arXivSub Start free trial

AAAI 2025 Papers with Code β€” Page 11

AAAI Conference on Artificial Intelligence Β· 1442 papers

Pioneer: Physics-informed Riemannian Graph ODE for Entropy-increasing Dynamics

Li Sun (North China Electric Power University), Philip S. Yu (University of Illinois at Chicago)

CodeGraph Neural NetworkGraphTime SeriesPhysics RelatedOrdinary Differential Equation

🎯 What it does: A physical information-based Riemannian graph ODE model called Pioneer is proposed to simulate dynamic interactive systems of entropy increase and achieve trajectory prediction on Riemannian manifolds.

PIXELS: Progressive Image Xemplar-based Editing with Latent Surgery

Shristi Das Biswas (Purdue University), Kaushik Roy (Amazon Fashion)

CodeImage TranslationGenerationDiffusion modelImageMultimodality

🎯 What it does: The PIXELS framework is proposed, enabling example-based image editing. This method utilizes a pre-trained text-image diffusion model during the inference phase, combined with user-defined pixel/region-level editing maps, to perform progressive editing of the source image with any number of example images, supporting multimodal prompts.

PlanLLM: Video Procedure Planning with Refinable Large Language Models

Dejie Yang (Peking University), Yang Liu (Peking University)

CodeSegmentationGenerationTransformerLarge Language ModelContrastive LearningVideoTextMultimodality

🎯 What it does: This paper proposes PlanLLM, a cross-modal joint learning framework that utilizes a trainable LLM for video program planning, capable of handling both closed-set single-step classification and open vocabulary free-text planning simultaneously.

Planning in the Dark: LLM-Symbolic Planning Pipeline Without Experts

Sukai Huang (University of Melbourne), Trevor Cohn (Google)

CodeTransformerLarge Language ModelText

🎯 What it does: A LLM-symbolic planning pipeline has been constructed that does not require expert intervention, capable of generating various executable action plans and automatically filtering and ranking them.

Plug-and-Play Tri-Branch Invertible Block for Image Rescaling

Jingwei Bao (University of Electronic Science and Technology of China), Shuyuan Zhu (Kuaishou Technology)

CodeRestorationSuper ResolutionCompressionFlow-based ModelImage

🎯 What it does: This paper proposes a three-branch invertible block (T-InvBlock) for image rescaling, which splits the low-frequency branch into luminance and chrominance channels, and uses a zero-high-frequency mapping during upsampling to achieve unified downsampling and upsampling.

PNVC: Towards Practical INR-based Video Compression

Ge Gao (University of Bristol), David Bull (University of Bristol)

CodeCompressionAuto EncoderOptical FlowVideo

🎯 What it does: A practical implicit neural representation (INR) video compression framework PNVC is proposed, which combines the foundation of autoencoders with overfitting techniques, supporting both low-latency and random access coding modes.

Point Cloud Mamba: Point Cloud Learning via State Space Model

Tao Zhang (Wuhan University), Xiangtai Li (Nanyang Technological University)

CodeRecognitionSegmentationTransformerPoint Cloud

🎯 What it does: A point cloud learning framework called Point Cloud Mamba is proposed, which utilizes consistent traversal serialization, sequential prompts, and spatial coordinate mapping position encoding to achieve global modeling of point clouds.

Point Cloud Semantic Segmentation with Sparse and Inhomogeneous Annotations

Zhiyi Pan (Peking University), Ge Li (Tencent)

CodeSegmentationPoint Cloud

🎯 What it does: In the weakly supervised point cloud semantic segmentation task, the AADNet model is proposed, which can adapt to any sparse annotation distribution, addressing the bias issue in gradient estimation caused by non-uniform sparse annotations.

PointDGMamba: Domain Generalization of Point Cloud Classification via Generalized State Space Model

Hao Yang (Shanghai Jiao Tong University), Shuicheng Yan (Skywork AI)

CodeClassificationDomain AdaptationPoint CloudBenchmark

🎯 What it does: A PointDGMamba framework based on the State Space Model (SSM) is proposed for domain generalization in point cloud classification, with core modules including Masked Sequence Denoising, Sequence-wise Cross-domain Feature Aggregation, and Dual-level Domain Scanning.

PokerBench: Training Large Language Models to Become Professional Poker Players

Richard Zhuang (University of California Berkeley), Gopala Anumanchipalli (University of California Berkeley)

CodeTransformerLarge Language ModelSupervised Fine-TuningReinforcement LearningTextBenchmarkFinance Related

🎯 What it does: This paper introduces a benchmark called POKERBENCH, designed to evaluate the decision-making capabilities of LLMs in Texas No-Limit Hold'em poker, and fine-tunes various LLMs on this benchmark to enhance poker performance.

POPoS: Improving Efficient and Robust Facial Landmark Detection with Parallel Optimal Position Search

Chong-Yang Xiang (Southwest Jiaotong University), Xian-Sheng Hua (Zhejiang University)

CodeRecognitionPose EstimationComputational EfficiencyConvolutional Neural NetworkImageBenchmark

🎯 What it does: A parallel optimal position search (POPoS) framework is proposed to enhance the accuracy and efficiency of facial keypoint detection.

Population Aware Diffusion for Time Series Generation

Yang Li (William and Mary), Haipeng Chen (William and Mary)

CodeGenerationData SynthesisTransformerDiffusion modelTime SeriesFinance Related

🎯 What it does: A time series generation framework based on diffusion models, PaD-TS, is proposed, focusing on preserving the overall properties of the original dataset (such as value distribution and functional dependency distribution).

PoseLLaVA: Pose Centric Multimodal LLM for Fine-Grained 3D Pose Manipulation

Dong Feng (Intel Labs), Peng Wang (Nanjing University of Posts and Telecommunications)

CodePose EstimationTransformerLarge Language ModelDiffusion modelContrastive LearningMultimodality

🎯 What it does: Design PoseLLaVA, which combines SMPL pose representation with a multimodal large language model to achieve language-based pose estimation, generation, and adjustment;

PoseMamba: Monocular 3D Human Pose Estimation with Bidirectional Global-Local Spatio-Temporal State Space Model

Yunlong Huang (Huazhong University of Science and Technology), Robert Caiming Qiu (Huazhong University of Science and Technology)

CodePose EstimationConvolutional Neural NetworkVideo

🎯 What it does: A pure state space model-based 3D human pose estimation framework called PoseMamba is proposed, utilizing bidirectional global-local spatiotemporal scanning to capture skeletal relationships and temporal dependencies.

PowerMLP: An Efficient Version of KAN

Ruichen Qiu (University of Chinese Academy of Sciences), Xiao-Shan Gao (University of Chinese Academy of Sciences)

CodeComputational EfficiencyImageText

🎯 What it does: PowerMLP is proposed, a ReLU power-based MLP network that replaces the spline activation function in KAN with a non-iterative approach, significantly improving training and inference speed while maintaining or exceeding the expressive power of KAN.

Practicable Black-Box Evasion Attacks on Link Prediction in Dynamic Graphsβ€”a Graph Sequential Embedding Method

Jiate Li (Nanchang University), Binghui Wang (Illinois Institute of Technology)

CodeAdversarial AttackRecurrent Neural NetworkGraph Neural NetworkReinforcement LearningGraph

🎯 What it does: A practical black-box evasion attack method for dynamic graph link prediction models is proposed, which can significantly reduce the prediction performance of the target model with only limited interactions and perturbations.

Practical Offloading for Fine-Tuning LLM on Commodity GPU via Learned Sparse Projectors

Siyuan Chen (Carnegie Mellon University), Phillip B. Gibbons (Carnegie Mellon University)

CodeTransformerLarge Language ModelSupervised Fine-TuningText

🎯 What it does: This paper proposes a framework called LSPOffload for large-scale language model fine-tuning on consumer-grade GPUs. It utilizes learned sparse projectors to achieve high-dimensional subspace updates within limited GPU memory, thereby avoiding memory overflow while maintaining performance close to the original model.

PRAGA: Prototype-aware Graph Adaptive Aggregation for Spatial Multi-modal Omics Analysis

Xinlei Huang (Great Bay University), Ziyue Qiao (Great Bay University)

CodeGraph Neural NetworkContrastive LearningMultimodalityBiomedical Data

🎯 What it does: The PRAGA framework is proposed, which uses a dynamically learnable omics-specific graph and dynamic prototype contrastive learning to achieve unified representation of multimodal spatial omics data.

Pre-Trained Vision-Language Models as Noisy Partial Annotators

Qian-Wei Wang (Tsinghua University), Shu-Tao Xia (Tsinghua University)

CodeClassificationKnowledge DistillationTransformerVision Language ModelContrastive LearningImage

🎯 What it does: This paper studies the use of the pre-trained vision-language model CLIP to automatically generate partial labels for downstream image classification tasks and proposes the Co-Reg method to train dedicated small models in the noise partial label learning (NPLL) scenario.

Pre-training a Density-Aware Pose Transformer for Robust LiDAR-based 3D Human Pose Estimation

Xiaoqi An (Nanjing University of Science and Technology), Jian Yang (Nanjing University of Science and Technology)

CodePose EstimationTransformerPoint Cloud

🎯 What it does: This study proposes a Density-Aware Pose Transformer (DAPT) and a comprehensive LiDAR human synthesis and occlusion enhancement scheme based on SMPL ray casting, achieving robust 3D human pose estimation from a single frame of LiDAR point cloud.

Pre-Training Graph Neural Networks on Molecules by Using Subgraph-Conditioned Graph Information Bottleneck

Van Thuy Hoang (Catholic University of Korea), O-Joun Lee (Catholic University of Korea)

CodeRepresentation LearningDrug DiscoveryGraph Neural NetworkContrastive LearningGraphBiomedical Data

🎯 What it does: A self-supervised pre-training method S-CGIB is proposed, which can automatically identify molecular core subgraphs and important subgraphs without the need for labels or prior knowledge of functional groups, and use them to generate molecular-level representations.

Predicting the Original Appearance of Damaged Historical Documents

Zhenhua Yang (South China University of Technology), Lianwen Jin (South China University of Technology)

CodeImage TranslationRestorationDiffusion modelImage

🎯 What it does: This paper proposes the Historical Document Restoration (HDR) task, constructs the HDR28K large-scale dataset, and introduces the DiffHDR diffusion network to achieve high-quality restoration of damaged images.

Preference-Oriented Supervised Fine-Tuning: Favoring Target Model over Aligned Large Language Models

Yuchen Fan (Zuoyebang Education Technology), Yang Song (Zuoyebang Education Technology)

CodeRecommendation SystemTransformerLarge Language ModelSupervised Fine-TuningText

🎯 What it does: A supervised fine-tuning method based on the Bradley–Terry preference model, called PoFT, is proposed to encourage the target model to outperform the aligned LLM on the same data.

PriFold: Biological Priors Improve RNA Secondary Structure Predictions

Chenchen Yang (Fudan University), Siqi Sun (Zelixir Biotech)

CodeProtein Structure PredictionTransformerSupervised Fine-TuningBiomedical Data

🎯 What it does: This paper proposes PriFold, which improves RNA secondary structure prediction using biological priors.

Prior-Constrained Association Learning for Fine-Grained Generalized Category Discovery

Menglin Wang (Nanjing Normal University), Xiaojin Gong (Zhejiang University)

CodeClassificationRecognitionTransformerContrastive LearningImage

🎯 What it does: Proposes a prior-constrained greedy association learning combined with non-parametric prototype comparison and parametric classification to achieve fine-grained general category discovery.

Privacy-and-Utility-Aware Publishing of Schedules

Maike Basmer (Humboldt-Universitat zu Berlin), Matthias Weidlich (Humboldt-Universitat zu Berlin)

CodeOptimizationSafty and PrivacyTabular

🎯 What it does: This paper studies the privacy leakage risks when releasing scheduling plans in parallel machine scheduling and proposes a mechanism for releasing schedules under the premise of meeting privacy and utility thresholds.

Privacy-Preserving Low-Rank Adaptation Against Membership Inference Attacks for Latent Diffusion Models

Zihao Luo (University of Auckland), Jingfeng Zhang (University of Auckland)

CodeGenerationOptimizationSafty and PrivacyDiffusion modelImage

🎯 What it does: Under privacy leakage attacks, a low-rank adaptation for privacy (MP-LoRA) and a stable version (SMP-LoRA) are proposed for secure fine-tuning on the Latent Diffusion Model, reducing the success rate of membership inference attacks.

Private Blotto: Viewpoint Competition with Polarized Agents

Kate Donahue (Cornell University), Jon Kleinberg (Cornell University)

Code

🎯 What it does: This paper proposes and analyzes the 'Private Blotto' game, studying how decentralized agents allocate limited effort across multiple projects and form pure Nash equilibria under two different aggregation functions (median and mean).

PrivDNFIS: Privacy-preserving and Efficient Deep Neuro-Fuzzy Inference System

Hao Ren (Sichuan University), Xingshu Chen (Sichuan University)

CodeSafty and PrivacyComputational EfficiencyReinforcement LearningImage

🎯 What it does: Developed PrivDNFIS, a privacy-preserving deep neural fuzzy inference system based on lattice-based homomorphic encryption;

Probability-Density-aware Semi-supervised Learning

Shuyang Liu (East China Normal University), Shaohui Lin (East China Normal University)

CodeClassificationSegmentationContrastive LearningImage

🎯 What it does: A probability density-aware semi-supervised learning method is proposed, which improves similarity measurement using density information and constructs a new label propagation algorithm called PMLP.

Procedure Knowledge Decoupled Distillation Strategy for Procedure Planning in Instructional Videos

Xiaotian Pan (Harbin Institute of Technology), Weigang Zhang (Harbin Institute of Technology)

CodeKnowledge DistillationTransformerDiffusion modelVideo

🎯 What it does: By allowing the teacher model to obtain real intermediate visual information and transferring its distributed knowledge to the student model through decoupled knowledge distillation (single action distillation and sequence distribution distillation), the program planning performance in teaching videos is improved.

Promising Multi-Granularity Linguistic Steganography by Jointing Syntactic and Lexical Manipulations

Chengfu Ou (Changsha University of Science and Technology), Yangfan Liu (Changsha University of Science and Technology)

CodeGenerationData SynthesisSafty and PrivacyTransformerLarge Language ModelText

🎯 What it does: A multi-granularity modified language steganography framework MMLS is proposed, which embeds the key into the syntactic space and symbolic space using both syntactic transformation and lexical replacement.

Prompt Compression with Context-Aware Sentence Encoding for Fast and Improved LLM Inference

Barys Liskavets (Alterra AI), Shane K. Luke (Workday Inc.)

CodeCompressionTransformerLarge Language ModelPrompt EngineeringContrastive LearningText

🎯 What it does: A sentence-level context-aware prompt compression (CPC) method is proposed, which uses a sentence encoder to evaluate the relevance of each sentence to the question, thereby trimming irrelevant sentences to shorten the input prompt.

Prompt Tuning In a Compact Attribute Space

Shiyu Hou (Beijing Institute of Technology), Guoren Wang (Beijing Institute of Technology)

CodeClassificationRecognitionExplainability and InterpretabilityTransformerLarge Language ModelPrompt EngineeringVision Language ModelImageMultimodality

🎯 What it does: A new prompt tuning method called PTinCAS is proposed, aimed at improving the performance of visual-language models through a compact attribute space for prompt tuning.

Prompt-based Unifying Inference Attack on Graph Neural Networks

Yuecen Wei (Beihang University), Chunming Hu (Guangxi Normal University)

CodeAdversarial AttackGraph Neural NetworkPrompt EngineeringContrastive LearningGraph

🎯 What it does: A unified prompt-based reasoning attack framework named ProIA has been designed and implemented for attribute inference and membership inference attacks on Graph Neural Networks (GNNs); it retains graph topology information through pre-training, generates attack samples using prompts, and incorporates a disentanglement mechanism in downstream tasks to enhance attack effectiveness.

Prompt-SID: Learning Structural Representation Prompt via Latent Diffusion for Single Image Denoising

Huaqiu Li (Tsinghua University), Haoqian Wang (Tsinghua University)

CodeRestorationTransformerPrompt EngineeringDiffusion modelImage

🎯 What it does: A self-supervised single image denoising framework based on prompt learning, Prompt-SID, is proposed, which utilizes structural representation generated diffusion (RG-Diff) and a Structural Attention Module (SAM) to preserve details and eliminate structural loss caused by sampling.

Promptable Anomaly Segmentation with SAM Through Self-Perception Tuning

Hui-Yue Yang (Tsinghua University), Guiguang Ding (Tsinghua University)

CodeSegmentationAnomaly DetectionSupervised Fine-TuningPrompt EngineeringImage

🎯 What it does: A Self-Perception Tuning (SPT) method is proposed, which enhances the perception and generalization capabilities of SAM in industrial defect segmentation tasks through two steps: self-sketch tuning and a visual relationship perception adapter.

Promptable Representation Distribution Learning and Data Augmentation for Gigapixel Histopathology WSI Analysis

Kunming Tang (Beihang University), Yushan Zheng (Beihang University)

CodeClassificationRepresentation LearningTransformerContrastive LearningBiomedical Data

🎯 What it does: This paper proposes the Promptable Representation Distribution Learning (PRDL) framework, which combines prompt-based representation distribution estimation with Promptable Representation Sampling (PRS) to achieve feature space data augmentation for pathological whole-slide images (WSI) and subsequently perform WSI classification.

PromptDet: A Lightweight 3D Object Detection Framework with LiDAR Prompts

Kun Guo (University of Science and Technology of China), Qiang Ling (University of Science and Technology of China)

CodeObject DetectionAutonomous DrivingPrompt EngineeringMultimodalityPoint Cloud

🎯 What it does: The PromptDet framework is proposed for multi-camera 3D detection, which injects LiDAR point clouds as prompts into a baseline camera detector using a small number of parameters, achieving lightweight multimodal fusion.

Proportional Representation in Practice: Quantifying Proportionality in Ordinal Elections

Tuva Bardal (University of Warwick), Jannik Peters (National University of Singapore)

CodeTabular

🎯 What it does: A quantifiable measure of proportionality is proposed, and the proportional representation of multi-vote methods is evaluated using real Scottish local election data.

ProPose: Probabilistic 3D Human Pose Estimation with Instance-Level Distribution and Normalizing Flow

Jumin Han (Korea University), Seong-Whan Lee (Korea University)

CodePose EstimationFlow-based ModelAuto EncoderImage

🎯 What it does: ProPose framework is proposed, implementing multi-hypothesis inference from 2D pose to 3D pose using instance-level Gaussian distribution and regularized flow.

ProsodyFM: Unsupervised Phrasing and Intonation Control for Intelligible Speech Synthesis

Xiangheng He (Imperial College London), BjΓΆrn Schuller (Technical University of Munich)

CodeGenerationData SynthesisTransformerFlow-based ModelAudio

🎯 What it does: An unsupervised ProsodyFM speech synthesis model is proposed, which can accurately control phrase breaks and terminal pitch, thereby enhancing intelligibility.

Protecting Model Adaptation from Trojans in the Unlabeled Data

Lijun Sheng (University of Science and Technology of China), Tieniu Tan (Chinese Academy of Sciences)

CodeClassificationDomain AdaptationAdversarial AttackImage

🎯 What it does: This study investigates the use of unlabelled target data to implant backdoor attacks during the model adaptation process and its defense methods.

ProtoOcc: Accurate, Efficient 3D Occupancy Prediction Using Dual Branch Encoder-Prototype Query Decoder

Jungho Kim (Seoul National University), Jun Won Choi (Hanyang University)

CodeSegmentationAutonomous DrivingConvolutional Neural NetworkTransformerPoint Cloud

🎯 What it does: ProtoOcc is a 3D occupancy prediction model based on a dual-branch encoder and a prototype query decoder.

Prototypical Calibrating Ambiguous Samples for Micro-Action Recognition

Kun Li (Hefei University of Technology), Meng Wang (Zhejiang University)

CodeRecognitionContrastive LearningVideoMultimodality

🎯 What it does: This paper proposes the Prototypical Calibrating Ambiguous Network (PCAN), which identifies and calibrates ambiguous samples in micro-action recognition through hierarchical prototype learning.

Provable Discriminative Hyperspherical Embedding for Out-of-Distribution Detection

Zhipeng Zou (Nanjing University of Science and Technology), Chen Gong (Shanghai Jiao Tong University)

CodeClassificationAnomaly DetectionConvolutional Neural NetworkContrastive LearningImage

🎯 What it does: The DHE (Provable Discriminative Hyperspherical Embedding) framework is proposed, which first maximizes the global class prototype distance through angular spread loss, and then uses prototype-enhanced contrastive (PEC) loss to bring ID samples closer to their corresponding prototypes, thereby obtaining a more discriminative embedding space for OOD detection.

Pruning Large Language Models with Semi-Structural Adaptive Sparse Training

Weiyu Huang (Tsinghua University), Jianfei Chen (Tsinghua University)

CodeCompressionComputational EfficiencyKnowledge DistillationTransformerLarge Language ModelText

🎯 What it does: The Adaptive Sparse Trainer (AST) framework is proposed, achieving efficient compression of large language models through semi-structured sparse training and knowledge distillation.

PScalpel: A Machine Learning-based Guider for Protein Phase-Separating Behaviour Alteration

Jia Wang (Shenzhen University), Jianqiang Li (Sun Yat-sen University)

CodeOptimizationProtein Structure PredictionGraph Neural NetworkContrastive LearningBiomedical Data

🎯 What it does: PScalpel is proposed, a tool that utilizes machine learning to assess and recommend changes in protein phase transition behavior.

PSMGD: Periodic Stochastic Multi-Gradient Descent for Fast Multi-Objective Optimization

Mingjing Xu (Rochester Institute of Technology), Haibo Yang (Rochester Institute of Technology)

CodeOptimizationGraph Neural NetworkTabular

🎯 What it does: A periodic updating dynamic weight multi-objective optimization algorithm PSMGD is proposed, which utilizes gradient fusion to quickly solve multi-objective problems.

Public Opinion Field Effect and Hawkes Process Join Hands for Information Popularity Prediction

Junliang Li (Tianjin University), Hong Gao (Zhejiang Normal University)

CodeRepresentation LearningRecurrent Neural NetworkGraph Neural NetworkTransformerTime SeriesSequential

🎯 What it does: A prediction model for information heat based on the Hawkes process that integrates public opinion field effects is proposed.

Putting People in LLMs’ Shoes: Generating Better Answers via Question Rewriter

Junhao Chen (Osaka University), Yuta Nakashima (Osaka University)

CodeGenerationOptimizationTransformerLarge Language ModelReinforcement LearningPrompt EngineeringText

🎯 What it does: Improving the answer quality of black-box LLMs in long-text question-answering tasks through question rewriting, addressing the issue of imprecise answers caused by vague user questions.

PVTree: Realistic and Controllable Palm Vein Generation for Recognition Tasks

Sheng Shang (Hefei University of Technology), Wei Jia (Hefei University of Technology)

CodeRecognitionGenerationData SynthesisConvolutional Neural NetworkGenerative Adversarial NetworkImage

🎯 What it does: Proposes the PVTree framework, which first generates a 3D vascular tree using an improved CCO algorithm, then projects it into multi-view vascular patterns, and subsequently converts it into realistic finger vein images through the improved PCE-Palm, achieving identity consistency and diversity;

QCS:Feature Refining from Quadruplet Cross Similarity for Facial Expression Recognition

Chengpeng Wang (Wisesoft Inc.), Xuebin Lv (Sichuan University)

CodeRecognitionKnowledge DistillationTransformerContrastive LearningImage

🎯 What it does: A four-branch symmetric network QCS is proposed, which refines facial expression features through cross-similarity attention, achieving suppression of redundant features.

QJL: 1-Bit Quantized JL Transform for KV Cache Quantization with Zero Overhead

Amir Zandieh (Google Research), Insu Han (KAIST)

CodeCompressionOptimizationLarge Language ModelText

🎯 What it does: A 1-bit quantization method QJL for LLM KV caching is proposed, providing an unbiased estimator without additional storage overhead.

QuARF: Quality-Adaptive Receptive Fields for Degraded Image Perception

Fei Gao (Xidian University), Nannan Wang (Xidian University)

CodeRestorationSegmentationGenerationConvolutional Neural NetworkGenerative Adversarial NetworkImage

🎯 What it does: A pluggable quality adaptive receptive field module QUARF is proposed, which can automatically select multi-scale convolution kernels based on the degradation level of the input image, thereby enhancing the perception and generation effects for low-quality images.

Queries, Representation & Detection: The Next 100 Model Fingerprinting Schemes

Augustin Godinot (University of Rennes), Gilles Tredan

CodeClassificationObject DetectionAnomaly DetectionImageBenchmark

🎯 What it does: The study investigates model fingerprinting techniques to detect cases of model theft and proposes a simple AKH baseline and QuRD framework to systematize fingerprint design.

Query-centric Audio-Visual Cognition Network for Moment Retrieval, Segmentation and Step-Captioning

Yunbin Tu (University of Chinese Academy of Sciences), Qingming Huang (University of Chinese Academy of Sciences)

CodeSegmentationRetrievalTransformerContrastive LearningVideoMultimodalityAudio

🎯 What it does: A query-centered audio-visual cognitive network QUAG is proposed for instant retrieval, segmentation, and step description in videos.

Queryable Prototype Multiple Instance Learning with Vision-Language Models for Incremental Whole Slide Image Classification

Jiaxiang Gou (University of Electronic Science and Technology of China), Mao Ye (University of Electronic Science and Technology of China)

CodeClassificationTransformerVision Language ModelImageBiomedical Data

🎯 What it does: A queryable prototype multi-instance learning framework based on a visual-language model (QPMIL-VL) is proposed, achieving bufferless incremental whole slide image classification.

Quickest Detection of Adversarial Attacks Against Correlated Equilibria

Kiarash Kazari (KTH Royal Institute of Technology), GyΓΆrgy DΓ‘n

CodeAnomaly DetectionOptimizationAdversarial AttackGraph

🎯 What it does: The study utilizes rapid change detection to timely identify and suppress situations where public signals are attacked under Cooperative Equilibrium (CE), and proposes a detection strategy based on generalized CUSUM.

R-DTI: Drug Target Interaction Prediction Based on Second-Order Relevance Exploration

Yang Hua (Jiangnan University), Xiaojun Wu (Jiangnan University)

CodeDrug DiscoveryGraph Neural NetworkBiomedical Data

🎯 What it does: This paper proposes a drug-target interaction prediction framework R-DTI based on second-order correlation exploration, combining protein structure and drug bimodal features, utilizing Riemannian space learning to enhance representation and make predictions based on second-order correlation.

RA-GAR: A Richly Annotated Benchmark for Gait Attribute Recognition

Chenye Wang (Beijing Normal University), Yongzhen Huang (Beijing Normal University)

CodeRecognitionTransformerContrastive LearningVideoBenchmark

🎯 What it does: A large-scale and attribute-rich gait attribute dataset RA-GAR (533 individuals, 120k+ sequences, 15 attributes) was collected and constructed, and a two-stage CLIP-GAR method was proposed for gait attribute recognition.

RaDIO: Real-Time Hallucination Detection with Contextual Index Optimized Query Formulation for Dynamic Retrieval Augmented Generation

Jia Zhu (Zhejiang Normal University), Pasquale De Meo (University of Messina)

CodeGenerationRetrievalTransformerLarge Language ModelTextRetrieval-Augmented Generation

🎯 What it does: This paper presents RaDIO, a dynamic retrieval-augmented generation framework that determines when to retrieve and optimizes queries using multi-head attention through real-time hallucination detection.

RAP-SR: RestorAtion Prior Enhancement in Diffusion Models for Realistic Image Super-Resolution

Jiangang Wang (Sun Yat-sen University), Wenqi Ren (Fuzhou University)

CodeRestorationSuper ResolutionDiffusion modelImage

🎯 What it does: The RAP-SR method is proposed, which achieves higher quality reconstruction of real image super-resolution by enhancing the recovery prior of the pre-trained diffusion model.

Rapid Learning in Constrained Minimax Games with Negative Momentum

Zijian Fang (Sun Yat-sen University), Chaohao Hu (Sun Yat-sen University)

CodeOptimizationReinforcement LearningTabular

🎯 What it does: This paper studies and implements the application of Negative Momentum in constrained minimax games, proposing a negative momentum update framework that can seamlessly integrate with classical algorithms (OMD, FTRL, RM+, etc.), and extends it to generalized limit games and extensive form games (EFG).

Rare Event Detection in Imbalanced Multi-Class Datasets Using an Optimal MIP-Based Ensemble Weighting Approach

Georgios Tertytchny (KIOS Research and Innovation Center of Excellence), Maria K. Michael (KIOS Research and Innovation Center of Excellence)

CodeAnomaly DetectionOptimizationTabular

🎯 What it does: This paper proposes a weighted voting ensemble method based on Mixed Integer Programming (MIP) combined with elastic net regularization for rare event detection in critical cyber-physical systems, aimed at selecting a predetermined number of classifiers and assigning class-level weights in imbalanced multi-class datasets.

RAT: Adversarial Attacks on Deep Reinforcement Agents for Targeted Behaviors

Fengshuo Bai (Shanghai Jiao Tong University), Yaodong Yang (Peking University)

CodeAdversarial AttackRobotic IntelligenceReinforcement LearningSequential

🎯 What it does: This paper proposes a general attack framework named RAT, which can achieve precise targeted attacks on the behavior of deep reinforcement learning (DRL) agents by making subtle perturbations to their observations without changing the target rewards.

RAZOR: Sharpening Knowledge by Cutting Bias with Unsupervised Text Rewriting

Shuo Yang (Technical University of Munich), Gjergji Kasneci (Technical University of Munich)

CodeTransformerLarge Language ModelText

🎯 What it does: The RAZOR method is proposed, utilizing large language models for unsupervised text rewriting to eliminate shortcut biases in the dataset and enhance the model's generalization ability.

RCTrans: Radar-Camera Transformer via Radar Densifier and Sequential Decoder for 3D Object Detection

Yiheng Li (Chinese Academy of Sciences), Zhen Lei (Chinese Academy of Sciences)

CodeObject DetectionAutonomous DrivingTransformerMultimodalityPoint Cloud

🎯 What it does: A query-based radar-camera Transformer framework RCTrans is proposed, which addresses the issue of sparse noise in radar point clouds through a radar dense encoder and a stepwise pruning decoder, achieving 3D object detection.

Re-Attentional Controllable Video Diffusion Editing

Yuanzhi Wang (Nanjing University of Science and Technology), Antoni B. Chan (City University of Hong Kong)

CodeGenerationData SynthesisDiffusion modelVideoText

🎯 What it does: This paper proposes a video diffusion editing framework based on re-attention, ReAtCo, which can precisely control the spatial position and quantity of foreground objects in a video through text prompts without training the model, while keeping the background unchanged.

Ready for You When You Are Back: Content-Driven Session-Based Recommendation for Continuity of Experience

Brijraj Singh (Sony Research India), Niranjan Pedanekar (Sony Research India)

CodeRecommendation SystemRecurrent Neural NetworkLarge Language ModelTextSequential

🎯 What it does: A conversation partitioning method based on content homogeneity is proposed to enhance the continuous experience and accuracy of conversational recommendation systems.

Real-Time Recurrent Reinforcement Learning

Julian Lemmel (Vienna University of Technology), Radu Grosu (Vienna University of Technology)

CodeMeta LearningRecurrent Neural NetworkReinforcement LearningSequential

🎯 What it does: A real-time recursive reinforcement learning framework (RTRRL) based on biological interpretability is proposed, which combines the Meta-RL structure of the basal ganglia, TD(λ) reinforcement learning, and RFLO/RTRL that can compute gradients online, to solve partially observable Markov decision process (POMDP) tasks.

RealisHuman: A Two-Stage Approach for Refining Malformed Human Parts in Generated Images

Benzhi Wang (Chinese Academy of Sciences), Zhen Lei (Chinese Academy of Sciences)

CodeRestorationGenerationPose EstimationDiffusion modelImage

🎯 What it does: In the human body images generated by diffusion models, a two-stage local reconstruction and seamless fusion is performed on deformed local areas such as hands and faces.

RealisID: Scale-Robust and Fine-Controllable Identity Customization via Local and Global Complementation

Zhaoyang Sun (Wuhan University of Technology), Shengwu Xiong (Interdisciplinary Artificial Intelligence Research Institute)

CodeImage TranslationGenerationPose EstimationDiffusion modelImage

🎯 What it does: This paper proposes the RealistID framework, which utilizes local and global branches to customize identity in Stable Diffusion, achieving fine-grained control over facial details, posture, expressions, and facial positioning, while maintaining high identity fidelity even with small face sizes.

Realistic Noise Synthesis with Diffusion Models

Qi Wu (Megvii Technology Inc), Shuaicheng Liu (University of Electronic Science and Technology of China)

CodeRestorationGenerationData SynthesisDiffusion modelImage

🎯 What it does: This paper proposes a real noise synthesis method based on diffusion models, RNSD, which generates RGB noise that conforms to different camera settings through conditional encoding, and utilizes the generated noise to enhance the performance of image denoising models.

Reasoning over Uncertain Text by Generative Large Language Models

Aliakbar Nafar (Michigan State University), Parisa Kordjamshidi (Michigan State University)

CodeTransformerLarge Language ModelPrompt EngineeringTextChain-of-Thought

🎯 What it does: This study investigates the challenges of LLMs in text reasoning that includes probabilistic information, proposes a new dataset called BLInD, and designs various prompting and symbolic methods to enhance reasoning capabilities.

Rebalancing Multi-Label Class-Incremental Learning

Kaile Du (Southeast University), Guangcan Liu (Southeast University)

CodeClassificationKnowledge DistillationConvolutional Neural NetworkImage

🎯 What it does: This paper addresses the issue of positive and negative sample imbalance in multi-label incremental learning and proposes the RebLL framework to achieve positive-negative balance.

Recording for Eyes, Not Echoing to Ears: Contextualized Spoken-to-Written Conversion of ASR Transcripts

Jiaqing Liu (Alibaba Group), Wen Wang (Alibaba Group)

CodeGenerationData SynthesisTransformerLarge Language ModelPrompt EngineeringTextBenchmarkRetrieval-Augmented Generation

🎯 What it does: This paper proposes the context-aware spoken-to-written (CoS2W) task, aimed at correcting recognition errors and grammatical mistakes in ASR transcriptions, as well as rewriting colloquial text into a formal written style. It also constructs a cross-lingual (Chinese-English) and cross-scenario (meetings, podcasts, lectures) SWAB dataset for training and evaluation with human-corrected targets.

Recoverable Compression: A Multimodal Vision Token Recovery Mechanism Guided by Text Information

Yi Chen (University of Chinese Academy of Sciences), Cheng-Lin Liu (University of Chinese Academy of Sciences)

CodeCompressionComputational EfficiencyTransformerLarge Language ModelVision Language ModelTextMultimodality

🎯 What it does: A training-independent multimodal visual token recovery mechanism is proposed, which dynamically filters and recovers important visual tokens using textual information, ultimately compressing the number of visual tokens to about 10% of the original without significant performance loss.

Reducing Divergence in Batch Normalization for Domain Adaptation

Ellen Yi-Ge (Carnegie Mellon University), Zhenghan Chen (Microsoft)

CodeDomain AdaptationConvolutional Neural NetworkImage

🎯 What it does: To address the estimation bias problem of Batch Normalization (BN) during training and inference in Unsupervised Domain Adaptation (UDA), we propose the Refined Batch Normalization (RBN) module, which replaces the BN layers in the bottleneck of residual networks with RBNBlock to reduce the accumulation of estimation bias caused by the stacking of BN layers, thereby improving cross-domain transfer performance.

ReFF: Reinforcing Format Faithfulness in Language Models Across Varied Tasks

Jiashu Yao (Beijing Institute of Technology), Yuhang Guo (Beijing Institute of Technology)

CodeReinforcement Learning from Human FeedbackTransformerLarge Language ModelSupervised Fine-TuningReinforcement LearningTextBenchmark

🎯 What it does: The FORMATBENCH benchmark is proposed to comprehensively evaluate the format fidelity of LLMs, and the REFF method is utilized to enhance the format adherence capability of LLMs within a reinforcement learning framework using a decidable format checker, while maintaining or improving overall quality.

Region-Based Optimization in Continual Learning for Audio Deepfake Detection

Yujie Chen (Anhui University), Xiaohui Zhang (Institute of Automation Chinese Academy of Sciences)

CodeAnomaly DetectionOptimizationBenchmarkAudio

🎯 What it does: A continuous learning method called Region-Based Optimization (RegO) is proposed for audio deepfake detection.

Regional Expected Improvement for Efficient Trust Region Selection in High-Dimensional Bayesian Optimization

Nobuo Namura (Fujitsu Limited), Sho Takemori (Fujitsu Limited)

CodeOptimizationTabular

🎯 What it does: This paper proposes a Regional Expectation Improvement (REI) and its variant qREI, aimed at better selecting trust regions in high-dimensional Bayesian optimization to enhance global search capabilities.

RegMixMatch: Optimizing Mixup Utilization in Semi-Supervised Learning

Haorong Han (Beijing Jiaotong University), Zhongyang Yu (Communication University of China)

CodeClassificationOptimizationTransformerContrastive LearningImage

🎯 What it does: Proposes the RegMixMatch framework, improving the use of Mixup in semi-supervised learning by jointly training with high-confidence cleaned samples and mixed samples, and further utilizing low-confidence samples through class-aware Mixup.

REGNav: Room Expert Guided Image-Goal Navigation

Pengna Li (Xi'an Jiaotong University), Sanping Zhou (Xi'an Jiaotong University)

CodeRobotic IntelligenceConvolutional Neural NetworkRecurrent Neural NetworkReinforcement LearningContrastive LearningImage

🎯 What it does: The REGNav model is proposed, utilizing the unsupervised training of Room Expert to help agents determine whether the target and the observed image are in the same room, thereby improving the efficiency of image target navigation.

Relation Also Knows: Rethinking the Recall and Editing of Factual Associations in Auto-Regressive Transformer Language Models

Xiyu Liu (Institute of Information Engineering, Chinese Academy of Sciences), Weiping Wang (University of Electronic Science and Technology of China)

CodeTransformerLarge Language ModelText

🎯 What it does: A single knowledge editing method based on relational information, RETS, is proposed, which utilizes the MLP sublayer of the autoregressive Transformer to modify weights at the last position of relational tags, significantly reducing the phenomenon of overgeneralization.

Relation-Aware Equivariant Graph Networks for Epitope-Unknown Antibody Design and Specificity Optimization

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

CodeOptimizationDrug DiscoveryGraph Neural NetworkGraphBiomedical Data

🎯 What it does: A relationship-aware equivariant graph network framework RAAD is proposed to generate antibody CDR sequences and structures in a single step under unknown display conditions, optimizing antibody specificity through comparative constraints.

Relational Neurosymbolic Markov Models

Lennert De Smet (KU Leuven), Giuseppe Marra (KU Leuven)

CodeGenerationData SynthesisOptimizationSequential

🎯 What it does: A new relational neural symbolic Markov model (NeSy-MMs) is proposed, which integrates deep sequence models with satisfiable relational logic constraints.

Relaxed Rotational Equivariance via G-Biases in Vision

Zhiqiang Wu (East China Normal University), Xian Wei

CodeClassificationObject DetectionConvolutional Neural NetworkImage

🎯 What it does: This paper proposes a Relaxed Rotational Equivariant Convolution method (RREConv) in the visual domain, which breaks the strict rotational symmetry by incorporating learnable G-Biases into Group Equivariant Convolution (GConv), constructing RRENet (for classification) and RREDet (for detection) networks;

Relieving Universal Label Noise for Unsupervised Visible-Infrared Person Re-Identification by Inferring from Neighbors

Xiao Teng (National University of Defense Technology), Nan Yin (National University of Defense Technology)

CodeRecognitionRetrievalGraph Neural NetworkContrastive LearningImageMultimodality

🎯 What it does: This paper proposes a method to alleviate global label noise in unsupervised visible-infrared person re-identification (USL-VI-ReID) using neighbor information, which mainly includes the Neighbor-guided Universal Label Calibration (N-ULC) and Neighbor-guided Dynamic Weighting (N-DW) modules.

REPEAT: Improving Uncertainty Estimation in Representation Learning Explainability

Kristoffer K. WickstrΓΈm (UiT The Arctic University of Norway), Robert Jenssen (University of Copenhagen)

CodeExplainability and InterpretabilityRepresentation LearningConvolutional Neural NetworkTransformerImage

🎯 What it does: A new unsupervised representation learning interpretability method called REPEAT is proposed, which can provide confidence estimates for pixel importance.

RepeatLeakage: Leak Prompts from Repeating as Large Language Model Is a Good Repeater

Yu Peng (Institute of Information Engineering, Chinese Academy of Sciences), Kai Chen (Institute of Information Engineering, Chinese Academy of Sciences)

CodeTransformerLarge Language ModelPrompt EngineeringText

🎯 What it does: It was found that LLMs primarily rely on context rather than model parameters in repetitive tasks. Based on this phenomenon, the RepeatLeakage method was proposed to achieve high-confidence single leakage of system prompts and dialogue context.

Replicating Electoral Success

Kiran Tomlinson (Cornell University), Jon Kleinberg (Cornell University)

Code

🎯 What it does: A candidate location replicator dynamics model based on the successful replicator strategy is proposed, and it is theoretically proven that when the number of candidates k ≀ 4, candidate locations will converge to the center, while for k β‰₯ 5, they will not.

Representing Sounds as Neural Amplitude Fields: A Benchmark of Coordinate-MLPs and a Fourier Kolmogorov-Arnold Framework

Linfei Li (Tongji University), Ying Shen (Shanghai Jiao Tong University)

CodeGenerationData SynthesisRepresentation LearningBenchmarkAudio

🎯 What it does: This paper first establishes a benchmark for audio implicit representation based on Coordinate-MLP, and then proposes the Fourier-ASR framework, which includes Fourier-KAN and a frequency adaptive learning strategy to achieve high-quality modeling of continuous audio signals.

ResAdapter: Domain Consistent Resolution Adapter for Diffusion Models

Jiaxiang Cheng (ByteDance Inc), Lean Fu (ByteDance Inc)

CodeGenerationData SynthesisDomain AdaptationDiffusion modelImage

🎯 What it does: Designed and trained a lightweight ResAdapter adapter, enabling any diffusion model to achieve resolution interpolation and extrapolation generation while maintaining the original style domain.

Resolving Multi-Condition Confusion for Finetuning-Free Personalized Image Generation

Qihan Huang (Zhejiang University), Jie Song (Alibaba Group)

CodeGenerationData SynthesisTransformerContrastive LearningImage

🎯 What it does: For personalized image generation with multiple reference images without fine-tuning, a weighted merge scheme is proposed to address the object confusion problem, and based on this, the model is further fine-tuned to improve generation quality.

Responsive Dynamic Graph Disentanglement for Metro Flow Forecasting

Qiang Gao (Southwestern University of Finance and Economics), Xueqin Chen (Kash Institute of Electronics and Information Industry)

CodeGraph Neural NetworkGraphTime Series

🎯 What it does: A responsive dynamic graph neural network without a fixed graph structure, ReDyNet, is proposed for accurate prediction of subway passenger flow.

RESQUE: Quantifying Estimator to Task and Distribution Shift for Sustainable Model Reusability

Vishwesh Sangarya (North Carolina State University), Jung-Eun Kim (North Carolina State University)

CodeClassificationDomain AdaptationComputational EfficiencyConvolutional Neural NetworkTransformerImage

🎯 What it does: A prediction quantifier named RESQUE is proposed to estimate the resource cost required for model retraining in the case of distribution shifts or task changes, and this metric can be obtained with a single forward pass before training.

Restabilizing Diffusion Models with Predictive Noise Fusion Strategy for Image Super-Resolution

Luoqian Jiang (South China University of Technology), Jian Chen (South China University of Technology)

CodeRestorationSuper ResolutionTransformerDiffusion modelImage

🎯 What it does: A Predictive Noise Fusion Strategy (PNFS) is proposed, which stabilizes the performance of diffusion models in image super-resolution tasks by predicting pixel-level errors and fusing different noises.

Rethinking Cancer Gene Identification Through Graph Anomaly Analysis

Yilong Zang (Hong Kong Polytechnic University), Junhang Wu (University of Queensland)

CodeRecognitionAnomaly DetectionDrug DiscoveryGraph Neural NetworkTransformerGraphBiomedical Data

🎯 What it does: This paper studies the weight heterogeneity of cancer genes in protein-protein interaction (PPI) networks, discovering that it leads to the abnormal feature of spectral energy 'flattening out'. Based on this, a hierarchical perspective graph neural network (HIPGNN) is proposed for cancer gene identification, utilizing dual perspectives of spectral feature encoding and spatial context decoding.

Rethinking High-speed Image Reconstruction Framework with Spike Camera

Kang Chen (Peking University), Zhaofei Yu (Peking University)

CodeRestorationSpiking Neural NetworkPrompt EngineeringContrastive LearningImage

🎯 What it does: Proposes the SpikeCLIP framework, which utilizes CLIP's text-image alignment as weak supervision to reconstruct high-quality images from low-light spike camera data.