is proposed, which is based on global robustness properties. It allows for post-hoc weight adjustments of pre-trained models to meet given global robustness constraints (including individual fairness).
Regret Analysis of Policy Gradient Algorithm for Infinite Horizon Average Reward Markov Decision Processes
Qinbo Bai (Purdue University), Vaneet Aggarwal (Purdue University)
Reinforcement Learning
🎯 What it does: A strategy gradient algorithm based on universal parameterization is proposed for infinite average reward Markov decision processes, and its convergence and sublinear scheduling loss (scheduling loss is ˜O(T^{3/4})) are theoretically proven.
Regret Analysis of Repeated Delegated Choice
Mohammad Hajiaghayi, Suho Shin (University of Maryland)
Optimization
🎯 What it does: The study investigates how the principal can delegate the selection of options to the agent through the publication of an acceptable set of options in a multi-round interaction under information asymmetry, achieving online learning and optimization without knowing the distribution of options.
Regroup Median Loss for Combating Label Noise
Fengpeng Li (University of Macau), Jiantao Zhou (University of Macau)
ClassificationImage
🎯 What it does: This paper proposes the Regroup Median Loss (RML) method for robustly estimating sample loss under label noise conditions, and further constructs a semi-supervised learning framework based on RML to enhance the model's generalization performance on noisy data.
Regulating AI: Applying Insights from Behavioural Economics and Psychology to the Application of Article 5 of the EU AI Act
Huixin Zhong (University of Bath), Janina A. Hoffmann (University of Bath)
Review/Survey Paper
🎯 What it does: By combining theories from psychology and behavioral economics, this paper clarifies key terms in Article 5 of the EU AI Act (covert techniques, manipulation techniques, deception techniques) and demonstrates the practical application of these techniques in AI through case analysis. Finally, it proposes targeted revision suggestions to enhance the legal protection effectiveness.
Regulating Intermediate 3D Features for Vision-Centric Autonomous Driving
Junkai Xu (Zhejiang University), Deng Cai (Zhejiang University)
Object DetectionSegmentationAutonomous DrivingNeural Radiance FieldPoint Cloud
🎯 What it does: The Vampire framework is proposed, utilizing volume rendering to adjust intermediate dense 3D features under multiple cameras, making them suitable for both dense tasks and object detection.
Reinforced Adaptive Knowledge Learning for Multimodal Fake News Detection
Litian Zhang (Beihang University), Chaozhuo Li (Beijing University of Posts and Telecommunications)
ClassificationGraph Neural NetworkReinforcement LearningMultimodalityGraph
🎯 What it does: This paper proposes AKA-Fake, a model that utilizes reinforcement learning to generate adaptive knowledge subgraphs and implements multimodal fake news detection through heterogeneous graph learning.
Reinforcement Learning and Data-Generation for Syntax-Guided Synthesis
Julian Parsert (University of Oxford), Elizabeth Polgreen (University of Edinburgh)
Data SynthesisOptimizationReinforcement LearningTabular
🎯 What it does: This paper proposes a Monte Carlo Tree Search (MCTS) algorithm based on reinforcement learning to solve the Syntax-Guided Synthesis (SyGuS) problem, and automatically generates training data through a technique called inverse unification.
Reinforcement Learning as a Parsimonious Alternative to Prediction Cascades: A Case Study on Image Segmentation
Bharat Srikishan (Stevens Institute of Technology), Nikhil Muralidhar (Oak Ridge National Laboratory)
SegmentationComputational EfficiencyKnowledge DistillationConvolutional Neural NetworkReinforcement LearningImage
🎯 What it does: This paper proposes a multi-model segmentation pipeline called PaSeR based on reinforcement learning, which can select the most suitable segmentation model for each image block on demand without using a cascade structure, thereby significantly reducing inference costs while maintaining high segmentation accuracy.
Relational Distant Supervision for Image Captioning without Image-Text Pairs
Yayun Qi (Beijing Institute of Technology), Xinxiao Wu (Shenzhen MSU-BIT University)
GenerationRetrievalGraph Neural NetworkTransformerImageText
🎯 What it does: Through an unsupervised relational distant supervision method, external textual corpus is utilized to infer object relationships in images, thereby generating pseudo image-sentence pairs for image description training;
Relational Programming with Foundational Models
Ziyang Li (University of Pennsylvania), Mayur Naik (University of Pennsylvania)
RetrievalRecommendation SystemTransformerLarge Language ModelPrompt EngineeringImageTextMultimodalityRetrieval-Augmented Generation
🎯 What it does: This paper presents VIEIRA, a declarative framework based on relational programming, designed to unify the invocation and combination of multimodal foundational models (such as GPT, CLIP, SAM, etc.), enabling seamless interaction and reasoning from text, images to structured databases.
Relative Policy-Transition Optimization for Fast Policy Transfer
Jiawei Xu (Tencent Robotics X), Lei Han (Tencent Robotics X)
OptimizationReinforcement LearningTabular
🎯 What it does: A strategy-transfer optimization framework based on the theory of value relativity is proposed to achieve rapid and robust policy transfer between the source and target environments.
Relax Image-Specific Prompt Requirement in SAM: A Single Generic Prompt for Segmenting Camouflaged Objects
Jian Hu (Queen Mary University of London), Weitong Cai (Queen Mary University of London)
Object DetectionSegmentationTransformerPrompt EngineeringVision Language ModelImage
🎯 What it does: An unsupervised CAMO detection method called GenSAM is proposed, which utilizes general task descriptions to automatically generate image-specific visual prompts for the segmentation of concealed objects.
Relaxed Stationary Distribution Correction Estimation for Improved Offline Policy Optimization
Woosung Kim (Korea University), Byung-Jun Lee (Korea University)
Reinforcement LearningTabularBenchmark
🎯 What it does: An offline reinforcement learning algorithm named PORelDICE is proposed, which alleviates estimation bias and performance degradation caused by the shape of conjugate functions by relaxing the positivity constraint within an implicit policy optimization framework and introducing an additional function approximator.
Relevant Intrinsic Feature Enhancement Network for Few-Shot Semantic Segmentation
Xiaoyi Bao (University of Chinese Academy of Sciences), Yun Zheng (Alibaba Group)
SegmentationConvolutional Neural NetworkTransformerImage
🎯 What it does: Designed and trained a multi-layer prototype enhancement network RiFeNet with an unlabeled branch for few-shot semantic segmentation, improving pixel-level binary classification performance by reinforcing intra-class consistency and inter-class separability.
Reliable Conflictive Multi-View Learning
Cai Xu (Xidian University), Xiyue Gao (Xidian University)
ClassificationRecognitionImageVideo
🎯 What it does: This paper proposes the Reliable Conflict Multi-Perspective Learning (RCML) problem and designs an evidence theory-based conflict opinion aggregation method, ECML, to provide reliable decisions and their confidence in the presence of conflicting perspectives in the data.
Reliable Data Generation and Selection for Low-Resource Relation Extraction
Junjie Yu (Soochow University), Wenliang Chen (Soochow University)
GenerationData SynthesisTransformerLarge Language ModelText
🎯 What it does: Proposes the Self-RDGS method, which utilizes LLM to automatically generate sentences and selects reliable training data through ranking to construct a high-quality training set for low-resource relation extraction tasks.
Relightable and Animatable Neural Avatars from Videos
Wenbin Lin (Tsinghua University), Feng Xu (Tsinghua University)
GenerationData SynthesisPose EstimationNeural Radiance FieldVideo
🎯 What it does: Reconstructing a re-lightable and animatable neural human avatar from sparse multi-view videos can generate realistic images under arbitrary poses, lighting, and viewpoints.
Removing Interference and Recovering Content Imaginatively for Visible Watermark Removal
Yicheng Leng (Xidian University), Guanbin Li (Sun Yat-sen University)
RestorationImage
🎯 What it does: A two-stage visible watermark removal framework RIRCI is proposed, which first separates the watermark components and then restores the background.
Repeated Fair Allocation of Indivisible Items
Ayumi Igarashi (University of Tokyo), Arianna Novaro (University of Paris 1 Panthéon-Sorbonne)
Optimization
🎯 What it does: This paper proposes and studies the trade-off between fairness and efficiency in the context of repeated allocation, providing several results on the existence and algorithms for multi-round allocation.
Reproduce, Replicate, Reevaluate. The Long but Safe Way to Extend Machine Learning Methods
Luisa Werner (Univ. Grenoble Alpes), Damien Graux (Trinity College Dublin)
Graph Neural NetworkGraph
🎯 What it does: Using a three-step process of reproduction-reimplementation-reevaluation, we systematically reproduce, reimplement, and cross-evaluate the Knowledge Enhanced Neural Network (KENN) experiments on Citation graphs, summarizing experiences and obstacles.
REPrune: Channel Pruning via Kernel Representative Selection
Mincheol Park (Yonsei University), Suhyun Kim (Hyundai MOBIS)
Object DetectionCompressionComputational EfficiencyConvolutional Neural NetworkImage
🎯 What it does: This paper proposes REPrune, a method for channel pruning that clusters similar convolution kernels during training and selects filters containing representative kernels.
ResDiff: Combining CNN and Diffusion Model for Image Super-resolution
Shuyao Shang (Shandong University), Jinglin Zhang (Shandong University)
RestorationSuper ResolutionConvolutional Neural NetworkDiffusion modelImage
🎯 What it does: Proposes the ResDiff framework, which combines CNN with diffusion models, first using CNN to predict low-frequency content and then using the diffusion model to predict the residuals for single-image super-resolution.
Residual Hyperbolic Graph Convolution Networks
Yangkai Xue (Beijing Institute of Technology), Yunde Jia (Shenzhen MSU-BIT University)
ClassificationGraph Neural NetworkGraph
🎯 What it does: This paper proposes a Residual Hypercurve Graph Convolutional Network (R-HGCN) to address the over-smoothing problem of traditional Hypercurve GCNs through hypercurve residual connections and product manifolds, and introduces HyperDrop regularization to reduce overfitting.
Resisting Backdoor Attacks in Federated Learning via Bidirectional Elections and Individual Perspective
Zhen Qin (Zhejiang University), Shuiguang Deng (Zhejiang University)
Federated LearningAuto EncoderImage
🎯 What it does: A dual-election-based backdoor attack framework called Snowball is proposed, which utilizes individual perspectives to vote on model updates and perform incremental filtering, and employs VAE to learn model differences to further exclude malicious updates.
ResMatch: Residual Attention Learning for Feature Matching
Yuxin Deng (Wuhan University), Jiayi Ma (Wuhan University)
Pose EstimationRetrievalGraph Neural NetworkTransformerImage
🎯 What it does: A residual attention learning framework ResMatch and its sparse version sResMatch are designed to improve the accuracy of image feature matching.
Resource Democratization: Is Compute the Binding Constraint on AI Research?
Rebecca Gelles (Georgetown University), James Dunham (Georgetown University)
🎯 What it does: A large-scale survey of AI researchers in the United States explores the use and limitations of computing, data, and talent resources.
Resource Efficient Deep Learning Hardware Watermarks with Signature Alignment
Joseph Clements (Applied Research Associates), Yingjie Lao (Tufts University)
Convolutional Neural NetworkTransformerImage
🎯 What it does: To protect the ownership of deep learning hardware, a method is proposed that aligns watermark key samples with hardware modifications, achieving a higher success rate and lower hardware overhead.
Responding to the Call: Exploring Automatic Music Composition Using a Knowledge-Enhanced Model
Zhejing Hu (Hong Kong Polytechnic University), Qianwen Luo (Shenzhen University)
GenerationTransformerAudio
🎯 What it does: This paper studies the application of call-and-response in automatic music creation, constructing the first CRD dataset containing 19,155 call-and-response pairs, and proposes a knowledge-enhanced Seq2Seq generator (CRG) to produce creatively responsive music.
Response Enhanced Semi-supervised Dialogue Query Generation
Jianheng Huang (Xiamen University), Jinsong Su (Xiamen University)
GenerationTransformerLarge Language ModelReinforcement LearningText
🎯 What it does: A semi-supervised dialogue query generation framework, SemiDQG, is designed to utilize a response-enhanced query generator, RA, to generate and filter pseudo-queries from unlabeled dialogues, and then improve the performance of the standard query generator, QP, through rewards provided by RA.
Responsibility in Extensive Form Games
Qi SHI
🎯 What it does: Formalizes responsibility in generalized extensive form games and proposes a new definition of 'seeing responsibility' that combines strategic and achievement perspectives.
Restoring Speaking Lips from Occlusion for Audio-Visual Speech Recognition
Jiadong Wang (National University of Singapore), Haizhou Li (Shenzhen Research Institute of Big Data)
RecognitionRestorationConvolutional Neural NetworkAuto EncoderGenerative Adversarial NetworkContrastive LearningVideoAudio
🎯 What it does: This paper addresses the issue of recognition performance degradation caused by mouth occlusion in audio-visual speech recognition, proposing an Audio-Visual Lip Recovery framework (AVLR) that can detect and mask occluded areas, and then synthesize-match reconstruct using synchronized audio and unobstructed reference images to recover the occluded lips.
Rethinking Causal Relationships Learning in Graph Neural Networks
Hang Gao (Institute of Software Chinese Academy of Sciences), Huaping Liu (Tsinghua University)
Graph Neural NetworkGraph
🎯 What it does: Construct a synthetic graph dataset with controllable causal relationships, CRCG, and evaluate the causal learning ability of GNN based on this. Subsequently, propose a lightweight R-CAM module to enhance the causal modeling effectiveness of GNN.
Rethinking Dimensional Rationale in Graph Contrastive Learning from Causal Perspective
Qirui Ji (Institute of Software Chinese Academy of Sciences), Fanjiang Xu (Institute of Software Chinese Academy of Sciences)
Representation LearningMeta LearningGraph Neural NetworkContrastive LearningGraph
🎯 What it does: This study proposes a Dimensional Rationale framework in graph contrastive learning, utilizing a learnable dimension weight network and redundancy reduction regularization, and employs meta-learning for backdoor adjustment to eliminate irrelevant information, enhancing the discriminability and transferability of graph representations.
Rethinking Graph Masked Autoencoders through Alignment and Uniformity
Liang Wang (Chinese Academy of Sciences)
ClassificationRepresentation LearningGraph Neural NetworkAuto EncoderContrastive LearningGraph
🎯 What it does: This paper establishes a connection between Graph Masked Autoencoders (GraphMAE) and Graph Contrastive Learning (GCL) through theoretical analysis, pointing out their limitations in alignment and uniformity, and subsequently proposes the AUG-MAE model to overcome these issues.
Rethinking Mesh Watermark: Towards Highly Robust and Adaptable Deep 3D Mesh Watermarking
Xingyu Zhu (Southern University of Science and Technology), Xuetao Wei (Southern University of Science and Technology)
Graph Neural NetworkAuto EncoderMesh
🎯 What it does: A three-dimensional mesh watermarking algorithm called DEEP3DMARK based on graph attention networks is proposed, which can embed binary information into the distribution of mesh vertices while maintaining geometric imperceptibility, and possesses robustness against various attacks and adaptability to different mesh sizes and geometries.
Rethinking Multi-Scale Representations in Deep Deraining Transformer
Hongming Chen (Shenyang Aerospace University), Yufeng Li (Shenyang Aerospace University)
RestorationTransformerImage
🎯 What it does: A multi-scale input multi-scale output Transformer network (MSDT) is proposed, which implements raindrop removal in a pyramid manner from coarse to fine.
Rethinking Peculiar Images by Diffusion Models: Revealing Local Minima’s Role
Jinhyeok Jang (Korea Advanced Institute of Science and Technology), Changha Lee (Korea Advanced Institute of Science and Technology)
GenerationPose EstimationOptimizationDiffusion modelImage
🎯 What it does: The researchers introduced momentum (ordinary momentum and positive-negative momentum) into the diffusion model to escape local minima, thereby reducing the probability of generating singular images.
Rethinking Propagation for Unsupervised Graph Domain Adaptation
Meihan Liu (Zhejiang University), Jiajun Bu (Tsinghua University)
Domain AdaptationGraph Neural NetworkGraph
🎯 What it does: A heterogeneous A2GNN model is proposed, which re-evaluates the role of the propagation layer in GNNs for unsupervised graph domain adaptation, providing both theoretical and experimental support.
Rethinking Reverse Distillation for Multi-Modal Anomaly Detection
Zhihao Gu (Shanghai Jiao Tong University), Lizhuang Ma (Shanghai Jiao Tong University)
Anomaly DetectionKnowledge DistillationConvolutional Neural NetworkImageMultimodalityPoint Cloud
🎯 What it does: A multi-modal reverse distillation (MMRD) framework is proposed, utilizing a frozen multi-modal teacher encoder and a learnable student decoder for joint anomaly detection and localization of RGB and auxiliary modalities such as depth/normal.
Rethinking Robustness of Model Attributions
Sandesh Kamath (Indian Institute of Technology), Vineeth N Balasubramanian (Microsoft Research)
Explainability and InterpretabilityAdversarial AttackConvolutional Neural NetworkImage
🎯 What it does: This paper addresses the robustness issue of model interpretation (feature importance mapping) by proposing more reasonable evaluation metrics and improvement methods, and conducting experimental validation on different datasets and models.
Rethinking the Paradigm of Content Constraints in Unpaired Image-to-Image Translation
Xiuding Cai (Chengdu Institute of Computer Application Chinese Academy of Sciences), Yu Yao (Chengdu Institute of Computer Application Chinese Academy of Sciences)
Image TranslationGenerationConvolutional Neural NetworkGenerative Adversarial NetworkImage
🎯 What it does: Proposes the EnCo framework, which utilizes the similarity constraint of representations in the latent space of generator encoders and decoders with the same-order features to achieve content preservation in unpaired image-to-image translation; simultaneously designs a parameter-independent discriminator-guided patch sampling strategy DAG;
Rethinking Two-Stage Referring Expression Comprehension: A Novel Grounding and Segmentation Method Modulated by Point
Peizhi Zhao (Guangxi University), Qingbao Huang (Guangxi University)
RecognitionObject DetectionSegmentationConvolutional Neural NetworkTransformerImageText
🎯 What it does: A point-based two-stage referential expression understanding framework is proposed, transforming the localization and understanding process into point-level cross-modal understanding and point-level localization.
RetouchFormer: Semi-supervised High-Quality Face Retouching Transformer with Prior-Based Selective Self-Attention
Xue Wen, Hau-San Wong (City University of Hong Kong)
Image TranslationRestorationTransformerGenerative Adversarial NetworkImage
🎯 What it does: A semi-supervised facial beautification framework called RetouchFormer based on Transformer is proposed, which can achieve high-quality facial defect recognition and repair on unpaired data.
Retrieval-Augmented Primitive Representations for Compositional Zero-Shot Learning
Chenchen Jing (Zhejiang University), Chunhua Shen (Zhejiang University)
ClassificationRecognitionRetrievalTransformerPrompt EngineeringImageRetrieval-Augmented Generation
🎯 What it does: A retrieval-enhanced combination zero-shot learning method is proposed, which constructs two retrieval databases using attributes and object representations from the training set. When inputting an image, the retrieval module obtains image features of the same attributes/objects and fuses them with the original primitive features, thereby improving the recognition of unseen combinations.
RetroOOD: Understanding Out-of-Distribution Generalization in Retrosynthesis Prediction
Yemin Yu (City University of Hong Kong), Xinhai Ye (Shanghai Institute for Advanced Study of Zhejiang University)
Domain AdaptationData-Centric LearningDrug DiscoveryGraph Neural NetworkGraph
🎯 What it does: This paper systematically studies the generalization problem of single-step inverse synthesis prediction in out-of-distribution (OOD) environments, first categorizing OOD into label (retro-strategy) shift and covariate (target molecule) shift, and constructing corresponding benchmark datasets for each; subsequently, five mainstream models are retrained and evaluated, and two model-agnostic improvement methods are proposed—Invariant Risk Minimization (IRM) to enhance robustness against covariate shift, and an energy-based model (EBM) concept enhancement method to alleviate the recall deficiency of label shift.
Revealing the Proximate Long-Tail Distribution in Compositional Zero-Shot Learning
Chenyi Jiang (Nanjing University of Science and Technology), Haofeng Zhang (Nanjing University of Science and Technology)
ClassificationRecognitionConvolutional Neural NetworkContrastive LearningImageBenchmark
🎯 What it does: This paper views Combinatorial Zero-Shot Learning (CZSL) as an approximate long-tail distribution problem and enhances model performance by introducing attribute priors and a logit adjustment method (ProLT).
Reverse Multi-Choice Dialogue Commonsense Inference with Graph-of-Thought
Li Zheng (Wuhan University), Chong Teng (Wuhan University)
TransformerLarge Language ModelPrompt EngineeringTextChain-of-Thought
🎯 What it does: A three-step reverse exclusion graph thinking framework (ReX-GoT) is proposed for dialogue commonsense multiple-choice question answering.
Review-Enhanced Hierarchical Contrastive Learning for Recommendation
Ke Wang (Shanghai Jiao Tong University), Mengyuan Jing (Nanjing University of Aeronautics and Astronautics)
Recommendation SystemGraph Neural NetworkContrastive LearningText
🎯 What it does: This paper proposes a graph-based hierarchical contrastive learning framework, ReHCL, which enhances the embedding quality of users and items by constructing a topic graph and a semantic graph, and performing contrastive learning between the two views as well as across modalities (rating-comment), thereby significantly improving item recommendation performance.
Reviewing the Forgotten Classes for Domain Adaptation of Black-Box Predictors
Shaojie Zhang (Jilin University), Zeyu Zhang (Jilin University)
Domain AdaptationKnowledge DistillationImage
🎯 What it does: A method to address the issue of minority class forgetting in domain adaptation black-box predictors, called RFC, is proposed, which includes two modules: Selective Training (ST) and Neighborhood Clustering (NC).
Revisiting Disentanglement in Downstream Tasks: A Study on Its Necessity for Abstract Visual Reasoning
Ruiqian Nai (Tsinghua University), Yang Gao (Tsinghua University)
Representation LearningTransformerAuto EncoderContrastive LearningImage
🎯 What it does: This study investigates the necessity of causal disentangled representations for downstream performance in abstract visual reasoning tasks.
Revisiting Document-Level Relation Extraction with Context-Guided Link Prediction
Monika Jain (Indraprastha Institute of Information Technology), Kuldeep Singh (Cerence)
Explainability and InterpretabilityKnowledge DistillationGraph Neural NetworkText
🎯 What it does: This paper reframes the document-level relation extraction task as knowledge graph link prediction and improves relation prediction performance by integrating document reasoning with external knowledge (Wikidata, WordNet).
Revisiting Gradient Pruning: A Dual Realization for Defending against Gradient Attacks
Lulu Xue (Huazhong University of Science and Technology), Dezhong Yao (Huazhong University of Science and Technology)
Federated LearningSafty and PrivacyAdversarial AttackImage
🎯 What it does: Proposes the Dual Gradient Pruning (DGP) method, which enhances defense against gradient inversion attacks in collaborative learning by simultaneously pruning the maximum and minimum parameters in the gradient;
Revisiting Graph-Based Fraud Detection in Sight of Heterophily and Spectrum
Fan Xu (Beijing Jiaotong University), Hai Wan (Tsinghua University)
Anomaly DetectionOptimizationGraph Neural NetworkContrastive LearningGraphFinance Related
🎯 What it does: A semi-supervised graph neural network named SEC-GFD is proposed to address the highly heterogeneous and class-imbalanced fraud detection task.
Revisiting Open-Set Panoptic Segmentation
Yufei Yin (University of Science and Technology of China), Houqiang Li (University of Science and Technology of China)
Object DetectionSegmentationConvolutional Neural NetworkImage
🎯 What it does: A new long-tail distribution open-set panoptic segmentation dataset, LVIS-PS, is proposed, and based on this, the training and evaluation standards for the open-set panoptic segmentation (OPS) task are redefined.
Reward Penalties on Augmented States for Solving Richly Constrained RL Effectively
Hao Jiang (Singapore Management University), Huy Hoang (Singapore Management University)
Reinforcement LearningTabularBenchmark
🎯 What it does: This paper proposes an unconstrained reinforcement learning framework that incorporates cumulative costs in the state space and applies reward penalties to trajectories that violate constraints, effectively addressing safety reinforcement learning problems under various constraints (expected cost, Value-at-Risk, Conditional Value-at-Risk, worst-case). This framework is embedded in DQN and SAC, resulting in two algorithms: Safe DQN and Safe SAC.
RewriteLM: An Instruction-Tuned Large Language Model for Text Rewriting
Lei Shu (Google Research), Lei Meng (Google Research)
GenerationReinforcement Learning from Human FeedbackTransformerLarge Language ModelSupervised Fine-TuningReinforcement LearningTextBenchmarkChain-of-Thought
🎯 What it does: A large language model named RewriteLM has been developed for instruction tuning, specifically designed for cross-sentence text rewriting tasks.
RG-GAN: Dynamic Regenerative Pruning for Data-Efficient Generative Adversarial Networks
Divya Saxena (Hong Kong Polytechnic University), Tarun Kulshrestha (Hong Kong Polytechnic University)
GenerationData SynthesisGenerative Adversarial NetworkImage
🎯 What it does: A GAN training method called RG-GAN is proposed, which enhances generation quality and reduces model parameters through dynamic weight pruning and regeneration (recursive regeneration and incremental regeneration).
RGMComm: Return Gap Minimization via Discrete Communications in Multi-Agent Reinforcement Learning
Jingdi Chen (George Washington University), Carlee Joe-Wong (Carnegie Mellon University)
Reinforcement LearningSequential
🎯 What it does: A multi-agent reinforcement learning framework RGMComm based on discrete messages is proposed, which utilizes an online clustering method to minimize the return gap between the ideal fully observable policy and the limited communication policy.
Risk-Conditioned Reinforcement Learning: A Generalized Approach for Adapting to Varying Risk Measures
Gwangpyo Yoo (Sungkyunkwan University), Honguk Woo (Sungkyunkwan University)
Autonomous DrivingOptimizationReinforcement LearningAuto EncoderGenerative Adversarial NetworkTime SeriesSequentialFinance Related
🎯 What it does: A general risk conditional reinforcement learning framework (GRIPS) is proposed, which can quickly adapt to any weighted VaR (WV@R) risk measure; it generates risk measures that meet risk preferences through a risk proposal network and ensures the monotonicity of quantiles using non-crossing quantile regression; this framework is implemented in a distributed reinforcement learning based on SAC.
RL-SeqISP: Reinforcement Learning-Based Sequential Optimization for Image Signal Processing
Xinyu Sun (Beijing Jiaotong University), Yuxuan Guo
Object DetectionSegmentationOptimizationConvolutional Neural NetworkReinforcement LearningImage
🎯 What it does: A reinforcement learning-based sequence ISP parameter optimization model, RL-SeqISP, is proposed to automatically adjust the parameters of hardware ISP to enhance image quality and downstream task performance.
RLfOLD: Reinforcement Learning from Online Demonstrations in Urban Autonomous Driving
Daniel Coelho (University of Aveiro), Vitor Santos (University of Aveiro)
Autonomous DrivingReinforcement LearningImageBenchmark
🎯 What it does: An algorithm named RLfOLD is proposed, which trains driving strategies in the CARLA urban driving environment using online demonstrations combined with reinforcement learning.
Robust 3D Tracking with Quality-Aware Shape Completion
Jingwen Zhang (Harbin Institute of Technology), Wenjie Pei (Shenyang Institute of Automation, Chinese Academy of Sciences)
Object TrackingAutonomous DrivingConvolutional Neural NetworkPoint Cloud
🎯 What it does: Construct a synthetic target point cloud and generate a dense and complete target point cloud through quality-aware shape completion, followed by 3D single-object tracking in a voxel-based framework.
Robust Beamforming for Downlink Multi-Cell Systems: A Bilevel Optimization Perspective
Xingdi Chen (Tongji University), Kai Yang (Tongji University)
Optimization
🎯 What it does: A robust beamforming method based on bi-level optimization (BLRBF and BLADRBF) is proposed for maximizing the worst-case weighted sum rate in multi-cell multi-user MISO systems.
Robust Blind Text Image Deblurring via Maximum Consensus Framework
Zijian Min (Inha University), Geun-Sik Jo (Inha University)
RestorationImage
🎯 What it does: A robust blind text image deblurring method based on the maximum consensus framework is proposed, which can simultaneously remove noise and blur.
Robust Communicative Multi-Agent Reinforcement Learning with Active Defense
Lebin Yu (Tsinghua University), Jian Wang (Tsinghua University)
Adversarial AttackRecurrent Neural NetworkReinforcement LearningSequential
🎯 What it does: Designed and implemented the ADMAC framework, which actively assesses the reliability of received messages in multi-agent reinforcement learning and dynamically reduces the impact of malicious messages on decision-making;
Robust Distributed Gradient Aggregation Using Projections onto Gradient Manifolds
Kwang In Kim (POSTECH)
OptimizationFederated LearningImage
🎯 What it does: A method for robust distributed gradient aggregation using gradient manifold projection is proposed.
Robust Evaluation Measures for Evaluating Social Biases in Masked Language Models
Yang Liu (Tianjin University)
TransformerLarge Language ModelText
🎯 What it does: This paper proposes the use of Gaussian distribution to represent the pseudo-log-likelihood (PLL) score of PLM, and constructs two new bias assessment metrics, KLS and JSS, based on KL and JS divergence, aimed at measuring the social bias of masked language models (MLM).
Robust Few-Shot Named Entity Recognition with Boundary Discrimination and Correlation Purification
Xiaojun Xue (Beijing Institute of Technology), Zhendong Niu (Beijing Institute of Technology)
RecognitionAdversarial AttackTransformerContrastive LearningText
🎯 What it does: A robust two-stage few-shot named entity recognition method based on boundary discrimination and correlation purification, called BDCP, is proposed to resist text adversarial attacks.
Robust Loss Functions for Training Decision Trees with Noisy Labels
Jonathan Wilton (University of Queensland), Nan Ye (University of Queensland)
ClassificationOptimizationImageTabular
🎯 What it does: This study proposes a robust loss function for training decision trees with noisy labeled data, introducing a conservative loss and distribution loss framework, and designing a negative exponential (NE) loss based on this, achieving an analyzable impurity calculation.
Robust Node Classification on Graph Data with Graph and Label Noise
Yonghua Zhu (University of Auckland), Michael Witbrock (University of Auckland)
ClassificationGraph Neural NetworkContrastive LearningGraph
🎯 What it does: A robust node classification method RNCGLN is proposed, which addresses both graph structure noise and label noise.
Robust Nonparametric Regression under Poisoning Attack
Puning Zhao (Zhejiang Lab), Zhiguo Wan (Zhejiang Lab)
OptimizationAdversarial AttackTabular
🎯 What it does: The study focuses on robust non-parametric regression methods under adversarial poisoning attacks (with up to q samples being attacked).
Robust Policy Learning via Offline Skill Diffusion
Woo Kyung Kim (Sungkyunkwan University), Honguk Woo (Sungkyunkwan University)
Robotic IntelligenceReinforcement LearningDiffusion modelSequential
🎯 What it does: The DuSkill framework is proposed, utilizing offline skill diffusion to generate diverse skills to enhance cross-domain reinforcement learning performance.
Robust Test-Time Adaptation for Zero-Shot Prompt Tuning
Ding-Chu Zhang (Nanjing University), Yu-Feng Li (Nanjing University)
ClassificationDomain AdaptationTransformerPrompt EngineeringContrastive LearningImage
🎯 What it does: For zero-shot classification with the CLIP model, a robust adaptive framework called ADAPROMPT is proposed, which integrates manually designed multiple prompts and dynamically tunes the prompts on unlabeled test data to address data bias and model bias issues.
Robust Visual Imitation Learning with Inverse Dynamics Representations
Siyuan Li (Harbin Institute of Technology), Zhe Ma (Intelligent Science and Technology Academy Limited of CASIC)
Robotic IntelligenceReinforcement LearningImage
🎯 What it does: A robust visual imitation learning framework RILIR is proposed, which aligns the expert environment and the learning environment using inverse dynamics representation, and generates effective rewards by combining trajectory matching and discriminator rewards, achieving robustness against visual disturbances.
Robust Visual Recognition with Class-Imbalanced Open-World Noisy Data
Na Zhao (Singapore University of Technology and Design), Gim Hee Lee (National University of Singapore)
ClassificationRecognitionGraph Neural NetworkContrastive LearningImage
🎯 What it does: This paper proposes a robust visual recognition method in an environment with class imbalance and the coexistence of open-world noise, primarily addressing the issues of label noise correction and sample selection.
Robustly Improving Bandit Algorithms with Confounded and Selection Biased Offline Data: A Causal Approach
Wen Huang (University of Arkansas), Xintao Wu (University of Arkansas)
Recommendation SystemOptimizationReinforcement LearningTabular
🎯 What it does: The study investigates how to utilize causal inference methods to obtain prior bounds on the reward distribution of each arm in the presence of confounding and sampling biases in offline data, and how to embed these bounds into contextual and non-contextual bandit algorithms to achieve more robust online decision-making.
Robustly Train Normalizing Flows via KL Divergence Regularization
Kun Song (Mohamed bin Zayed University of Artificial Intelligence), Fakhri Karray (Mohamed bin Zayed University of Artificial Intelligence)
Anomaly DetectionFlow-based ModelImage
🎯 What it does: This paper proposes a training method based on KL divergence regularization, enhancing the robustness of regularization, especially for datasets with insufficient samples or the presence of outliers.
Robustness Verification of Deep Reinforcement Learning Based Control Systems Using Reward Martingales
Dapeng Zhi (East China Normal University), Min Zhang (East China Normal University)
Reinforcement LearningTabular
🎯 What it does: A robust verification method based on reward martingales is proposed, which can provide upper and lower bounds on expected cumulative rewards and tail probability bounds in deep reinforcement learning control systems affected by state disturbances.
Robustness-Guided Image Synthesis for Data-Free Quantization
Jianhong Bai (Zhejiang University), Haoji Hu
GenerationData SynthesisConvolutional Neural NetworkGenerative Adversarial NetworkImage
🎯 What it does: A robust-guided image synthesis method (RIS) is proposed for data-independent quantization;
ROG_PL: Robust Open-Set Graph Learning via Region-Based Prototype Learning
Qin Zhang (Shenzhen University), Junyang Chen (Shenzhen University)
ClassificationGraph Neural NetworkGraph
🎯 What it does: A robust open set node classification method for mixed IND noise and OOD noise in graph data is proposed (ROG PL), achieved through label propagation denoising and region-based prototype learning.
Roll with the Punches: Expansion and Shrinkage of Soft Label Selection for Semi-supervised Fine-Grained Learning
Yue Duan (Nanjing University), Yinghuan Shi (Nanjing University)
ClassificationConvolutional Neural NetworkContrastive LearningImage
🎯 What it does: This paper proposes a framework called SoC (Soft Label Selection with Confidence-Aware Clustering based on Class Transition Tracking) to address the challenges of high difficulty in identifying unlabeled data and significant noise in pseudo-labels in semi-supervised fine-grained visual classification (SS-FGVC);
Rolling-Unet: Revitalizing MLP’s Ability to Efficiently Extract Long-Distance Dependencies for Medical Image Segmentation
Yutong Liu (Beijing University of Chemical Technology), Jie Gao (Beijing University of Chemical Technology)
SegmentationConvolutional Neural NetworkImageBiomedical DataUltrasound
🎯 What it does: This paper proposes Rolling-Unet, which effectively captures long-range dependencies in medical images and improves segmentation accuracy by combining convolution with the Rolling-MLP module.
Root Cause Analysis in Microservice Using Neural Granger Causal Discovery
Cheng-Ming Lin (National Yang Ming Chiao Tung University), Wen-Chih Peng (National Yang Ming Chiao Tung University)
Anomaly DetectionContrastive LearningTime Series
🎯 What it does: A root cause analysis framework RUN based on neural Granger causality discovery and contrastive learning is proposed for root cause localization of anomalies in microservice systems.
Root Cause Explanation of Outliers under Noisy Mechanisms
Phuoc Nguyen (Deakin University), Svetha Venkatesh (Deakin University)
Anomaly DetectionExplainability and InterpretabilityComputational EfficiencyGraph
🎯 What it does: A noise functional causal model is proposed, attributing the root cause of leaf node anomalies to the noise of both nodes and edges, and a BIGEN attribution method based on Bayesian integral gradient is designed.
RoPDA: Robust Prompt-Based Data Augmentation for Low-Resource Named Entity Recognition
Sihan Song (Nanjing University), Jian Zhao (Nanjing University)
RecognitionTransformerSupervised Fine-TuningPrompt EngineeringText
🎯 What it does: This paper proposes RoPDA—a robust data augmentation method based on continuous prompts for low-resource named entity recognition.
RPSC: Robust Pseudo-Labeling for Semantic Clustering
Sihang Liu (South China University of Technology), Zhiwen Yu (Peng Cheng Laboratory)
Representation LearningContrastive LearningImage
🎯 What it does: A two-stage robust pseudo-label semantic clustering (RPSC) method is proposed, which first mines reliable semantic pseudo-labels through self-supervised learning, and then uses these pseudo-labels for semi-supervised learning to further enhance clustering performance.
RR-PU: A Synergistic Two-Stage Positive and Unlabeled Learning Framework for Robust Tax Evasion Detection
Shuzhi Cao (Xi'an Jiaotong University), Qinghua Zheng (Xi'an Jiaotong University)
ClassificationAnomaly DetectionTabularFinance Related
🎯 What it does: This paper proposes a two-stage positive and negative sample learning framework called RR-PU for detecting tax evasion with only a small number of labeled tax evaders and a large amount of unlabeled taxpayers' data.
RRL: Recommendation Reverse Learning
Xiaoyu You (Fudan University), Min Yang (Fudan University)
Recommendation SystemTabular
🎯 What it does: This paper proposes a recommendation system forgetting framework based on reverse learning, RRL, which can delete user-marked interaction records from the recommendation model without retraining the model, thus meeting the 'right to be forgotten' requirement.
Runtime Analysis of the (μ + 1) GA: Provable Speed-Ups from Strong Drift towards Diverse Populations
Benjamin Doerr (Ecole Polytechnique), Martin S. Krejca (Ecole Polytechnique)
OptimizationTabularBenchmark
🎯 What it does: A rigorous analysis of the running time of (µ+1) GA on the JUMP function.
Runtime Analysis of the SMS-EMOA for Many-Objective Optimization
Weijie Zheng (Harbin Institute of Technology), Benjamin Doerr (Institut Polytechnique de Paris)
OptimizationBenchmark
🎯 What it does: A rigorous theoretical analysis of the running time of SMS-EMOA in multi-objective optimization is conducted, proposing the m-objective OJZJ multimodal benchmark, and proving that the expected running time of the algorithm on this benchmark is O(M·n·2^k); upper bounds on the running time of the algorithm on the classic two-objective benchmarks ONEMINMAX and LOTZ are also provided; further, the impact of two techniques, random population update and heavy-tailed mutation, on running time is studied.
Runtime vs. Extracted Proof Size: An Exponential Gap for CDCL on QBFs
Olaf Beyersdorff (Friedrich Schiller University Jena), Meena Mahajan (Homi Bhabha National Institute)
🎯 What it does: This paper theoretically proves that there is an exponential gap between the running time and the proof size extracted from the solving trajectory in the Quantified Boolean Formula (QBF) Conflict-Driven Clause Learning (QCDCL) algorithm, revealing the inherent search overhead of QCDCL on certain instances.
RWMS: Reliable Weighted Multi-Phase for Semi-supervised Segmentation
Wensi Liu (Zhejiang University), Chunjie Yang (Zhejiang University)
SegmentationImage
🎯 What it does: A reliability-weighted multi-stage semi-supervised semantic segmentation method based on a dual-teacher model (RWMS) is proposed. It selects reliable samples for multi-stage training through image-level and pixel-level reliability assessments, re-weights pseudo-labels at the pixel level, and introduces strong data augmentation.
s-ID: Causal Effect Identification in a Sub-population
Amir Mohammad Abouei (École Polytechnique Fédérale de Lausanne), Negar Kiyavash (École Polytechnique Fédérale de Lausanne)
🎯 What it does: A new problem of identifying causal effects in subpopulations (subsamples affected by selection bias) is proposed (S-ID), and its necessary and sufficient conditions are provided.
S2CycleDiff: Spatial-Spectral-Bilateral Cycle-Diffusion Framework for Hyperspectral Image Super-resolution
Jiahui Qu (Xidian University), Jingyu Zhao (Xidian University)
RestorationSuper ResolutionDiffusion modelImage
🎯 What it does: A spatial-spectral bidirectional cyclic diffusion framework S CycleDiff is proposed for high-resolution hyperspectral image reconstruction.
S2WAT: Image Style Transfer via Hierarchical Vision Transformer Using Strips Window Attention
Chiyu Zhang (Nanjing University of Aeronautics and Astronautics), Jun Yang (Sichuan Normal University)
Image TranslationTransformerImage
🎯 What it does: This paper proposes an image style transfer framework S2WAT based on hierarchical visual Transformer, utilizing Strips Window Attention (SpW Attention) combined with multi-scale window attention and dynamically merging through Attn Merge, successfully addressing the localization (grid-like) problem caused by traditional window attention, achieving style transfer that captures both local and global features simultaneously.