ICLR 2025 Papers — Page 28
International Conference on Learning Representations · 3704 papers
Reassessing How to Compare and Improve the Calibration of Machine Learning Models
Muthu Chidambaram (Duke University), Rong Ge (Duke University)
OptimizationTransformerImage
🎯 What it does: This paper re-examines the evaluation and improvement methods for the calibration of machine learning models, pointing out that reporting only ECE and accuracy can lead to misjudgments. It proposes a calibration and generalization metric combined with Bregman divergence, as well as a new calibration-sharpness visualization diagram.
ReAttention: Training-Free Infinite Context with Finite Attention Scope
Xiaoran Liu (Fudan University), Xipeng Qiu (Fudan University)
TransformerLarge Language ModelTextRetrieval-Augmented Generation
🎯 What it does: A training-independent ReAttention method is proposed, which allows LLMs to handle infinite-length contexts under a limited attention window by performing position-independent top-k cache selection before self-attention.
REBIND: Enhancing Ground-state Molecular Conformation Prediction via Force-Based Graph Rewiring
Taewon Kim (Korea Advanced Institute of Science and Technology), Eunho Yang (Korea Advanced Institute of Science and Technology)
Drug DiscoveryGraph Neural NetworkTransformerGraph
🎯 What it does: This paper proposes a framework based on force-aware graph re-binding (REBIND) to predict the ground state three-dimensional conformation of molecules from two-dimensional molecular graphs.
RECAST: Reparameterized, Compact weight Adaptation for Sequential Tasks
Nazia Tasnim (Boston University), Bryan A. Plummer (Boston University)
ClassificationCompressionConvolutional Neural NetworkTransformerImage
🎯 What it does: The RECAST method is proposed, which re-parameterizes the weights of pre-trained models through shared templates and a small number of coefficients, achieving extremely high parameter compression and task-specific adaptation.
RecDreamer: Consistent Text-to-3D Generation via Uniform Score Distillation
Chenxi Zheng (South China University of Technology), Shengfeng He (Singapore Management University)
GenerationData SynthesisPose EstimationDiffusion modelScore-based ModelImageText
🎯 What it does: A text-to-3D generation method called RecDreamer is proposed, which eliminates the Multi-Face Janus geometric inconsistency problem through uniform distribution correction.
RecFlow: An Industrial Full Flow Recommendation Dataset
Qi Liu (University of Science and Technology of China), Kun Gai (Independent)
Recommendation SystemTransformerContrastive LearningVideo
🎯 What it does: The first industrial-grade end-to-end recommendation dataset, RecFlow, has been constructed and made public. It collects samples from six stages, including retrieval and ranking, and experiments have been conducted on retrieval, coarse ranking, and ranking stages using this dataset.
Recite, Reconstruct, Recollect: Memorization in LMs as a Multifaceted Phenomenon
USVSN Sai Prashanth (EleutherAI), Naomi Saphra (Harvard University)
TransformerLarge Language ModelText
🎯 What it does: This study explores the multifaceted nature of language model memory, proposing three intuitive classification methods: recitation, reconstruction, and recollection, and uses them to predict memory probabilities.
ReCogLab: a framework testing relational reasoning & cognitive hypotheses on LLMs
Andrew Liu (Google Deepmind), Kenneth Marino (Google Deepmind)
GenerationData SynthesisTransformerLarge Language ModelPrompt EngineeringTextChain-of-Thought
🎯 What it does: A configurable generative framework, ReCogLab, is proposed for automatically constructing relationship reasoning tasks with adjustable difficulty, and systematic experiments are conducted on various models.
Recognize Any Surgical Object: Unleashing the Power of Weakly-Supervised Data
Jiajie Li (University at Buffalo), Jinjun Xiong (University at Buffalo)
RecognitionObject DetectionTransformerLarge Language ModelVision Language ModelImageVideoText
🎯 What it does: Developed the RASO foundational model to achieve open surgical object recognition, and realized high performance under zero-shot and supervised conditions through weakly supervised data generation.
Reconciling Model Multiplicity for Downstream Decision Making
Ally Yalei Du (Carnegie Mellon University), Steven Wu
ClassificationOptimizationConvolutional Neural NetworkReinforcement LearningImage
🎯 What it does: An algorithm named ReDCal is proposed, which utilizes multi-calibration techniques to harmonize multiple predictive models that have similar accuracy but differ in decision-making for downstream decision tasks, making them nearly consistent in individual predictions and optimal response actions.
Reconsidering Faithfulness in Regular, Self-Explainable and Domain Invariant GNNs
Steve Azzolin (University of Trento), Andrea Passerini (University of Trento)
Explainability and InterpretabilityGraph Neural NetworkGraph
🎯 What it does: Systematically evaluate the authenticity metrics of graph neural network explanations, theoretically analyze their limitations, and study the implementation and impact of interpretability under different architectures (regular, interpretable, domain-invariant), proposing improvement methods and validating their effectiveness.
Reconstruction-Guided Policy: Enhancing Decision-Making through Agent-Wise State Consistency
Liang Qifan, Yuan Tian (Jilin University)
Recurrent Neural NetworkReinforcement LearningDiffusion modelTabular
🎯 What it does: A Reconstruction-Guided Policy (RGP) framework is proposed, which constructs agent-wise states using a decision module and a guidance module, maintaining state consistency during training and execution to enhance decision-making performance in multi-agent reinforcement learning under partially observable environments.
Reconstructive Visual Instruction Tuning
Haochen Wang (Chinese Academy of Sciences), Zhaoxiang Zhang (Chinese Academy of Sciences)
Image TranslationRestorationGenerationTransformerLarge Language ModelVision Language ModelDiffusion modelImageMultimodality
🎯 What it does: A constructive visual instruction tuning (ROSS) is proposed, which supervises visual output by having multimodal models reconstruct input images.
Recovering Manifold Structure Using Ollivier Ricci Curvature
Tristan Luca Saidi (Columbia University), Andrew J. Blumberg (Yale University)
Point CloudBiomedical Data
🎯 What it does: The ORC-MANL algorithm is proposed, which identifies and prunes 'shortcut' edges in the nearest neighbor graph by combining Ollivier-Ricci curvature with a graph distance threshold.
Recovery of Causal Graph Involving Latent Variables via Homologous Surrogates
Xiu-Chuan Li (Sydney AI Centre, University of Sydney), Tongliang Liu (Sydney AI Centre, University of Sydney)
GraphTabular
🎯 What it does: Proposes the concept of 'homologous substitute variables' to eliminate the reliance on the pure child assumption in traditional causal discovery, allowing for partial or complete recovery of causal graphs with latent variables under weaker or stronger conditions;
Rectified Diffusion: Straightness Is Not Your Need in Rectified Flow
Fu-Yun Wang (Chinese University of Hong Kong), Hongsheng Li (Chinese University of Hong Kong)
GenerationData SynthesisOptimizationKnowledge DistillationDiffusion modelRectified FlowImageOrdinary Differential Equation
🎯 What it does: The Rectified Diffusion method is proposed, simplifying and generalizing the idea of Rectified Flow. It supports any diffusion model, staged training, and consistency distillation by using pre-collected noise-sample pairs for retraining.
ReDeEP: Detecting Hallucination in Retrieval-Augmented Generation via Mechanistic Interpretability
ZhongXiang Sun, Han Li (Kuaishou Technology Co., Ltd.)
RetrievalExplainability and InterpretabilityTransformerLarge Language ModelTextRetrieval-Augmented Generation
🎯 What it does: This paper proposes a mechanism-interpretable RAG hallucination detection method called ReDeEP, and based on this, designs AARF to mitigate hallucinations.
Redefining the task of Bioactivity Prediction
Yanwen Huang (Peking University), Yanyan Lan (Tsinghua University)
Drug DiscoveryConvolutional Neural NetworkGraph Neural NetworkBiomedical Data
🎯 What it does: Redefine the bioactivity prediction task, construct a large-scale structural small molecule-protein interaction dataset SIU, and propose Pearson/Spearman evaluation metrics grouped by protein pockets.
Reducing Hallucinations in Large Vision-Language Models via Latent Space Steering
Sheng Liu (Stanford University), James Zou (Stanford University)
Object DetectionGenerationTransformerVision Language ModelImageTextMultimodality
🎯 What it does: A method for intervention in the latent space during inference (VTI) is proposed in large-scale vision-language models, which reduces hallucination generation by applying pre-computed offset vectors to the visual and text hidden layers.
REEF: Representation Encoding Fingerprints for Large Language Models
Jie Zhang (Shanghai Artificial Intelligence Laboratory), Jing Shao (Shanghai Artificial Intelligence Laboratory)
Representation LearningTransformerLarge Language ModelText
🎯 What it does: A no-training fingerprint method based on representation layer similarity, REEF, is proposed to identify whether an open-source LLM is a subsequent development of the target model.
RefactorBench: Evaluating Stateful Reasoning in Language Agents Through Code
Dhruv Gautam (University of California Berkeley), Roshanak Zilouchian Moghaddam (Microsoft)
AI Code AssistantTransformerLarge Language ModelAgentic AITextBenchmark
🎯 What it does: A benchmark called RefactorBench is proposed to evaluate the reasoning and state-awareness capabilities of language model agents in multi-file code refactoring.
Refine Knowledge of Large Language Models via Adaptive Contrastive Learning
Yinghui Li (Tsinghua University), Philip S. Yu (University of Illinois Chicago)
TransformerLarge Language ModelSupervised Fine-TuningContrastive LearningText
🎯 What it does: Proposes the Adaptive Contrastive Learning (ACL) method, which fine-tunes large language models using knowledge boundaries (I Know and I Don't Know thresholds) to enhance their ability to provide honest answers and reduce hallucination issues.
Refine-by-Align: Reference-Guided Artifacts Refinement through Semantic Alignment
Yizhi Song (Purdue University), Daniel Aliaga (Adobe Research)
Image TranslationRestorationDiffusion modelImageBenchmark
🎯 What it does: A reference image-based local defect (artifact) automatic repair framework called Refine-by-Align is proposed, which employs a two-stage process of alignment followed by repair. This framework can correct local defects in the generated image using details from the reference image while keeping the original background unchanged.
REFINE: Inversion-Free Backdoor Defense via Model Reprogramming
Yukun Chen (Zhejiang University), Kui Ren (Zhejiang University)
OptimizationAdversarial AttackConvolutional Neural NetworkAuto EncoderContrastive LearningImage
🎯 What it does: In response to backdoor attacks on third-party deep learning models, a backdoor defense method called REFINE is proposed, which is based on model reprogramming and does not require reverse engineering of triggers.
Refining CLIP's Spatial Awareness: A Visual-Centric Perspective
Congpei Qiu (Xi'an Jiaotong University), Tong Zhang (University of Chinese Academy of Sciences)
SegmentationKnowledge DistillationRepresentation LearningTransformerContrastive LearningImage
🎯 What it does: A spatially correlated distillation framework (SCD) and a lightweight Refiner are proposed to maintain the spatial awareness of CLIP ViT during the Region-Language Alignment (RLA) process, enhancing the performance of open vocabulary dense prediction.
Reflective Gaussian Splatting
Yuxuan Yao (Fudan University), Li Zhang (Fudan University)
Gaussian SplattingImage
🎯 What it does: This paper studies a real-time high-quality neural rendering framework suitable for reflective objects—Ref-Gaussian.
Reflexive Guidance: Improving OoDD in Vision-Language Models via Self-Guided Image-Adaptive Concept Generation
Jihyo Kim (Seoul National University of Science and Technology), Sangheum Hwang (Seoul National University of Science and Technology)
Anomaly DetectionTransformerPrompt EngineeringVision Language ModelImageMultimodalityBenchmark
🎯 What it does: This study investigates the performance of visual language large models in out-of-distribution detection (OoDD) and proposes a self-guided prompting (ReGuide) method to enhance their OoDD capabilities.
Reframing Structure-Based Drug Design Model Evaluation via Metrics Correlated to Practical Needs
Bowen Gao (Tsinghua University), Yanyan Lan (Beijing Academy of Artificial Intelligence)
Drug DiscoveryBiomedical Data
🎯 What it does: A framework for evaluating structure-based drug design (SBDD) models based on practical needs is proposed, and various mainstream generative models are systematically evaluated within this framework.
ReGen: Generative Robot Simulation via Inverse Design
Phat Tan Nguyen (Massachusetts Institute of Technology), Daniela Rus (Massachusetts Institute of Technology)
GenerationAutonomous DrivingRobotic IntelligenceTransformerLarge Language ModelReinforcement LearningVision-Language-Action ModelMultimodality
🎯 What it does: Using reverse design and large language models to automatically generate situational graphs and convert them into symbolic programs, thereby inferring and constructing diverse, controllable simulation environments based on robot behaviors (trajectories or objective functions).
ReGenesis: LLMs can Grow into Reasoning Generalists via Self-Improvement
XIANGYU PENG, Chen Xing (Salesforce AI Research)
TransformerLarge Language ModelSupervised Fine-TuningText
🎯 What it does: Utilize the LLM's own generated reasoning paths as post-training data to enhance its general reasoning ability.
REGENT: A Retrieval-Augmented Generalist Agent That Can Act In-Context in New Environments
Kaustubh Sridhar (University of Pennsylvania), Insup Lee (University of Pennsylvania)
RetrievalRobotic IntelligenceReinforcement Learning from Human FeedbackTransformerAgentic AISequentialRetrieval-Augmented Generation
🎯 What it does: A semi-parametric retrieval-enhanced general agent REGENT is proposed, which can quickly adapt to new environments through retrieval context without fine-tuning;
RegMix: Data Mixture as Regression for Language Model Pre-training
Qian Liu (Sea AI Lab), Min Lin (Sea AI Lab)
TransformerLarge Language ModelText
🎯 What it does: Proposes the REGMIX method, which automatically identifies suitable data mixing schemes for large-scale language model pre-training;
Regressing the Relative Future: Efficient Policy Optimization for Multi-turn RLHF
Zhaolin Gao (Cornell University), Wen Sun (Princeton University)
OptimizationReinforcement Learning from Human FeedbackTransformerLarge Language ModelReinforcement LearningText
🎯 What it does: This paper proposes a multi-round RLHF method called REFUEL, which utilizes a single model to regress relative future rewards, achieving online alignment and addressing the covariance shift problem encountered in traditional single-round RLHF in multi-round dialogues.
Regret Bounds for Episodic Risk-Sensitive Linear Quadratic Regulator
Wenhao XU, Xuedong He (Chinese University of Hong Kong)
Reinforcement LearningTime SeriesFinance Related
🎯 What it does: Two online adaptive control algorithms based on least squares are proposed for risk-sensitive linear quadratic regulation (LEQR) in a finite time discrete-time domain, with analysis and corresponding upper bounds provided.
Regret-Optimal List Replicable Bandit Learning: Matching Upper and Lower Bounds
Michael Chen (Iowa State University), Lin Yang
OptimizationReinforcement Learning
🎯 What it does: This paper studies list replicability in multi-armed bandits (MAB) and proposes a new algorithm A that can maintain a limited number of execution trajectories with high probability under different random conditions.
Regretful Decisions under Label Noise
Sujay Nagaraj (University of Toronto), Berk Ustun (University of California San Diego)
Anomaly DetectionOptimizationData-Centric LearningBiomedical DataElectronic Health Records
🎯 What it does: This paper studies the 'error lottery' problem that may arise in individual predictions when training machine learning models on datasets with label noise. It introduces the concept of 'Regret' to measure the unpredictable errors caused by noise and proposes using plausible noise sampling to train models to estimate individual-level regret probabilities (Ambiguity), thereby achieving data cleaning, selective prediction, and safe inference in scientific discovery.
Regularization by Texts for Latent Diffusion Inverse Solvers
Jeongsol Kim (Korea Advanced Institute of Science and Technology), Jong Chul Ye (Korea Advanced Institute of Science and Technology)
RestorationSuper ResolutionDiffusion modelImage
🎯 What it does: This paper proposes a latent diffusion inverse solver TReg based on text regularization, which can embed text descriptions into the latent space during the inverse problem-solving process. By using adaptive negation constraints, it reduces ambiguity in the solution space, resulting in reconstructions that are more semantically aligned and consistent with measurements.
Regularizing Energy among Training Samples for Out-of-Distribution Generalization
Yiting Chen (Shanghai Jiao Tong University), Junchi Yan (Shanghai Jiao Tong University)
ClassificationDomain AdaptationImageBenchmark
🎯 What it does: This paper proposes regularizing the energy distribution among training samples to enhance the model's robustness in out-of-distribution (OOD) tasks.
Regulatory DNA Sequence Design with Reinforcement Learning
Zhao Yang (Renmin University of China), Ji-Rong Wen (Renmin University of China)
OptimizationReinforcement LearningBiomedical Data
🎯 What it does: A design method for cis-regulatory elements (CRE) called TACO is proposed, which combines reinforcement learning and autoregressive DNA generation models, capable of generating highly functional and diverse promoter and enhancer sequences from scratch.
Reinforcement Learning for Control of Non-Markovian Cellular Population Dynamics
Josiah C Kratz, Jacob Adamczyk (University of Massachusetts)
Drug DiscoveryReinforcement LearningBiomedical Data
🎯 What it does: The study uses reinforcement learning to control drug dosage in non-Markovian cell populations, proposing a memory-enabled model to obtain the optimal strategy.
Reinforcement Learning from Imperfect Corrective Actions and Proxy Rewards
Zhaohui JIANG, Changjie Fan (NetEase)
Reinforcement Learning from Human FeedbackReinforcement LearningSequential
🎯 What it does: This paper studies a novel reinforcement learning framework called ICoPro, which utilizes dual signals of imperfect agent rewards and human corrective actions to enhance the alignment of agents with human preferences and sample efficiency.
Reinforcement learning with combinatorial actions for coupled restless bandits
Lily Xu (Harvard University), Milind Tambe (Harvard University)
OptimizationReinforcement Learning
🎯 What it does: A reinforcement learning algorithm named SEQUOIA is proposed, which combines deep Q-networks with mixed-integer programming to directly search for the maximum expected return in the combinatorial action space at each step.
Relation-Aware Diffusion for Heterogeneous Graphs with Partially Observed Features
Daeho Um (Samsung Electronics), Seong Jin Ahn (KAIST)
Graph Neural NetworkGraphBiomedical Data
🎯 What it does: The HetGFD method is proposed for feature diffusion-based missing value imputation in heterogeneous graphs by utilizing virtual features and edge type heterogeneity.
Relax and Merge: A Simple Yet Effective Framework for Solving Fair $k$-Means and $k$-sparse Wasserstein Barycenter Problems
Shihong Song (University of Science and Technology of China), Hu Ding (University of Science and Technology of China)
OptimizationTabular
🎯 What it does: This paper proposes the 'Relax and Merge' framework to address the issues of fair k-means and k-sparse Wasserstein Barycenter;
Relaxed Recursive Transformers: Effective Parameter Sharing with Layer-wise LoRA
Sangmin Bae (KAIST), Tal Schuster (Google DeepMind)
CompressionComputational EfficiencyKnowledge DistillationTransformerLarge Language ModelSupervised Fine-TuningText
🎯 What it does: This paper proposes the Recursive Transformer, which compresses large language models by reusing layers through parameter sharing. It further introduces the Relaxed Recursive Transformer, which incorporates low-rank LoRA modules into the shared layers to achieve flexible parameter relaxation. Additionally, it presents an inference scheme that combines Continuous Depth-wise Batching with early exiting.
RelCon: Relative Contrastive Learning for a Motion Foundation Model for Wearable Data
Maxwell A Xu, Shirley You Ren
ClassificationRecognitionRepresentation LearningConvolutional Neural NetworkContrastive LearningTime Series
🎯 What it does: The RelCon framework is proposed, which uses a learnable motif distance and relative contrastive loss for self-supervised pre-training on wearable accelerometer data to build a foundational model for movement.
Release the Powers of Prompt Tuning: Cross-Modality Prompt Transfer
Ningyuan Zhang (Australian Artificial Intelligence Institute, University of Technology Sydney), Guangquan Zhang (Australian Artificial Intelligence Institute, University of Technology Sydney)
ClassificationDomain AdaptationTransformerPrompt EngineeringImageText
🎯 What it does: This study explores Cross-Modality Prompt Transfer, which transfers prompts pre-trained in natural language to visual tasks to enhance prompt tuning performance in data-scarce tasks.
Reliable and Diverse Evaluation of LLM Medical Knowledge Mastery
Yuxuan Zhou (Tsinghua University), Ji Wu (Tsinghua University)
TransformerLarge Language ModelPrompt EngineeringTextBiomedical Data
🎯 What it does: The PretexEval framework is proposed, which dynamically generates reliable and diverse evaluation samples from medical knowledge bases using predicate equivalence transformation and prototype paraphrasing techniques, aimed at assessing large language models' mastery of medical factual knowledge.
RelitLRM: Generative Relightable Radiance for Large Reconstruction Models
Tianyuan Zhang (Massachusetts Institute of Technology), Fujun Luan (Adobe Research)
GenerationData SynthesisTransformerDiffusion modelGaussian SplattingImage
🎯 What it does: Utilizing Transformer and diffusion models, we end-to-end generate 3D Gaussian Splatting volumes that can be re-lit under arbitrary lighting conditions with only 4-8 sparse camera images;
ReMatching Dynamic Reconstruction Flow
Sara Oblak (NVIDIA), Matan Atzmon (NVIDIA)
RestorationOptimizationGaussian SplattingVideoOrdinary Differential Equation
🎯 What it does: A framework named ReMatching is proposed to integrate velocity field priors into models for dynamic scene reconstruction, thereby enhancing generalization capabilities to unknown viewpoints and time points while maintaining high fidelity.
REMEDY: Recipe Merging Dynamics in Large Vision-Language Models
Didi Zhu (Zhejiang University), Chao Wu (East China Normal University)
OptimizationTransformerLarge Language ModelVision Language ModelImageTextMultimodality
🎯 What it does: The REMEDY method is proposed, which constructs reusable recipes (projectors and low-rank adaptation of shallow LLMs) and uses a modality-aware allocator to dynamically fuse during inference, achieving multi-task learning and zero-shot generalization for large-scale visual language models.
ReMoE: Fully Differentiable Mixture-of-Experts with ReLU Routing
Ziteng Wang (Tsinghua University), Jianfei Chen (Tsinghua University)
Mixture of ExpertsText
🎯 What it does: This paper proposes ReMoE, a fully differentiable Mixture-of-Experts model using ReLU routing, serving as a direct alternative to TopK+Softmax routing.
Remove Symmetries to Control Model Expressivity and Improve Optimization
Liu Ziyin (Massachusetts Institute of Technology), Isaac L. Chuang (Massachusetts Institute of Technology)
OptimizationReinforcement LearningImage
🎯 What it does: By adding random static bias to the loss function and combining it with weight decay, the reflection symmetry in the network is eliminated, avoiding low capacity traps and improving optimization results.
ReNovo: Retrieval-Based \emph{De Novo} Mass Spectrometry Peptide Sequencing
Shaorong Chen (Zhejiang University), Stan Z. Li (Westlake University)
RetrievalTransformerBiomedical Data
🎯 What it does: A retrieval-based de novo peptide sequencing method called ReNovo is proposed, which constructs a retrieval library using training data and retrieves contextual information during the inference phase to enhance prediction accuracy.
Repetition Improves Language Model Embeddings
Jacob Mitchell Springer (Carnegie Mellon University), Aditi Raghunathan (Carnegie Mellon University)
Representation LearningTransformerLarge Language ModelPrompt EngineeringText
🎯 What it does: Utilizing autoregressive language models (such as Mistral, LLaMA) to extract text embeddings from the form of input text that appears twice (i.e., 'echo'), allowing the model to see the complete context during the second occurrence, thereby achieving an effect similar to bidirectional attention.
RepoGraph: Enhancing AI Software Engineering with Repository-level Code Graph
Siru Ouyang (University of Illinois), Dong Yu (Tencent)
AI Code AssistantGraph Neural NetworkLarge Language ModelPrompt EngineeringTextGraphRetrieval-Augmented Generation
🎯 What it does: The REPOGRAPH plugin module is proposed to construct a repository-level code graph (line-level dependency graph) and provide it as a subgraph for retrieval to LLMs, enhancing the effectiveness of AI software engineering tasks.
Representation Alignment for Generation: Training Diffusion Transformers Is Easier Than You Think
Sihyun Yu (Korea Advanced Institute of Science and Technology), Saining Xie (New York University)
GenerationData SynthesisTransformerDiffusion modelContrastive LearningImage
🎯 What it does: This paper studies a regularization method called REPA, which aligns the clear image features of a pre-trained self-supervised visual encoder to the internal representations of a diffusion transformer, thereby improving generation quality and training efficiency.
Representational Similarity via Interpretable Visual Concepts
Neehar Kondapaneni (California Institute of Technology), Pietro Perona (California Institute of Technology)
Explainability and InterpretabilityRepresentation LearningConvolutional Neural NetworkTransformerImage
🎯 What it does: An interpretable representation similarity method RSVC is proposed, which compares the representations of two networks using visual concepts and reveals the specific reasons for similarities and differences.
Representative Guidance: Diffusion Model Sampling with Coherence
Anh-Dung Dinh (University of Sydney), Chang Xu (University of Sydney)
GenerationData SynthesisDiffusion modelContrastive LearningImage
🎯 What it does: Proposes Representative Guidance to address the inconsistency in the diffusion sampling process.
Repulsive Latent Score Distillation for Solving Inverse Problems
Nicolas Zilberstein (Rice University), Santiago Segarra (Rice University)
RestorationGenerationDiffusion modelImage
🎯 What it does: A variational posterior sampler is constructed using a pre-trained latent diffusion model to address mode collapse and latent space inversion issues in inverse problems.
RESfM: Robust Deep Equivariant Structure from Motion
Fadi Khatib (Weizmann Institute of Science), Ronen Basri (Weizmann Institute of Science)
Pose EstimationOptimizationSimultaneous Localization and MappingPoint Cloud
🎯 What it does: A robust multi-view structure from motion (SfM) method based on set-set equivariant networks is proposed, which can simultaneously recover camera poses and 3D scene structures, and handles numerous outliers generated by common matching heuristics through the incorporation of a multi-view inlier/outlier classification module and robust bundle adjustment.
ReSi: A Comprehensive Benchmark for Representational Similarity Measures
Max Klabunde (University of Passau), Florian Lemmerich (University of Passau)
Representation LearningGraph Neural NetworkImageTextGraphBenchmark
🎯 What it does: A comprehensive benchmark for representation similarity measurement, ReSi, has been constructed, which includes six designed tests, 24 types of similarity measures, 14 different domain neural network models (graph, language, visual), and seven publicly available datasets, with open code for community use.
Residual Connections and Normalization Can Provably Prevent Oversmoothing in GNNs
Michael Scholkemper (RWTH Aachen University), Michael T Schaub
ClassificationGraph Neural NetworkGraph
🎯 What it does: This paper theoretically analyzes and proves that residual connections and normalization layers can fundamentally prevent over-smoothing in GNNs, and proposes an improved normalization layer, GraphNormv2, which further avoids the detrimental effects of centering on graph signals.
Residual Deep Gaussian Processes on Manifolds
Kacper Wyrwal (ETH Zurich), Viacheslav Borovitskiy (ETH Zurich)
Gaussian SplattingTabular
🎯 What it does: A new model for implementing residual deep Gaussian processes on Riemannian manifolds is proposed, along with corresponding Bayesian inference methods.
Residual Kernel Policy Network: Enhancing Stability and Robustness in RKHS-Based Reinforcement Learning
Yixian Zhang (Tsinghua University), Wenbo Ding (Tsinghua University)
Reinforcement LearningSequential
🎯 What it does: This paper proposes and implements the Residual Kernel Policy Network (ResKPN), which significantly enhances the stability and robustness of RL training by incorporating representation learning and residual layers into the RKHS policy.
Residual Stream Analysis with Multi-Layer SAEs
Tim Lawson (University of Bristol), Laurence Aitchison (University of Bristol)
TransformerAuto Encoder
🎯 What it does: A multi-layer sparse autoencoder (MLSAE) trained on the residual flow of all layers of the Transformer is proposed and trained to analyze the flow of information between layers.
Residual-MPPI: Online Policy Customization for Continuous Control
Pengcheng Wang (University of California), Wei Zhan (University of California)
OptimizationRobotic IntelligenceReinforcement LearningSequential
🎯 What it does: An online planning algorithm called Residual-MPPI is proposed, which can adaptively customize pre-trained continuous control policies during execution without the need for retraining.
Resolution Attack: Exploiting Image Compression to Deceive Deep Neural Networks
Wangjia Yu (Institute of Information Engineering), Xiaodan Zhang (Institute of Information Engineering)
ClassificationGenerationAdversarial AttackDiffusion modelImage
🎯 What it does: A 'resolution attack' framework based on diffusion models is proposed and implemented, capable of automatically generating images that display one semantic at high resolution and another semantic at low resolution, thereby misleading classifiers and human observers;
Restructuring Vector Quantization with the Rotation Trick
Christopher Fifty (Stanford University), Christopher Re
GenerationData SynthesisAuto EncoderGenerative Adversarial NetworkImage
🎯 What it does: A new gradient propagation method called the Rotation Trick is proposed in VQ-VAE. This method involves rotating and scaling the encoder output with the nearest codebook vector during forward propagation, while keeping the angle between the two unchanged during backpropagation, allowing the gradient to bypass the non-differentiable vector quantization layer without losing information.
Restyling Unsupervised Concept Based Interpretable Networks with Generative Models
Jayneel Parekh (ISIR Sorbonne Université), Florence d'Alché-Buc (ISIR Sorbonne Université)
GenerationExplainability and InterpretabilityConvolutional Neural NetworkAuto EncoderGenerative Adversarial NetworkImage
🎯 What it does: The VisCoIN architecture is proposed, which achieves high-quality visualization and interpretation of unsupervised concept networks by mapping concept vectors to the latent space of a pre-trained generative model.
RESuM: A Rare Event Surrogate Model for Physics Detector Design
Ann-Kathrin Schuetz (Lawrence Berkeley National Laboratory), Aobo Li (University of California San Diego)
OptimizationTabularPhysics Related
🎯 What it does: Proposed and implemented the RESuM rare event surrogate model to optimize the neutron shield design in the LEGEND experiment, significantly reducing the background rate.
Rethinking and Improving Autoformalization: Towards a Faithful Metric and a Dependency Retrieval-based Approach
Qi Liu (Shanghai Jiao Tong University), Junchi Yan (Shanghai Jiao Tong University)
Large Language ModelTextRetrieval-Augmented Generation
🎯 What it does: This paper proposes a credible automatic evaluation metric BEq and a retrieval-enhanced statement automation method RAutoformalizer, addressing the issues of traditional evaluations being imprecise and lacking contextual information.
Rethinking Artistic Copyright Infringements In the Era Of Text-to-Image Generative Models
Mazda Moayeri (University of Maryland), Soheil Feizi (University of Maryland)
ClassificationExplainability and InterpretabilityTransformerContrastive LearningImage
🎯 What it does: The ArtSavant tool is constructed to automatically identify the unique styles of artists and detect whether such styles appear in works generated by text-to-image models, thereby providing quantifiable and interpretable evidence for art style infringement.
Rethinking Audio-Visual Adversarial Vulnerability from Temporal and Modality Perspectives
Zeliang Zhang (University of Rochester), Chenliang Xu (University of Rochester)
Adversarial AttackVideoMultimodalityAudio
🎯 What it does: Two novel attack methods for audio-video multimodal models are proposed (temporal invariant attack and modality mismatch attack), and an efficient adversarial training framework is designed to enhance robustness.
Rethinking Classifier Re-Training in Long-Tailed Recognition: Label Over-Smooth Can Balance
Siyu Sun (Shanghai Jiao Tong University), Junchi Yan (Shanghai Jiao Tong University)
ClassificationRecognitionImage
🎯 What it does: This paper reevaluates classifier retraining methods in long-tail recognition and proposes new evaluation metrics and a Label Over-Smoothing (LOS) method based on logits magnitude balancing.
Rethinking Diffusion Posterior Sampling: From Conditional Score Estimator to Maximizing a Posterior
Tongda Xu (Tsinghua University), Yan Wang (Tsinghua University)
RestorationSuper ResolutionDiffusion modelImage
🎯 What it does: This paper studies the performance of Diffusion Posterior Sampling (DPS) in high-resolution image inverse problems, finding that it is not a conditional score estimator as previously theorized, but rather a process closer to maximizing the posterior (MAP). Based on this, two improvement schemes are proposed: 1) using multi-step gradient ascent and projection to achieve explicit MAP maximization; 2) employing a lightweight conditional score estimator (CSE) to enhance initialization.
Rethinking Evaluation of Sparse Autoencoders through the Representation of Polysemous Words
Gouki Minegishi (University of Tokyo), Yutaka Matsuo (University of Tokyo)
Representation LearningAuto EncoderText
🎯 What it does: This paper proposes Poly-Semantic Evaluation (PS-Eval) to assess whether Sparse Autoencoders (SAE) can decompose the polysemous activations of LLMs into monosemantic features.
Rethinking Fair Representation Learning for Performance-Sensitive Tasks
Charles Jones (Imperial College London), Ben Glocker (Imperial College London)
Representation LearningConvolutional Neural NetworkBiomedical DataMagnetic Resonance Imaging
🎯 What it does: This paper systematically explores the effectiveness of Fair Representation Learning (FRL) under different fairness analysis paradigms, proposes a unified description of data bias mechanisms from a causal perspective, and validates the performance of FRL in IID and distribution shift scenarios through theoretical proofs and large-scale medical imaging experiments.
Rethinking Graph Neural Networks From A Geometric Perspective Of Node Features
Feng Ji (Nanyang Technological University), Wee Peng Tay (Nanyang Technological University)
Graph Neural NetworkGraph
🎯 What it does: By constructing the centroid simplex of node features and applying rough geometric analysis, this paper proposes an explanation for the heterogeneity, oversmoothing, feature reshuffling, and other phenomena of GNNs from the perspective of features, and provides simple graph-independent improvement techniques.
Rethinking Invariance in In-context Learning
Lizhe Fang (Peking University), Yisen Wang (Peking University)
TransformerLarge Language ModelSupervised Fine-TuningText
🎯 What it does: This work proposes a new method for Invariant In-context Learning (InvICL), aimed at addressing the issue of large language models being sensitive to the order of context examples during ICL, while achieving model order invariance under the premise of information leakage prevention and context interdependence.
Rethinking Invariance Regularization in Adversarial Training to Improve Robustness-Accuracy Trade-off
Futa Kai Waseda (University of Tokyo), Isao Echizen (National Institute of Informatics)
OptimizationAdversarial AttackConvolutional Neural NetworkAuto EncoderImage
🎯 What it does: A new adversarial training method called AR-AT is proposed to reduce the robustness-accuracy trade-off;
Rethinking Light Decoder-based Solvers for Vehicle Routing Problems
Ziwei Huang (Singapore Management University), Yixin XU
OptimizationTransformerMixture of ExpertsTabular
🎯 What it does: This paper proposes ReLD, a method for improving the solution of the Vehicle Routing Problem (VRP) by enhancing a lightweight decoder neural solver.
Rethinking LLM Unlearning Objectives: A Gradient Perspective and Go Beyond
Qizhou Wang (Hong Kong Baptist University), Kilian Q Weinberger
TransformerLarge Language ModelText
🎯 What it does: This paper proposes a tool for evaluating the learning objectives of LLMs from the perspective of gradients—G-effect. Based on this, it analyzes existing learning methods and proposes improved new objectives such as WGA, TNPO, and WTNPO. Experiments are conducted on the TOFU dataset with various LLMs, demonstrating their advantages in removing target knowledge while maintaining model integrity.
Rethinking Multiple-Instance Learning From Feature Space to Probability Space
Zhaolong Du (Xidian University), Licheng Jiao (Xidian University)
Representation LearningConvolutional Neural NetworkImageBenchmark
🎯 What it does: A multi-instance learning framework called PSMIL is proposed and implemented, which aligns and pools in probability space, addressing the issue of instance representation drift in traditional feature space.
Rethinking Neural Multi-Objective Combinatorial Optimization via Neat Weight Embedding
Jinbiao Chen (Sun Yat-sen University), Yue-Jiao Gong (South China University of Technology)
OptimizationTransformerReinforcement LearningGraph
🎯 What it does: This paper proposes a 'weight embedding' method that directly learns weight-specific representations within a single-objective network, thereby efficiently solving multi-objective combinatorial optimization problems.
Rethinking Reward Model Evaluation: Are We Barking up the Wrong Tree?
Xueru Wen (Chinese Information Processing Laboratory, Institute of Software, Chinese Academy of Sciences), Le Sun (Chinese Information Processing Laboratory, Institute of Software, Chinese Academy of Sciences)
OptimizationReinforcement Learning from Human FeedbackTransformerSupervised Fine-TuningReinforcement LearningPrompt EngineeringText
🎯 What it does: This study evaluates the Reward Model (RM) and explores the relationship between RM accuracy and downstream policy regret.
Rethinking Reward Modeling in Preference-based Large Language Model Alignment
Hao Sun (University of Cambridge), Jean-Francois Ton (ByteDance Research)
Recommendation SystemTransformerLarge Language ModelReinforcement LearningText
🎯 What it does: This paper conducts a systematic study of reward modeling methods in aligning large language models (LLMs), comparing the traditional Bradley-Terry (BT) model with classification-based reward models, and exploring the impact of cross-prompt preference annotations on model performance.
Rethinking Self-Distillation: Label Averaging and Enhanced Soft Label Refinement with Partial Labels
Hyeonsu Jeong (Korea Advanced Institute of Science and Technology), Hye Won Chung (Korea Advanced Institute of Science and Technology)
Knowledge DistillationConvolutional Neural NetworkImage
🎯 What it does: This study investigates a multi-class self-distillation mechanism under a fixed feature extractor (linear probing) and proposes a partial label distillation method that can replicate the effects of multi-round distillation with just a single round.
Rethinking Shapley Value for Negative Interactions in Non-convex Games
Wonjoon Chang (Korea Advanced Institute of Science and Technology), Jaesik Choi (Korea Advanced Institute of Science and Technology)
Convolutional Neural NetworkTransformerImageText
🎯 What it does: This paper studies the interaction issues of Shapley values in non-convex games and proposes the Aggregated Positive Interaction (API) method to correct the underestimation of attribution caused by negative interactions, while also providing an efficient computational algorithm for gradient approximation.
Rethinking Spiking Neural Networks from an Ensemble Learning Perspective
Yongqi Ding (University of Electronic Science and Technology of China), Hanpu Deng (University of Electronic Science and Technology of China)
ClassificationKnowledge DistillationSpiking Neural NetworkImagePoint CloudAudio
🎯 What it does: Treating spiking neural networks (SNNs) as a collection of sub-networks with shared architectures and parameters, this paper addresses the output instability caused by differences in membrane potential distribution between sub-networks by proposing membrane potential smoothing and temporary neighboring sub-network guidance techniques, significantly enhancing the overall performance of SNNs.
Rethinking the generalization of drug target affinity prediction algorithms via similarity aware evaluation
Chenbin Zhang (MoleculeMind), Shaoting Zhang (Shanghai AI Laboratory)
Drug DiscoveryConvolutional Neural NetworkRecurrent Neural NetworkGraph Neural NetworkTransformerBiomedical Data
🎯 What it does: This paper proposes a similarity-based evaluation framework (SAE) that can partition the test set of drug-target affinity prediction models according to the expected similarity distribution (such as uniform or simulating external data distribution), thereby more accurately assessing the model's performance on low-similarity samples.
Rethinking the role of frames for SE(3)-invariant crystal structure modeling
Yusei Ito (OMRON SINIC X Corporation), Kanta Ono (OMRON SINIC X Corporation)
TransformerGraph
🎯 What it does: Proposed the concept of dynamic frames and implemented the CrystalFramer model, improving the SE(3) invariant encoding of crystal structures.
Rethinking Visual Counterfactual Explanations Through Region Constraint
Bartlomiej Sobieski (University of Warsaw), Przemyslaw Biecek (University of Warsaw)
Image TranslationExplainability and InterpretabilityScore-based ModelImage
🎯 What it does: This paper proposes Region-Constrained Visual Counterfactual Explanations (RVCE) and its generation method RCSB, which can modify images within a specified region to cause changes in classifier predictions, thereby providing more interpretable counterfactuals.
Reti-Diff: Illumination Degradation Image Restoration with Retinex-based Latent Diffusion Model
Chunming He (Duke University), Sina Farsiu (Duke University)
RestorationObject DetectionSegmentationTransformerDiffusion modelImage
🎯 What it does: A method called Reti-Diff, which combines a latent diffusion model based on Retinex theory with Transformer, is proposed to achieve high-quality image enhancement and color correction in various degradation scenarios such as low light, underwater, and backlight.
Retri3D: 3D Neural Graphics Representation Retrieval
Yushi Guan (University of Toronto), Nandita Vijaykumar (Intel)
RetrievalVision Language ModelPoint CloudMesh
🎯 What it does: The Retri3D framework is proposed, which enables the retrieval of pre-trained 3D neural graphic representations (NGR) from large-scale storage based on text queries, compatible with any NGR format.
Retrieval Augmented Diffusion Model for Structure-informed Antibody Design and Optimization
Zichen Wang (Global Institute of Future Technology), Shuangjia Zheng (Global Institute of Future Technology)
OptimizationDrug DiscoveryDiffusion modelBiomedical DataRetrieval-Augmented Generation
🎯 What it does: A retrieval-enhanced diffusion framework called RADAb is proposed, which guides antibody sequence generation using structurally homologous fragments, achieving antibody design and optimization under structural constraints.
Retrieval Head Mechanistically Explains Long-Context Factuality
Wenhao Wu (Peking University), Yao Fu (University of Edinburgh)
RetrievalTransformerLarge Language ModelText
🎯 What it does: This paper studies the internal mechanisms of Transformer long-context language models and proposes and examines a class of attention heads specifically responsible for information retrieval, referred to as retrieval heads.
RetroInText: A Multimodal Large Language Model Enhanced Framework for Retrosynthetic Planning via In-Context Representation Learning
Chenglong Kang (Central South University), Fei Guo (Central South University)
Representation LearningDrug DiscoveryTransformerLarge Language ModelPrompt EngineeringTextMultimodality
🎯 What it does: This paper presents RetroInText, a multimodal large language model framework that achieves more accurate single-step and multi-step retrosynthetic predictions by using ChatGPT to generate text descriptions in multi-step retrosynthetic pathways and integrating them with molecular 2D/3D representations.
Reveal Object in Lensless Photography via Region Gaze and Amplification
Yin Xiangjun, Huihui Yue (Nanyang Technological University)
Object DetectionTransformerImageBenchmark
🎯 What it does: This paper proposes RGANet, specifically designed for detecting hidden objects in lensless camera measurements, and introduces the benchmark dataset for this task for the first time.
Revealing and Mitigating Over-Attention in Knowledge Editing
Pinzheng Wang (Soochow University), Min Zhang (Soochow University)
TransformerLarge Language ModelText
🎯 What it does: This paper studies the issue of specificity failure that arises after knowledge editing in large language models and proposes a new regularization method called SADR to mitigate this problem.