NeurIPS 2024 Papers — Page 31
Conference on Neural Information Processing Systems · 4035 papers
Reinforcement Learning Policy as Macro Regulator Rather than Macro Placer
Ke Xue (Nanjing University), Chao Qian (Huawei Noah's Ark Lab)
OptimizationReinforcement LearningTabularBenchmark
🎯 What it does: This paper proposes a macro regulator method called MaskRegulate based on reinforcement learning, which makes local adjustments on an existing macro layout instead of starting from scratch.
Reinforcement Learning Under Latent Dynamics: Toward Statistical and Algorithmic Modularity
Philip Amortila (University of Illinois), Zakaria Mhammedi (Google)
Reinforcement Learning
🎯 What it does: A general framework is proposed to study reinforcement learning under latent dynamics, exploring the modularity of statistics and algorithms.
Reinforcement Learning with Adaptive Regularization for Safe Control of Critical Systems
Haozhe Tian (Imperial College London), Pietro Ferraro (Imperial College London)
Safty and PrivacyReinforcement LearningSequential
🎯 What it does: The RL-AR algorithm is proposed, which achieves safe control of critical systems by introducing a safety regularizer (MPC) and an adaptive weight combination of the RL policy during the reinforcement learning process.
Reinforcement Learning with Euclidean Data Augmentation for State-Based Continuous Control
Jinzhu Luo (University of South Carolina), Qi Zhang (University of South Carolina)
Robotic IntelligenceReinforcement Learning
🎯 What it does: A state space data augmentation method based on Euclidean symmetry is proposed, using body-based state representation instead;
Reinforcement Learning with Lookahead Information
Nadav Merlis (FairPlay Joint Team, CREST, ENSAE Paris)
Reinforcement Learning
🎯 What it does: This study investigates the scenario in reinforcement learning where immediate rewards or transition information can be observed before taking actions, and presents algorithms that can utilize this information.
Reinforcement Learning with LTL and $\omega$-Regular Objectives via Optimality-Preserving Translation to Average Rewards
Xuan-Bach Le (Nanyang Technological University), Luke Ong (Nanyang Technological University)
OptimizationReinforcement Learning
🎯 What it does: This paper presents an optimality-preserving translation that converts ω-regular (including LTL) objectives in reinforcement learning into finite memory reward machines, and based on this, implements an algorithm that can learn optimal policies in the limit.
Reinforcing LLM Agents via Policy Optimization with Action Decomposition
Muning Wen (Shanghai Jiao Tong University), Ying Wen (Shanghai Jiao Tong University)
OptimizationReinforcement Learning from Human FeedbackTransformerLarge Language ModelSupervised Fine-TuningReinforcement LearningText
🎯 What it does: Proposes Bellman Backup with Action Decomposition (BAD) and combines it with PPO to form POAD, assigning rewards to each intra-action token of the language agent at a finer granularity;
Rejection via Learning Density Ratios
Alexander Soen (Amazon), Vu Nguyen (Amazon)
ClassificationAnomaly DetectionOptimizationImage
🎯 What it does: A post-hoc rejection framework based on the idealized distribution density ratio is proposed, which determines when to reject predictions by learning an idealized distribution regularized by α-divergence that is aligned with the original data distribution.
Relating Hopfield Networks to Episodic Control
Hugo Chateau-Laurent (Inria centre of the University of Bordeaux), Frederic Alexandre
RetrievalReinforcement LearningImage
🎯 What it does: This paper proves that the Differentiable Neural Dictionary (DND) used in reinforcement learning is essentially an instance of the Universal Hopfield Network (UHN), and derives the energy function through continuous approximation; it then evaluates its performance in terms of capacity and retrieval on MNIST, CIFAR-10, and Tiny ImageNet.
Relational Concept Bottleneck Models
Pietro Barbiero (Università della Svizzera Italiana), Giuseppe Marra (KU Leuven)
Explainability and InterpretabilityGraph Neural NetworkGraph
🎯 What it does: An interpretable relational concept bottleneck model (R-CBM) is proposed, which combines concept bottlenecks with graph neural networks to achieve reasoning and explanation for multi-entity relational tasks.
Relational Verification Leaps Forward with RABBit
Tarun Suresh (University of Illinois Urbana-Champaign), Gagandeep Singh (VMware Research)
OptimizationAdversarial AttackConvolutional Neural NetworkImage
🎯 What it does: RABBit is proposed, a branch-and-bound based relational DNN verifier designed to assess the general robustness against adversarial perturbations.
Relationship Prompt Learning is Enough for Open-Vocabulary Semantic Segmentation
Jiahaoli, Yanyun Qu (Xiamen University)
SegmentationTransformerPrompt EngineeringVision Language ModelImage
🎯 What it does: This paper proposes a framework RPN for directly implementing open vocabulary semantic segmentation (OVSS) based on relational prompts (RPM), without the need for any explicit segmentation network.
Reliable Learning of Halfspaces under Gaussian Marginals
Ilias Diakonikolas (University of Wisconsin Madison), Nikos Zarifis (University of Wisconsin Madison)
Optimization
🎯 What it does: Under the Gaussian marginal distribution, a new reliable learning (reliable PAC) algorithm for half-spaces is proposed, which significantly outperforms traditional reliable learning and global reliable learning algorithms in terms of sample and computational complexity.
ReLIZO: Sample Reusable Linear Interpolation-based Zeroth-order Optimization
Xiaoxing Wang (Shanghai Jiao Tong University), Junchi Yan (Shanghai Jiao Tong University)
OptimizationAdversarial AttackNeural Architecture SearchImageText
🎯 What it does: A ReLIZO zero-order optimization method is proposed, which reduces the number of queries and computational complexity while maintaining a constant sample size by estimating gradients through linear interpolation and allowing the reuse of previous queries.
ReMAP: Neural Model Reprogramming with Network Inversion and Retrieval-Augmented Mapping for Adaptive Motion Forecasting
Sharmita Dey (University Medical Center Göttingen), Sarath Ravindran Nair (Georg-August University of Göttingen)
Pose EstimationDomain AdaptationRobotic IntelligenceTime SeriesRetrieval-Augmented Generation
🎯 What it does: Through model reprogramming, a gait prediction model based on healthy individuals is transferred to patients with lower limb disabilities to achieve joint motion prediction.
Remix-DiT: Mixing Diffusion Transformers for Multi-Expert Denoising
Gongfan Fang (National University of Singapore), Xinchao Wang (National University of Singapore)
RestorationGenerationTransformerMixture of ExpertsDiffusion modelImage
🎯 What it does: A multi-expert denoising method called Remix-DiT is designed, which generates N denoising experts by mixing K benchmark Transformer models to reduce training costs and improve image generation quality.
ReMoDetect: Reward Models Recognize Aligned LLM's Generations
Hyunseok Lee (Korea Advanced Institute of Science and Technology), Jinwoo Shin (Korea Advanced Institute of Science and Technology)
RecognitionGenerationTransformerLarge Language ModelReinforcement LearningText
🎯 What it does: A reward model-based alignment framework for detecting LLM-generated text, ReMoDetect, is proposed to distinguish between LLM-generated text and human text.
Remove that Square Root: A New Efficient Scale-Invariant Version of AdaGrad
Sayantan Choudhury (Mohammed Bin Zayed University of Artificial Intelligence), Eduard Gorbunov (Mohammed Bin Zayed University of Artificial Intelligence)
OptimizationImageTabular
🎯 What it does: KATE is proposed—a variant of AdaGrad that removes the square root and achieves scale invariance, and it is proven to have a convergence rate comparable to AdaGrad/Adam on smooth non-convex problems.
ReNO: Enhancing One-step Text-to-Image Models through Reward-based Noise Optimization
Luca Eyring (Technical University of Munich), Zeynep Akata (Technical University of Munich)
GenerationOptimizationReinforcement LearningImageText
🎯 What it does: A method is proposed to enhance the performance of single-step text-to-image models during the inference phase through Reward-guided Noise Optimization (ReNO).
Renovating Names in Open-Vocabulary Segmentation Benchmarks
Haiwen Huang (Bosch IoC Lab and University of Tubingen), Andreas Geiger (University of Tubingen)
SegmentationTransformerLarge Language ModelVision Language ModelImageBenchmark
🎯 What it does: Renaming the open-source segmentation benchmark and constructing the RENOVATE framework to achieve fine-grained automated naming for each instance.
Reparameterization invariance in approximate Bayesian inference
Hrittik Roy (Technical University of Denmark), Søren Hauberg (Technical University of Denmark)
ClassificationAnomaly DetectionOptimizationConvolutional Neural NetworkImageStochastic Differential Equation
🎯 What it does: This paper studies the issue of the lack of reparameterization invariance in approximate posteriors of Bayesian neural networks, proposing the use of Laplace diffusion with pseudo-Riemannian metrics to explain the geometric reasons behind the success of linearized Laplace, and providing a feasible sampling algorithm.
Reparameterized Multi-Resolution Convolutions for Long Sequence Modelling
Harry Jake Cunningham (University College London), Marc Peter Deisenroth (University College London)
RecognitionOptimizationComputational EfficiencyConvolutional Neural NetworkMultimodality
🎯 What it does: The MRConv model is proposed, utilizing reparameterizable multi-resolution convolution for long sequence modeling.
ReplaceAnything3D: Text-Guided Object Replacement in 3D Scenes with Compositional Scene Representations
Edward Bartrum (University College London), Lei Xiao (Meta Reality Labs)
RestorationGenerationDiffusion modelNeural Radiance FieldImageTextPoint Cloud
🎯 What it does: A text-prompt-based 3D scene object replacement method called RAM3D is proposed, which adopts an erase-and-replace strategy to fuse multi-view images with text input, generating new 3D objects that are consistent with the original scene's lighting and shadows.
Replay-and-Forget-Free Graph Class-Incremental Learning: A Task Profiling and Prompting Approach
Chaoxi Niu (University of Technology Sydney), Bing Liu (University of Illinois at Chicago)
ClassificationGraph Neural NetworkContrastive LearningGraph
🎯 What it does: This study investigates graph-class incremental learning (GCIL) without task IDs, achieving complete no-replay and no-forgetting node classification through task analysis and graph prompts.
Replicability in Learning: Geometric Partitions and KKM-Sperner Lemma
Jason Vander Woude (Sandia National Laboratories), N. V. Vinodchandran (University of Nebraska-Lincoln)
🎯 What it does: This paper studies the replicability of learning algorithms from a geometric perspective, proposing and analyzing the concept of 'secluded partition' and providing an approximate optimal bound for the parameters of this partition.
Replicable Uniformity Testing
Sihan Liu (University of California San Diego), Christopher Ye (University of California San Diego)
🎯 What it does: This paper re-examines the problem of uniformity testing and proposes a replicable uniformity testing algorithm that maintains consistency under any distribution.
Representation Noising: A Defence Mechanism Against Harmful Finetuning
Domenic Rosati (Dalhousie University), Frank Rudzicz (Dalhousie University)
Representation LearningAdversarial AttackTransformerLarge Language ModelSupervised Fine-TuningText
🎯 What it does: A defense mechanism called Representation Noising (RepNoise) is proposed, aimed at resisting harmful fine-tuning attacks (HFA) by eliminating intermediate representation information of harmful inputs, making it difficult for attackers to recover harmful capabilities even after obtaining the model weights.
Reproducibility of predictive networks for mouse visual cortex
Polina Turishcheva (University Göttingen), Alexander S Ecker
Representation LearningConvolutional Neural NetworkSupervised Fine-TuningBiomedical Data
🎯 What it does: Evaluate and enhance the consistency and reproducibility of visual cortical neuron embeddings based on deep prediction models.
Reprogramming Pretrained Target-Specific Diffusion Models for Dual-Target Drug Design
Xiangxin Zhou (University of Chinese Academy of Sciences), Jianzhu Ma (Tsinghua University)
Drug DiscoveryGraph Neural NetworkDiffusion modelGraphBiomedical Data
🎯 What it does: A generative model for dual-target drug design was developed, and a dual-target dataset based on synergistic drug combinations was constructed.
Repurposing Language Models into Embedding Models: Finding the Compute-Optimal Recipe
Albert Q. Jiang (University of Cambridge), Piotr Miłoś (University of Warsaw)
TransformerLarge Language ModelSupervised Fine-TuningContrastive LearningText
🎯 What it does: This study investigates the optimal configuration for fine-tuning a pre-trained decoder language model to produce high-quality text embeddings under computational constraints using contrastive loss.
Reranking Laws for Language Generation: A Communication-Theoretic Perspective
António Farinhas (Instituto Superior Técnico, Universidade de Lisboa), Andre Martins
GenerationAI Code AssistantTransformerLarge Language ModelText
🎯 What it does: This paper proposes viewing the generator-reorderer system as a communication system and proves that under multi-path redundancy generation and reordering, the probability of incorrect answers can decay exponentially/power-law with the number of candidates N, achieving asymptotically error-free generation.
ResAD: A Simple Framework for Class Generalizable Anomaly Detection
Xincheng Yao (Shanghai Jiao Tong University), Chongyang Zhang (Shanghai Jiao Tong University)
Anomaly DetectionConvolutional Neural NetworkImageVideoBenchmark
🎯 What it does: Proposes the ResAD framework, which utilizes residual features to achieve class-generalizable anomaly detection.
Resfusion: Denoising Diffusion Probabilistic Models for Image Restoration Based on Prior Residual Noise
Zhenning Shi (Nankai University), Huazhu Fu (Institute of High Performance Computing)
RestorationConvolutional Neural NetworkDiffusion modelImage
🎯 What it does: A general framework named Resfusion is proposed, which introduces residual terms into the diffusion forward process and directly starts from the noisy degraded image to complete the reverse recovery.
Reshuffling Resampling Splits Can Improve Generalization of Hyperparameter Optimization
Thomas Nagler (Ludwig Maximilian University of Munich), Matthias Feurer (Ludwig Maximilian University of Munich)
OptimizationHyperparameter SearchTabularBenchmark
🎯 What it does: This paper explores whether reshuffling the training/validation split in hyperparameter optimization (HPO) can improve model generalization performance, providing theoretical analysis, simulation experiments, and large-scale benchmark tests to validate the method.
Resolving Discrepancies in Compute-Optimal Scaling of Language Models
Tomer Porian (Tel Aviv University), Yair Carmon (Tel Aviv University)
TransformerLarge Language ModelText
🎯 What it does: This paper systematically reproduces and compares the optimal scaling laws of two language models by Kaplan et al. and Hoffmann et al., identifying three major factors that lead to differences: the computational cost of the final layer, the length of the warm-up period, and the tuning method of optimizer hyperparameters. Based on this, a more accurate scaling law has been redefined.
Resource-Aware Federated Self-Supervised Learning with Global Class Representations
Mingyi Li (Shandong University), Dongxiao Yu (Shandong University)
Federated LearningKnowledge DistillationRepresentation LearningContrastive LearningImage
🎯 What it does: A multi-teacher knowledge distillation framework FedMKD is proposed for learning global class representations in federated self-supervised learning scenarios with resource constraints, model heterogeneity, and imbalanced class distributions.
ReST-MCTS*: LLM Self-Training via Process Reward Guided Tree Search
Dan Zhang (Tsinghua University), Jie Tang (Tsinghua University)
Reinforcement Learning from Human FeedbackTransformerLarge Language ModelSupervised Fine-TuningReinforcement LearningTextBenchmark
🎯 What it does: Proposes ReST-MCTS, which utilizes tree search MCTS* and an automatically generated process reward model to perform dual training of the strategy and reward model for LLM in a self-supervised manner, generating high-quality reasoning trajectories.
RestoreAgent: Autonomous Image Restoration Agent via Multimodal Large Language Models
Haoyu Chen (Hong Kong University of Science and Technology), Lei Zhu (Hong Kong University of Science and Technology)
RestorationTransformerLarge Language ModelSupervised Fine-TuningAgentic AIImageMultimodality
🎯 What it does: This paper proposes an autonomous image restoration agent called RestoreAgent, based on a multimodal large language model (LLM), which can automatically identify multiple degradation types in input images, dynamically plan the execution order of models, and select the most suitable dedicated restoration model to achieve end-to-end multi-task image restoration.
Rethinking 3D Convolution in $\ell_p$-norm Space
Li Zhang (Hefei Institute of Physical Science), Liu Liu (Hefei University of Technology)
SegmentationPose EstimationOptimizationConvolutional Neural NetworkPoint Cloud
🎯 What it does: The study is based on ℓp-norm 3D convolution, proving its universal approximation capability, proposing ℓ1 convolution, and providing mixed gradient and dynamic learning rate optimization strategies, verifying its effectiveness in point cloud tasks.
Rethinking Decoders for Transformer-based Semantic Segmentation: A Compression Perspective
Qishuai Wen (Beijing University of Posts and Telecommunications), Chun-Guang Li (Beijing University of Posts and Telecommunications)
SegmentationCompressionTransformerImage
🎯 What it does: A Transformer decoder based on compression theory (DEPICT) is proposed, which completes semantic segmentation through PCA and a gradient-free expansion of coding rate, designing fully attention-based self-attention and cross-attention modules.
Rethinking Deep Thinking: Stable Learning of Algorithms using Lipschitz Constraints
Jay Bear (University of Southampton), Jonathon Hare (University of Southampton)
Convolutional Neural NetworkRecurrent Neural NetworkSupervised Fine-TuningTabularSequential
🎯 What it does: The Deep Thinking with Lipschitz Constraints (DT‑L) model is proposed, which addresses the numerical instability and convergence uncertainty of traditional Deep Thinking networks during training and inference by enforcing Lipschitz contraction constraints in the recursive part.
Rethinking Exploration in Reinforcement Learning with Effective Metric-Based Exploration Bonus
Yiming Wang (University of Macau), Leong Hou U (University of Macau)
Reinforcement Learning
🎯 What it does: This paper proposes a new exploration reward based on effective metrics—EME, aimed at overcoming the limitations of existing metric-based exploration methods in terms of counting scaling factors and approximation errors, thereby achieving more efficient exploration in sparse reward environments.
Rethinking Fourier Transform from A Basis Functions Perspective for Long-term Time Series Forecasting
Runze Yang (Shanghai Jiao Tong University), li jianxun
TransformerTime Series
🎯 What it does: This paper proposes and implements Fourier Basis Mapping (FBM), treating the discrete Fourier transform as an expansion of sine and cosine basis functions, generating features that incorporate both time and frequency information. It achieves long-term time series prediction through three mapping networks: linear, MLP, and Transformer, while addressing the issues of inconsistent starting periods and sequence lengths present in existing Fourier-based methods.
Rethinking Human Evaluation Protocol for Text-to-Video Models: Enhancing Reliability, Reproducibility, and Practicality
Tianle Zhang (Shanghai AI Laboratory), Kaipeng Zhang (Shanghai AI Laboratory)
GenerationData SynthesisPrompt EngineeringVideoText
🎯 What it does: A standardized human evaluation protocol for text-to-video generation models, T2VHE, is proposed to address the issues of repeatability, reliability, and practicality in existing evaluations.
Rethinking Imbalance in Image Super-Resolution for Efficient Inference
Wei Yu (Harbin Institute of Technology), Xiangyang Ji (Tsinghua University)
RestorationSuper ResolutionImage
🎯 What it does: This study investigates the imbalance problem of data distribution and model optimization in image super-resolution, proposing a weight balancing framework WBSR, which includes Hierarchical Equitable Sampling (HES) and Balanced Diversity Loss (BDLoss), and presents a gradient projection dynamic inference strategy.
Rethinking Inverse Reinforcement Learning: from Data Alignment to Task Alignment
Weichao Zhou (Boston University), Wenchao Li (Boston University)
Robotic IntelligenceReinforcement Learning from Human FeedbackReinforcement LearningGenerative Adversarial Network
🎯 What it does: A new Inverse Reinforcement Learning (IRL) framework is proposed, prioritizing task alignment over traditional data alignment, utilizing expert demonstrations as weak supervision signals, and deriving a set of candidate reward functions to better reflect task objectives.
Rethinking LLM Memorization through the Lens of Adversarial Compression
Avi Schwarzschild (Carnegie Mellon University), J Zico Kolter
CompressionOptimizationTransformerLarge Language ModelPrompt EngineeringText
🎯 What it does: Proposes a memory metric based on Adversarial Compression Ratio (ACR) to assess the 'memory' extent of large language models in their training data.
Rethinking Memory and Communication Costs for Efficient Data Parallel Training of Large Language Models
Hanxiao Zhang (Ant Group), JUN ZHOU
OptimizationComputational EfficiencyTransformerLarge Language ModelText
🎯 What it does: A novel partitioning and communication strategy for data parallel training of large language models, called PaRO, is proposed, including PaRO-DP and PaRO-CC, to better balance memory usage and communication costs, thereby improving training speed.
Rethinking Misalignment in Vision-Language Model Adaptation from a Causal Perspective
Yanan Zhang (University of Chinese Academy of Sciences), Wenwen Qiang (Institute of Software Chinese Academy of Sciences)
Domain AdaptationTransformerPrompt EngineeringVision Language ModelImage
🎯 What it does: In-depth research on the two-layer mismatch problem of CLIP in downstream tasks (task mismatch and data mismatch) is conducted, and a causal inference-based CDC method is proposed to enhance the model's generalization ability;
Rethinking Model-based, Policy-based, and Value-based Reinforcement Learning via the Lens of Representation Complexity
Guhao Feng (Peking University), Han Zhong (Peking University)
Reinforcement LearningSequential
🎯 What it does: This paper theoretically proves the hierarchical relationship of representation complexity among models, policies, and value functions by constructing multiple special MDPs (3-SAT MDP, NP MDP, CVP MDP, P MDP), and validates this hierarchy through experiments in common MuJoCo environments.
Rethinking No-reference Image Exposure Assessment from Holism to Pixel: Models, Datasets and Benchmarks
Shuai He (Beijing University of Posts and Telecommunications), Huadong Ma (Beijing University of Posts and Telecommunications)
Image TranslationRestorationData-Centric LearningConvolutional Neural NetworkTransformerImageBenchmark
🎯 What it does: A no-reference image exposure assessment model P-IEANet is proposed, and a large-scale pixel-level exposure assessment dataset IEA40K is constructed, along with a benchmark evaluation of 19 methods.
Rethinking Optimal Transport in Offline Reinforcement Learning
Arip Asadulaev (Artificial Intelligence Research Institute), Evgeny Burnaev
Reinforcement LearningTabular
🎯 What it does: This paper proposes an offline reinforcement learning algorithm based on Partial Optimal Transport, utilizing the value function as the transport cost to directly map states to the optimal subset of expert action distributions, thereby achieving 'stitched' optimal behavior.
Rethinking Out-of-Distribution Detection on Imbalanced Data Distribution
Kai Liu (Zhejiang University), Jieping Ye (Alibaba Cloud)
Anomaly DetectionConvolutional Neural NetworkImage
🎯 What it does: A framework for OOD detection called ImOOD is proposed, which can correct the bias caused by class imbalance through regularization during the training phase, improving the accuracy of OOD detection under long-tail data distribution.
Rethinking Parity Check Enhanced Symmetry-Preserving Ansatz
Ge Yan (Shanghai Jiao Tong University), Junchi Yan (Shanghai Jiao Tong University)
OptimizationTabularPhysics Related
🎯 What it does: This study investigates a variational quantum algorithm (VQA) that combines Hamming Weight Preserving (HWP) ansatz with parity checks, achieving error mitigation and hard constraints in both quantum chemistry and constrained combinatorial optimization (such as QAP).
Rethinking Reconstruction-based Graph-Level Anomaly Detection: Limitations and a Simple Remedy
Sunwoo Kim (Korea Advanced Institute of Science and Technology), Kijung Shin (Korea Advanced Institute of Science and Technology)
Anomaly DetectionGraph Neural NetworkAuto EncoderGraph
🎯 What it does: This study investigates the 'reconstruction flip' phenomenon in graph autoencoders for graph-level anomaly detection and proposes the GLAD method MUSE based on a multidimensional reconstruction error summary.
Rethinking Score Distillation as a Bridge Between Image Distributions
David McAllister (University of California Berkeley), Angjoo Kanazawa (University of California Berkeley)
GenerationData SynthesisDiffusion modelScore-based ModelNeural Radiance FieldImageText
🎯 What it does: By treating Score Distillation Sampling as a Schrödinger bridge problem, this paper analyzes its two major sources of error and proposes a method to correct these errors using only textual descriptions of the source distribution, improving the generation effects of images, NeRF, and paintings to real images.
Rethinking the Capacity of Graph Neural Networks for Branching Strategy
Ziang Chen (Massachusetts Institute of Technology), Wotao Yin (Alibaba)
OptimizationGraph Neural NetworkGraph
🎯 What it does: This paper studies the expressive power of Graph Neural Networks (GNN) for the strong branching (SB) scores of Mixed Integer Linear Programming (MILP) and proposes the class of MP-tractable MILP, proving that MP-GNN can universally approximate SB in this class; it also provides a non-approximable counterexample for MP-GNN in the non-MP-tractable case and proves that the second-order folklore GNN (2-FGNN) can approximate SB across all MILP distributions; additionally, a method for determining MP-tractability based on Weisfeiler-Lehman color refinement is proposed to guide model selection.
Rethinking the Diffusion Models for Missing Data Imputation: A Gradient Flow Perspective
Zhichao Chen (Zhejiang University), Hao Wang (Zhejiang University)
Diffusion modelScore-based ModelTabularOrdinary Differential Equation
🎯 What it does: Proposes NewImp, a framework for missing data imputation based on Wasserstein Gradient Flow in diffusion models.
Rethinking the Membrane Dynamics and Optimization Objectives of Spiking Neural Networks
Hangchi Shen (Southwest University), Gang Pan (Zhejiang University)
ClassificationOptimizationSpiking Neural NetworkSupervised Fine-TuningImage
🎯 What it does: A learnable initial membrane potential (IMP) is proposed to enhance the expressive capability of spiking neural networks, combined with LTS processing and label smoothing TET loss to improve performance on static and event data.
Rethinking the Power of Timestamps for Robust Time Series Forecasting: A Global-Local Fusion Perspective
Chengsen Wang (Beijing University of Posts and Telecommunications), Jianxin Liao (Beijing University of Posts and Telecommunications)
TransformerTime Series
🎯 What it does: A general plugin framework GLAFF is proposed, which utilizes global information from timestamps to enhance the robustness of mainstream time series forecasting models.
Rethinking The Training And Evaluation of Rich-Context Layout-to-Image Generation
Jiaxin Cheng (University of Macau), Yicong Zhou (Amazon Web Services)
GenerationData SynthesisTransformerLarge Language ModelDiffusion modelImageText
🎯 What it does: This paper proposes a Regional Cross-Attention module to handle complex, lengthy text descriptions in Layout-to-Image (L2I) generation, and redesigns evaluation metrics for open vocabulary scenarios; it also constructs a large-scale synthetic rich-context L2I dataset.
Rethinking Transformer for Long Contextual Histopathology Whole Slide Image Analysis
Honglin Li (Zhejiang University), Lin Yang (Research Center for Industries of the Future)
ClassificationSegmentationComputational EfficiencyRepresentation LearningTransformerImageBiomedical Data
🎯 What it does: A Local-Global Hybrid Transformer (LongMIL) for long sequence multi-instance learning of large-sized, diverse-deformation digital pathology slides (Whole Slide Image, WSI) has been designed and implemented, enhancing representation capability through local attention masks while significantly reducing computational complexity.
Rethinking Weight Decay for Robust Fine-Tuning of Foundation Models
Junjiao Tian (Georgia Institute of Technology), Zsolt Kira (Georgia Institute of Technology)
Domain AdaptationOptimizationSupervised Fine-TuningImageText
🎯 What it does: A new selective weight decay technique called Selective Projection Decay (SPD) is proposed for robust fine-tuning on powerful base models, selectively applying strong penalties to certain layers while allowing other layers to vary freely.
RETR: Multi-View Radar Detection Transformer for Indoor Perception
Ryoma Yataka (Mitsubishi Electric Research Laboratories), Ryuhei Takahashi (Mitsubishi Electric Corporation)
Object DetectionSegmentationTransformerPoint Cloud
🎯 What it does: A Transformer-based multi-view radar detection framework called RETR is proposed, which directly implements 3D frame prediction on radar heatmaps and projects it onto the image plane for object detection and segmentation.
Retrieval & Fine-Tuning for In-Context Tabular Models
Valentin Thomas (Layer6 Technologies), Anthony L. Caterini
RetrievalTransformerSupervised Fine-TuningTabular
🎯 What it does: Based on TabPFN, we introduce nearest neighbor retrieval and end-to-end fine-tuning to construct a locally calibrated ICL model called LoCalPFN;
Retrieval-Augmented Diffusion Models for Time Series Forecasting
Jingwei Liu (Peking University), Shenda Hong (Peking University)
RetrievalTransformerDiffusion modelTime SeriesBiomedical DataElectrocardiogram
🎯 What it does: This paper proposes a Retrieval-Augmented Time Series Diffusion Model (RATD), which uses samples from a database that are most similar to historical sequences as references to guide the diffusion process and improve prediction accuracy.
Retrieval-Retro: Retrieval-based Inorganic Retrosynthesis with Expert Knowledge
Heewoong Noh (Korea Advanced Institute of Science and Technology), Chanyoung Park (Korea Advanced Institute of Science and Technology)
RetrievalGraph Neural NetworkGraph
🎯 What it does: Developed the Retrieval-Retro method, which utilizes retrieved reference materials and implicitly extracts precursor information through attention mechanisms to achieve inverse synthesis prediction of inorganic materials.
Return of Unconditional Generation: A Self-supervised Representation Generation Method
Tianhong Li (Massachusetts Institute of Technology), Kaiming He (Massachusetts Institute of Technology)
GenerationData SynthesisRepresentation LearningDiffusion modelContrastive LearningImage
🎯 What it does: The Representation-Conditioned Generation (RCG) framework is proposed, which first generates low-dimensional semantic representations using a self-supervised encoder, then generates these representations in an unsupervised manner, and finally synthesizes images using a representation-conditioned image generator, achieving high-quality image generation without labels.
Revealing Distribution Discrepancy by Sampling Transfer in Unlabeled Data
Zhilin Zhao (Macquarie University), Wei-Shi Zheng (Sun Yat-sen University)
Domain AdaptationAnomaly DetectionImage
🎯 What it does: A metric called Importance Divergence (I-Div) is proposed to assess the applicability of trained models on unlabeled test data.
Reverse Transition Kernel: A Flexible Framework to Accelerate Diffusion Inference
Xunpeng Huang (Hong Kong University of Science and Technology), Tong Zhang (University of Illinois at Urbana-Champaign)
OptimizationComputational EfficiencyDiffusion modelStochastic Differential Equation
🎯 What it does: The Reverse Transition Kernel (RTK) framework is proposed, which decomposes the inference of diffusion models into several controllable subproblems, and implements two efficient inference algorithms, RTK-MALA and RTK-ULD, based on MALA and ULD MCMC methods.
Reversing the Forget-Retain Objectives: An Efficient LLM Unlearning Framework from Logit Difference
Jiabao Ji (University of California Santa Barbara), Shiyu Chang (University of California Santa Barbara)
TransformerLarge Language ModelSupervised Fine-TuningText
🎯 What it does: A novel LLM no-learning framework ULD based on auxiliary model adversarial objectives and logit difference calculation is proposed, which can forget specified document knowledge while retaining other knowledge.
ReVideo: Remake a Video with Motion and Content Control
Chong Mou (Peking University), Jian Zhang (Peking University)
GenerationData SynthesisDiffusion modelVideo
🎯 What it does: A method called ReVideo is proposed for simultaneously editing content and motion trajectories in specific areas of a video, allowing users to achieve precise local edits by simply modifying the first frame and drawing the trajectory.
Revisiting Adversarial Patches for Designing Camera-Agnostic Attacks against Person Detection
Hui Wei (Wuhan University), Zheng Wang (Wuhan University)
Object DetectionAdversarial AttackGenerative Adversarial NetworkImage
🎯 What it does: A cross-camera physical adversarial patch (Camera-Agnostic Patch, CAP) attack method for human detection is proposed, which maintains a high attack success rate across different cameras.
Revisiting Differentially Private ReLU Regression
Meng Ding (University at Buffalo), Jinhui Xu (University at Buffalo)
OptimizationSafty and PrivacyTabular
🎯 What it does: This paper proposes two differential privacy algorithms for ReLU regression, DP-GLMtron and DP-TAGLMtron, aimed at addressing the privacy-performance trade-off in high-dimensional, over-parameterized scenarios.
Revisiting Ensembling in One-Shot Federated Learning
Youssef Allouah (École Polytechnique Fédérale de Lausanne), Rishi Sharma (École Polytechnique Fédérale de Lausanne)
Federated LearningKnowledge DistillationImageText
🎯 What it does: A FENS framework is proposed that combines one-round federated learning with lightweight aggregation training, utilizing an aggregator network to integrate local models, maintaining low communication costs while approaching the accuracy of traditional iterative federated learning.
Revisiting K-mer Profile for Effective and Scalable Genome Representation Learning
Abdulkadir Celikkanat, Thomas Dyhre Nielsen
Representation LearningContrastive LearningBiomedical Data
🎯 What it does: This paper re-examines k-mer based genomic representation and proposes a lightweight, scalable model for metagenomic binning tasks.
Revisiting motion information for RGB-Event tracking with MOT philosophy
Tianlu Zhang (Xidian University), Jungong Han (Tsinghua University)
Object TrackingTransformerImageVideoMultimodality
🎯 What it does: This paper proposes a single-object tracking framework CSAM that combines RGB frames and event streams, drawing on the philosophy of multi-object tracking (MOT) to simultaneously track the target and interfering objects, utilizing motion information to enhance the distinction between the target and the interference.
Revisiting Score Propagation in Graph Out-of-Distribution Detection
Longfei Ma (Zhejiang University), Fei Wu (Zhejiang University)
Anomaly DetectionGraph Neural NetworkGraph
🎯 What it does: This study explores the out-of-distribution (OOD) node detection problem in graph data and proposes a simple and effective OOD score propagation method that improves detection performance by propagating the OOD scores of neighboring nodes within the graph structure.
Revisiting Self-Supervised Heterogeneous Graph Learning from Spectral Clustering Perspective
Yujie Mo (National University of Singapore), Xinchao Wang (University of Electronic Science and Technology of China)
ClassificationRepresentation LearningGraph Neural NetworkContrastive LearningGraph
🎯 What it does: A self-supervised heterogeneous graph learning framework called SCHOOL based on spectral clustering is proposed to address the issues of noisy graph structures and underutilized clustering information in traditional methods.
Revisiting the Integration of Convolution and Attention for Vision Backbone
Lei Zhu (City University of Hong Kong), Rynson W. H. Lau
ClassificationObject DetectionSegmentationComputational EfficiencyKnowledge DistillationConvolutional Neural NetworkTransformerImage
🎯 What it does: This paper proposes the GLMix structure, which allows convolution and multi-head self-attention (MHSA) to work in parallel at different granularities, and integrates fine-grained grid features with coarse-grained semantic slot features through a soft clustering and dispatch module, achieving efficient local-global feature mixing.
Revive Re-weighting in Imbalanced Learning by Density Ratio Estimation
Jiaan Luo (Shanghai Jiao Tong University), Yanfeng Wang (Shanghai Jiao Tong University)
ClassificationReinforcement LearningImage
🎯 What it does: Dynamic adjustment of class weights using density ratio estimation to alleviate the performance decline of deep learning models on long-tailed data.
Reward Machines for Deep RL in Noisy and Uncertain Environments
Andrew C Li, Sheila A. McIlraith (University of Toronto)
Reinforcement Learning
🎯 What it does: This paper proposes a method for deep reinforcement learning using Reward Machines in noisy and uncertain environments, and designs three inference modules (Naive, IBU, TDM) to handle uncertain interpretations of domain-specific vocabulary.
RFLPA: A Robust Federated Learning Framework against Poisoning Attacks with Secure Aggregation
Peihua Mai (National University of Singapore), Yan Pang (National University of Singapore)
Federated LearningSafty and PrivacyConvolutional Neural NetworkImage
🎯 What it does: A robust federated learning framework RFLPA is proposed under the secure aggregation (SecAgg) protocol, addressing the dual threats of privacy leakage and model poisoning.
RG-SAN: Rule-Guided Spatial Awareness Network for End-to-End 3D Referring Expression Segmentation
Changli Wu (Xiamen University), Rongrong Ji (Xiamen University)
Object DetectionSegmentationTransformerPoint Cloud
🎯 What it does: The RG-SAN model is proposed to achieve end-to-end 3D pointing expression segmentation.
RGFN: Synthesizable Molecular Generation Using GFlowNets
Michał Koziarski (Mila - Québec AI Institute), Robert A. Batey (University of Toronto)
GenerationDrug DiscoveryGraph Neural NetworkTransformerGraph
🎯 What it does: Proposes Reaction-GFlowNet (RGFN) to generate synthesizable molecules in the chemical reaction space;
RGMDT: Return-Gap-Minimizing Decision Tree Extraction in Non-Euclidean Metric Space
Jingdi Chen (George Washington University), Tian Lan (George Washington University)
Explainability and InterpretabilityReinforcement LearningTabular
🎯 What it does: A multi-agent reinforcement learning explanation method based on decision trees is proposed, introducing the RGMDT framework to minimize return gaps and providing a closed-form upper bound;
Richelieu: Self-Evolving LLM-Based Agents for AI Diplomacy
Zhenyu Guan (Peking University), Yizhou Wang (Peking University)
TransformerLarge Language ModelAgentic AIText
🎯 What it does: Richelieu, a self-evolving diplomatic agent based on large language models, has been developed to perform social reasoning, long-term planning, negotiation, and self-reflection in the game of Diplomacy, continuously improving its capabilities through self-play.
Right this way: Can VLMs Guide Us to See More to Answer Questions?
Li Liu (University of California Santa Cruz), Leilani H. Gilpin (University of California Santa Cruz)
TransformerSupervised Fine-TuningPrompt EngineeringVision Language ModelImageTextMultimodality
🎯 What it does: A 'directional guidance' task for visual question answering is proposed, along with the construction of a corresponding manually annotated test set; unsupervised data augmentation is performed using VLM to generate synthetic training data, which is then fine-tuned; the performance of the zero-shot and fine-tuned models is evaluated and compared.
Risk-Averse Fine-tuning of Large Language Models
Sapana Chaudhary (Amazon Web Services), Srinivas Shakkottai (Texas A&M University)
Reinforcement Learning from Human FeedbackTransformerLarge Language ModelSupervised Fine-TuningReinforcement LearningText
🎯 What it does: A risk-averse RLHF method (RA-RLHF) is proposed and implemented for fine-tuning large language models to reduce the probability of generating harmful outputs in response to negative or toxic prompts.
Risk-sensitive control as inference with Rényi divergence
Kaito Ito (University of Tokyo), Kenji Kashima (Kyoto University)
Reinforcement LearningSequential
🎯 What it does: This paper studies a risk-sensitive framework called RCaI that incorporates Rényi divergence into Control as Inference (CaI), and derives the risk-sensitive policy gradient and soft actor-critic (RSAC) algorithm based on this framework.
RL in Latent MDPs is Tractable: Online Guarantees via Off-Policy Evaluation
Jeongyeol Kwon (University of Wisconsin Madison), Yonathan Efroni (Meta AI)
Recommendation SystemReinforcement Learning
🎯 What it does: A new sample-efficient algorithm is proposed to address the online exploration problem in the latent Markov decision process (LMDP), particularly in the absence of additional distributional assumptions.
RL on Incorrect Synthetic Data Scales the Efficiency of LLM Math Reasoning by Eight-Fold
Amrith Setlur (Carnegie Mellon University), Aviral Kumar (Google DeepMind)
Computational EfficiencyLarge Language ModelSupervised Fine-TuningReinforcement LearningText
🎯 What it does: In mathematical reasoning tasks, the authors systematically evaluate and improve the fine-tuning methods of LLMs based on synthetic data, proposing the use of model-generated erroneous reasoning trajectories (negative samples) for stepwise credit assignment to enhance sample efficiency and mitigate overfitting to irrelevant steps.
RL-GPT: Integrating Reinforcement Learning and Code-as-policy
Shaoteng Liu (Chinese University of Hong Kong), Jiaya Jia (Hong Kong University of Science and Technology)
Robotic IntelligenceAI Code AssistantTransformerLarge Language ModelReinforcement LearningAgentic AITextBenchmark
🎯 What it does: Developed the RL-GPT framework, which combines large language models and reinforcement learning to achieve collaborative learning between high-level code strategies and low-level RL actions through a two-tier agent system.
RLE: A Unified Perspective of Data Augmentation for Cross-Spectral Re-Identification
Lei Tan (Xiamen University), Rongrong Ji (Xiamen University)
RecognitionRetrievalConvolutional Neural NetworkImage
🎯 What it does: This paper proposes a Random Linear Enhancement (RLE) method based on the Lambertian model for data augmentation in cross-spectral re-identification.
RMLR: Extending Multinomial Logistic Regression into General Geometries
Ziheng Chen (University of Trento), Nicu Sebe (University of Trento)
ClassificationOptimizationTabularBenchmark
🎯 What it does: A general polynomial logistic regression framework based on Riemannian logarithmic mapping is proposed, and five families of SPD MLR and Lie MLR are implemented on SPD and SO(n).
Road Network Representation Learning with the Third Law of Geography
Haicang Zhou (Nanyang Technological University), Yew-Soon Ong (Nanyang Technological University)
RetrievalRepresentation LearningGraph Neural NetworkContrastive LearningImageGraph
🎯 What it does: This paper proposes a novel road network representation learning framework that captures geographical configurations using street view images and combines contrastive learning to generate low-dimensional vector representations of road segments.
ROBIN: Robust and Invisible Watermarks for Diffusion Models with Adversarial Optimization
Huayang Huang (Wuhan University), Qian Wang (Wuhan University)
GenerationOptimizationAdversarial AttackDiffusion modelImage
🎯 What it does: Embed a robust watermark during the intermediate generation phase of the diffusion model, and generate hidden prompts through adversarial optimization to guide the model in embedding the watermark into the final image during subsequent sampling;
RobIR: Robust Inverse Rendering for High-Illumination Scenes
Ziyi Yang (Zhejiang University), Xiaogang Jin (Zhejiang University)
RestorationData SynthesisNeural Radiance FieldImage
🎯 What it does: RobIR is proposed, an implicit inverse rendering framework that utilizes ACES tone mapping and regularized visibility estimation to accurately separate shadows, ambient light, and the PBR materials of objects in high illumination scenes.
RoboMamba: Efficient Vision-Language-Action Model for Robotic Reasoning and Manipulation
Jiaming Liu (Peking University), Shanghang Zhang (Peking University)
Pose EstimationComputational EfficiencyRobotic IntelligenceTransformerLarge Language ModelSupervised Fine-TuningVision-Language-Action ModelImageTextMultimodality
🎯 What it does: We propose RoboMamba, an efficient visual-language-action model that combines the Mamba LLM with a visual encoder, capable of visual reasoning and low-level pose prediction.
Robot Policy Learning with Temporal Optimal Transport Reward
Yuwei Fu (McGill University), Benoit Boulet (McGill University)
Robotic IntelligenceReinforcement LearningVideo
🎯 What it does: In a robot control environment with only a few expert video demonstrations and no task rewards, the study explores a Temporal Optimal Transport reward learning strategy.