arXivSub Start free trial

NeurIPS 2024 Papers — Page 30

Conference on Neural Information Processing Systems · 4035 papers

QuadMamba: Learning Quadtree-based Selective Scan for Visual State Space Model

Fei Xie (Shanghai Jiao Tong University), Chao Ma (Shanghai Jiao Tong University)

ClassificationObject DetectionSegmentationTransformerImage

🎯 What it does: A visual Mamba model called QuadMamba based on a learnable quadtree scanning mechanism has been designed for visual tasks such as image classification, object detection, instance segmentation, and semantic segmentation.

Quadratic Quantum Variational Monte Carlo

Baiyu Su (University of Texas at Austin), qiang liu

OptimizationTransformerTabularPhysics RelatedStochastic Differential Equation

🎯 What it does: A new algorithm called Quadratic Quantum Variational Monte Carlo (Q2VMC) is proposed and implemented, which optimizes neural network wave functions using discrete imaginary time Schrödinger evolution and KL divergence projection.

Qualitative Mechanism Independence

Oliver Ethan Richardson, Joseph Halpern

🎯 What it does: This paper defines the qualitative compatibility between probability distributions and directed hypergraphs (QIM-compatibility) and proves its equivalence and constraint relationships with concepts such as causal models, functional dependencies, and interaction information in information theory.

Quality-Improved and Property-Preserved Polarimetric Imaging via Complementarily Fusing

Chu Zhou (National Institute of Informatics), Boxin Shi (Peking University)

RestorationConvolutional Neural NetworkImage

🎯 What it does: A three-stage neural network framework is designed to integrate noise and blurred polarization snapshots, restoring high-quality polarization images along with their degree and angle of polarization.

QuanTA: Efficient High-Rank Fine-Tuning of LLMs with Quantum-Informed Tensor Adaptation

Zhuo Chen (Massachusetts Institute of Technology), Marin Soljacic (Massachusetts Institute of Technology)

TransformerLarge Language ModelSupervised Fine-TuningText

🎯 What it does: Proposes QuanTA, a tensor adaptive method inspired by quantum circuits for high-rank, efficient LLM fine-tuning.

Quantifying Aleatoric Uncertainty of the Treatment Effect: A Novel Orthogonal Learner

Valentyn Melnychuk (Ludwig Maximilian University of Munich), Mihaela van der Schaar (University of Cambridge)

Flow-based ModelTabular

🎯 What it does: A novel orthogonal learner (AU-learner) is proposed to estimate the Markov boundary of the conditional treatment effect distribution (CDTE), thereby quantifying the Alitayi uncertainty of treatment effects.

Quantifying and Optimizing Global Faithfulness in Persona-driven Role-playing

Letian Peng (University of California, San Diego), Jingbo Shang (University of California, San Diego)

OptimizationTransformerLarge Language ModelReinforcement LearningText

🎯 What it does: Quantitatively assess the global fidelity in persona-driven role-playing (PRP) and propose an Active-Passive Constraint (APC) scoring and an APC-based Direct Preference Optimization (DPO) scheme.

Quantifying the Gain in Weak-to-Strong Generalization

Moses Charikar (Stanford University), Kirankumar Shiragur (Microsoft Research)

OptimizationRepresentation LearningTransformerLarge Language ModelText

🎯 What it does: Proposes and theorizes the mechanism of weak-to-strong generalization, proving that the error of a strong model using labels generated by a weak model is at least one 'misfit' order of magnitude lower than that of the weak model;

Quantitative Convergences of Lie Group Momentum Optimizers

Lingkai Kong (Georgia Institute of Technology), Molei Tao (Georgia Institute of Technology)

OptimizationTransformerImage

🎯 What it does: This paper proposes two momentum-based Lie group optimizers: Lie Heavy-Ball and Lie NAG-SC, and provides quantitative convergence rates under L-smooth and local strong convexity conditions, verifying the acceleration effect of Lie NAG-SC, while applying it to the optimization of visual Transformers with orthogonal constraints.

Quantum algorithm for large-scale market equilibrium computation

Po-Wei Huang (National University of Singapore), Patrick Rebentrost (National University of Singapore)

OptimizationComputational EfficiencyTabularFinance Related

🎯 What it does: The first quantum equilibrium computation algorithm for the Fisher market is proposed, and iterative updates are achieved through error-proportional response dynamics.

Quantum Algorithms for Non-smooth Non-convex Optimization

Chengchang Liu (Chinese University of Hong Kong), John C.S. Lui (Chinese University of Hong Kong)

OptimizationPhysics Related

🎯 What it does: A non-smooth non-convex optimization algorithm based on quantum zero-order methods is proposed to find Goldstein stationary points.

Quantum Deep Equilibrium Models

Philipp Schleich (University of Toronto), Alan Aspuru-Guzik

OptimizationImage

🎯 What it does: A Quantum Deep Equilibrium Model (QDEQ) is proposed, achieving parameter optimization through the use of deep equilibrium techniques on quantum models;

QuaRot: Outlier-Free 4-Bit Inference in Rotated LLMs

Saleh Ashkboos (ETH Zurich), James Hensman (Microsoft Research)

TransformerLarge Language ModelText

🎯 What it does: Proposes the QuaRot method, which uses Hadamard rotation to eliminate outliers in the hidden states and key-value caches of large language models, achieving end-to-end 4-bit quantization of the entire model (weights, activations, KV cache).

Quasi-Bayes meets Vines

David Huk (University of Warwick), Mark Steel (University of Warwick)

Hyperparameter SearchTabular

🎯 What it does: A high-dimensional joint density estimation method combining Bayesian prediction and Vine copula is proposed—Quasi-Bayesian Vine (QB-Vine).

QUEEN: QUantized Efficient ENcoding of Dynamic Gaussians for Streaming Free-viewpoint Videos

Sharath Girish (University of Maryland), Shalini De Mello (NVIDIA)

CompressionGaussian SplattingVideo

🎯 What it does: A streaming compression and encoding framework for online free viewpoint video (FVV) called QUEEN is proposed, which dynamically updates scene attributes using 3D Gaussian expansion (3D-GS) and learns residuals for each frame;

Query-Based Adversarial Prompt Generation

Jonathan Hayase (University of Washington), Milad Nasr (University of Washington)

Adversarial AttackLarge Language ModelPrompt EngineeringText

🎯 What it does: A query-based adversarial prompt generation method (GCQ) is proposed, which enables large language models to produce specified harmful strings or evade content moderation solely through API queries.

Query-Efficient Correlation Clustering with Noisy Oracle

Yuko Kuroki (CENTAI Institute), Wei Chen (Microsoft Research)

OptimizationGraph

🎯 What it does: A method for achieving query-efficient relevant clustering under unknown and noisy similarity is proposed, along with two pure exploration (fixed confidence / fixed budget) online learning frameworks, corresponding to the KC-FC and KC-FB algorithms.

QUEST: Quadruple Multimodal Contrastive Learning with Constraints and Self-Penalization

Qi Song (Beihang University), Jianxin Li (Beihang University)

RetrievalContrastive LearningMultimodalityAudio

🎯 What it does: The QUEST framework is proposed, which implements quaternion constraints and self-punishment for multimodal contrastive learning through a shared and unique information decoder.

QUEST: Quality-Aware Metropolis-Hastings Sampling for Machine Translation

Gonçalo Faria, Andre Martins

GenerationTransformerLarge Language ModelText

🎯 What it does: A quality-aware sampling method called QUEST based on Metropolis-Hastings is proposed to generate high-quality and diverse machine translation results.

QueST: Self-Supervised Skill Abstractions for Learning Continuous Control

Atharva Mete (Georgia Institute of Technology), Animesh Garg (Georgia Institute of Technology)

Robotic IntelligenceReinforcement Learning from Human FeedbackTransformerReinforcement LearningContrastive LearningSequential

🎯 What it does: We propose a self-supervised discrete latent variable model called Quantized Skill Transformer (QueST) for learning transferable low-level skill representations, and for decision-making in multi-task, few-shot, and long-horizon control tasks.

Questioning the Survey Responses of Large Language Models

Ricardo Dominguez-Olmedo (Max-Planck Institute for Intelligent Systems), Celestine Mendler-Dünner (Max-Planck Institute for Intelligent Systems)

Large Language ModelTextReview/Survey Paper

🎯 What it does: This study investigates the use of multiple-choice questions from the American Community Survey (ACS) to evaluate large language models through a 'questionnaire' approach, assessing the similarity of their answer distributions to human demographic data.

Queueing Matching Bandits with Preference Feedback

Jung-hun Kim (Seoul National University), Min-hwan Oh (Seoul National University)

Reinforcement Learning

🎯 What it does: A new queue matching Bandit model is proposed, allowing for learning and stabilizing queues in a multi-queue multi-server environment through service rate preference feedback.

QVAE-Mole: The Quantum VAE with Spherical Latent Variable Learning for 3-D Molecule Generation

Huaijin Wu, Junchi Yan (Shanghai Jiao Tong University)

GenerationData SynthesisDrug DiscoveryAuto EncoderPoint Cloud

🎯 What it does: Proposes the first fully quantum circuit-based Variational Autoencoder (VAE) framework for generating three-dimensional molecular structures, and extends to conditional generation.

QWO: Speeding Up Permutation-Based Causal Discovery in LiGAMs

Mohammad Shahverdikondori (École Polytechnique Fédérale de Lausanne), Negar Kiyavash (École Polytechnique Fédérale de Lausanne)

OptimizationComputational EfficiencyGraph

🎯 What it does: An efficient QWO module is proposed for quickly calculating the causal graph Gπ under a given permutation in linear Gaussian acyclic models (LiGAM), significantly accelerating the permutation search process.

R$^2$-Gaussian: Rectifying Radiative Gaussian Splatting for Tomographic Reconstruction

Ruyi Zha (Australian National University), Hongdong Li (Australian National University)

Gaussian SplattingImageComputed Tomography

🎯 What it does: The paper proposes a sparse-view CT reconstruction framework based on 3D Gaussian splatting, called R-Gaussian.

RA-PbRL: Provably Efficient Risk-Aware Preference-Based Reinforcement Learning

Yujie Zhao (University of California), Huazheng Wang (Oregon State University)

Reinforcement LearningTabularSequential

🎯 What it does: This paper proposes a risk-aware preference reinforcement learning algorithm named RA-PbRL, aimed at addressing the PbRL problem where only complete trajectory preference feedback is available. It demonstrates the theoretical feasibility and lower bounds for achieving nested and static quantile risk objectives.

Rad-NeRF: Ray-decoupled Training of Neural Radiance Field

Lidong Guo (Tsinghua University), Yu Wang (Tsinghua University)

GenerationData SynthesisNeural Radiance FieldGaussian SplattingImage

🎯 What it does: We propose a ray-decoupled training framework called Rad-NeRF, which utilizes sub-NeRF integration and a soft gating module to allocate different sub-models along the ray dimension, achieving better lighting and geometric consistency through deep mutual learning.

RadarOcc: Robust 3D Occupancy Prediction with 4D Imaging Radar

Fangqiang Ding (New York University), Chris Xiaoxuan Lu (University College London)

Object DetectionSegmentationAutonomous DrivingConvolutional Neural NetworkSupervised Fine-TuningPoint Cloud

🎯 What it does: This paper proposes the use of 4D imaging radar tensor (4DRT) for 3D occupancy prediction, constructing the RadarOcc framework.

RAGraph: A General Retrieval-Augmented Graph Learning Framework

Xinke Jiang (Peking University), Yasha Wang (Peking University)

RetrievalGraph Neural NetworkGraphRetrieval-Augmented Generation

🎯 What it does: A general retrieval-enhanced graph learning framework RAGRAPH is proposed, which improves the generalization ability of pre-trained graph neural networks on unseen graph data by utilizing external graph data through retrieval and information injection.

Rainbow Teaming: Open-Ended Generation of Diverse Adversarial Prompts

Mikayel Samvelyan (Meta), Roberta Raileanu (Meta)

GenerationSafty and PrivacyAdversarial AttackTransformerLarge Language ModelPrompt EngineeringText

🎯 What it does: A RAINBOW TEAMING method based on quality diversity search has been developed, utilizing LLM for mutation and evaluation to automatically generate diverse and effective adversarial prompts, helping to identify and fix security vulnerabilities in large language models.

RAMP: Boosting Adversarial Robustness Against Multiple $l_p$ Perturbations for Universal Robustness

Enyi Jiang (University of Illinois), Gagandeep Singh (University of Illinois)

Adversarial AttackConvolutional Neural NetworkImage

🎯 What it does: Proposes the RAMP framework to enhance joint robustness against various l_p transformations.

RandNet-Parareal: a time-parallel PDE solver using Random Neural Networks

Guglielmo Gattiglio (University of Warwick), Massimiliano Tamborrino (University of Warwick)

Time SeriesPhysics RelatedStochastic Differential EquationOrdinary Differential Equation

🎯 What it does: A novel parallel time solver RandNet-Parareal is proposed, which learns the error between coarse and fine solvers through a random neural network, thereby achieving rapid correction in the Parareal iteration.

Random Cycle Coding: Lossless Compression of Cluster Assignments via Bits-Back Coding

Daniel Severo (University of Toronto), Alireza Makhzani (University of Toronto)

Compression

🎯 What it does: Lossless compression of clustering partitions for any dataset, allowing the reconstruction of clustering structure by only transmitting clustering information;

Random Function Descent

Felix Benning (University of Mannheim), Leif Döring (University of Mannheim)

OptimizationConvolutional Neural NetworkImage

🎯 What it does: A gradient descent algorithm based on a random function framework, RFD, is proposed to replace the traditional convex function assumption;

Randomized algorithms and PAC bounds for inverse reinforcement learning in continuous spaces

Angeliki Kamoutsi (École Polytechnique Fédérale de Lausanne), John Lygeros (ETH Zürich)

Reinforcement Learning

🎯 What it does: This paper proposes a framework for inverse reinforcement learning (IRL) based on linear programming (LP) and stochastic scenario methods for continuous state and action spaces. It is capable of learning the cost function under known or unknown MDP dynamics, relying solely on offline expert trajectories and a generator model, and provides corresponding theoretical error and sample complexity upper bounds.

Randomized Exploration for Reinforcement Learning with Multinomial Logistic Function Approximation

Wooseong Cho (Seoul National University), Min-hwan Oh (Seoul National University)

OptimizationReinforcement LearningTabular

🎯 What it does: Two randomized exploration RL algorithms, RRL-MNL and ORRL-MNL, are proposed, which approximate the transition probabilities using a polynomial logistic regression model and achieve constant time computation.

Randomized Exploration in Cooperative Multi-Agent Reinforcement Learning

Hao-Lun Hsu (Duke University), Pan Xu (Duke University)

Reinforcement Learning

🎯 What it does: A unified framework is proposed for implementing Thompson Sampling-based stochastic exploration algorithms—CoopTS-PHE and CoopTS-LMC—in parallel MDP environments;

Randomized Sparse Matrix Compression for Large-Scale Constrained Optimization in Cancer Radiotherapy

Shima Adeli (Sharif University of Technology), Masoud Zarepisheh (Memorial Sloan Kettering Cancer Center)

OptimizationBiomedical Data

🎯 What it does: A dose impact matrix compression method based on random sparsification is proposed, enabling the solution of large-scale constrained nonlinear optimization problems in cancer radiotherapy within clinical time windows.

Randomized Strategic Facility Location with Predictions

Eric Balkanski (Columbia University), Golnoosh Shahkarami (Max Planck Institute for Informatics)

Optimization

🎯 What it does: The study investigates the strategic facility location problem under a learning-enhanced framework, exploring the impact of randomization mechanisms and different types of predictions on the approximation of fair social costs.

Randomized Truthful Auctions with Learning Agents

Gagan Aggarwal (Google Research), Grigoris Velegkas (Yale University)

Reinforcement LearningAgentic AI

🎯 What it does: This study investigates the behavior of agents using no-regret learning algorithms in repeated auctions, particularly focusing on the convergence and revenue maximization issues in truthful auctions.

RanDumb: Random Representations Outperform Online Continually Learned Representations

Ameya Prabhu (University of Oxford), Puneet K. Dokania (IIIT Hyderabad)

ClassificationRepresentation LearningImage

🎯 What it does: This paper presents RanDumb—a baseline for online continual learning that uses only random Fourier feature projections and a nearest class mean (with Mahalanobis distance) classifier, aimed at evaluating the comparison between random representations and traditional online learning representations.

RankRAG: Unifying Context Ranking with Retrieval-Augmented Generation in LLMs

Yue Yu (Georgia Institute of Technology), Bryan Catanzaro (NVIDIA Corporation)

GenerationRetrievalRecommendation SystemTransformerLarge Language ModelSupervised Fine-TuningTextBiomedical DataRetrieval-Augmented Generation

🎯 What it does: Proposes RankRAG, an instruction fine-tuning framework that unifies retrieval and generation within the same LLM;

RankUp: Boosting Semi-Supervised Regression with an Auxiliary Ranking Classifier

Pin-Yen Huang (Academia Sinica), Yu Tsao (Academia Sinica)

OptimizationContrastive LearningImageTextAudio

🎯 What it does: The RankUp framework is proposed, which transforms regression tasks into ranking problems, using an Auxiliary Ranking Classifier (ARC) trained together with semi-supervised classification methods like FixMatch, and introduces an optional Regression Distribution Alignment (RDA) to further improve the quality of pseudo-labels.

Rapid Plug-in Defenders

Kai Wu (Xidian University), Jing Liu (Xidian University)

OptimizationAdversarial AttackTransformerSupervised Fine-TuningImage

🎯 What it does: A fast plugin defense framework CeTaD is proposed, which utilizes a pre-trained Transformer to only fine-tune layer normalization parameters, enhancing adversarial robustness with very few adversarial samples without altering the already deployed model.

RashomonGB: Analyzing the Rashomon Effect and Mitigating Predictive Multiplicity in Gradient Boosting

Hsiang Hsu (JPMorganChase Global Technology Applied Research), Chun-Fu Chen (JPMorganChase AI Research)

OptimizationExplainability and InterpretabilityTabular

🎯 What it does: Analyzes the Rashomon effect in gradient boosting and proposes the RashomonGB method, which efficiently constructs a model ensemble using the residual Rashomon set, further evaluating prediction diversity, interpretability, and fairness.

RaVL: Discovering and Mitigating Spurious Correlations in Fine-Tuned Vision-Language Models

Maya Varma (Stanford University), Curtis Langlotz (Stanford University)

ClassificationObject DetectionSegmentationData-Centric LearningTransformerSupervised Fine-TuningVision Language ModelContrastive LearningImageMultimodality

🎯 What it does: This paper proposes the RAVL method, which analyzes fine-grained image regions to automatically discover and eliminate the false correlations learned by visual-language models during fine-tuning, thereby enhancing the model's robustness in zero-shot tasks.

RAW: A Robust and Agile Plug-and-Play Watermark Framework for AI-Generated Images with Provable Guarantees

Xun Xian (University of Minnesota), Jie Ding (University of Minnesota)

GenerationData SynthesisOptimizationConvolutional Neural NetworkDiffusion modelImageVideo

🎯 What it does: A watermarking framework named RAW is proposed, which can learn embeddable watermarks in AI-generated images and detect watermarks in real-time through a jointly trained classifier, supporting model-agnostic, real-time deployment, and providing provable false positive rate guarantees.

RCDN: Towards Robust Camera-Insensitivity Collaborative Perception via Dynamic Feature-based 3D Neural Modeling

Tianhang Wang (Tongji University), changjun jiang

Object DetectionSegmentationAutonomous DrivingNeural Radiance FieldOptical FlowImageVideo

🎯 What it does: This paper proposes RCDN, an algorithm capable of recovering failed viewpoints and maintaining high accuracy in multi-vehicle collaborative perception scenarios where cameras fail or noise is severe, through collaborative neural rendering.

Real-time Core-Periphery Guided ViT with Smart Data Layout Selection on Mobile Devices

Zhihao Shu (University of Georgia), Wei Niu (University of Georgia)

RecognitionOptimizationComputational EfficiencyTransformerImage

🎯 What it does: Real-time Vision Transformer inference is implemented on mobile devices, guided by the core-edge principle of brain networks, utilizing sparse self-attention and compiler-level data layout optimization.

Real-Time Recurrent Learning using Trace Units in Reinforcement Learning

Esraa Elelimy (University of Alberta), Martha White (University of Alberta)

Recurrent Neural NetworkTransformerReinforcement LearningSequential

🎯 What it does: A lightweight recurrent unit RTU is proposed and evaluated for real-time recursive learning (RTRL) in online reinforcement learning.

Real-Time Selection Under General Constraints via Predictive Inference

Yuyang Huo (Nankai University), Changliang Zou (Nankai University)

OptimizationSupervised Fine-TuningTabular

🎯 What it does: A real-time online sample selection algorithm II-COS is proposed, which can select samples that meet the target response interval while satisfying individual cost constraints and interaction diversity constraints.

Real-time Stereo-based 3D Object Detection for Streaming Perception

Changcai Li (Sun Yat-sen University), Huihui Zhou (Pengcheng Laboratory)

Object DetectionObject TrackingAutonomous DrivingOptical FlowImageVideo

🎯 What it does: A real-time 3D object detection framework called StreamDSGN based on stereo vision is proposed for streaming perception.

Real-world Image Dehazing with Coherence-based Pseudo Labeling and Cooperative Unfolding Network

Chengyu Fang (Shenzhen International Graduate School Tsinghua University), Xiu Li (Shenzhen International Graduate School Tsinghua University)

RestorationKnowledge DistillationImage

🎯 What it does: This paper proposes the Collaborative Unfolding Network (CORUN) and the consistency pseudo-label-based Mean-Teacher framework (Colabator) for achieving high-quality image dehazing in real-world scenarios.

RealCompo: Balancing Realism and Compositionality Improves Text-to-Image Diffusion Models

Xinchen Zhang (Tsinghua University), Bin CUI

GenerationData SynthesisTransformerLarge Language ModelDiffusion modelImageText

🎯 What it does: The RealCompo framework is proposed, which utilizes large language models to generate layouts and integrates pre-trained text-image diffusion models with spatially aware diffusion models (layouts, key points, segmentation, etc.) through a dynamic balancer without training, to achieve a balance between image realism and multi-object composability.

Realizable $H$-Consistent and Bayes-Consistent Loss Functions for Learning to Defer

Anqi Mao (New York University), Yutao Zhong (New York University)

OptimizationTabular

🎯 What it does: This paper studies the problem of Learning to Defer and proposes a family of parameterized alternative loss functions, proving their ability to achieve H-consistency, Bayesian consistency, and H-consistency bounds.

realSEUDO for real-time calcium imaging analysis

Iuliia Dmitrieva (Johns Hopkins University), Adam Shabti Charles

Object DetectionObject TrackingComputational EfficiencyVideo

🎯 What it does: A real-time multiphoton calcium imaging analysis algorithm named realSEUDO has been developed, capable of cell detection and temporal trajectory estimation on streaming video without the need for batch processing.

Reasoning Multi-Agent Behavioral Topology for Interactive Autonomous Driving

Haochen Liu (Nanyang Technological University), Hongyang Li (Shanghai AI Lab)

Autonomous DrivingOptimizationTransformerTabular

🎯 What it does: In autonomous driving, simultaneous multi-agent behavior prediction and planning are achieved by introducing a behavior topology (BeTop) based on winding theory to explicitly supervise the consistency of multi-agent future interactions. Furthermore, a collaborative prediction and planning Transformer network, BeTopNet, is designed to control uncertainty and conflicts.

Reasons and Solutions for the Decline in Model Performance after Editing

Xiusheng Huang (Chinese Academy of Sciences), Kang Liu (Chinese Academy of Sciences)

TransformerLarge Language ModelSupervised Fine-TuningText

🎯 What it does: Through experiments from both data and model perspectives, the reasons for the performance decline after knowledge editing were systematically analyzed, and a new sequence editing method called D4S was proposed to mitigate performance loss.

Reawakening knowledge: Anticipatory recovery from catastrophic interference via structured training

Yanlai Yang (New York University), Mengye Ren (New York University)

Convolutional Neural NetworkTransformerLarge Language ModelSupervised Fine-TuningImageText

🎯 What it does: The study investigates the expected recovery phenomenon observed in LLMs and visual models during cyclic serialized training (fixed document order).

REBEL: Reinforcement Learning via Regressing Relative Rewards

Zhaolin Gao (Cornell University), Wen Sun (Princeton University)

Reinforcement LearningText

🎯 What it does: This paper proposes a reinforcement learning algorithm named REBEL, which achieves policy optimization by regressing on 'relative rewards', simplifying the implementation difficulty of traditional RL schemes.

REBORN: Reinforcement-Learned Boundary Segmentation with Iterative Training for Unsupervised ASR

Liang-Hsuan Tseng (National Taiwan University), Shao-Hua Sun (National Taiwan University)

RecognitionSegmentationConvolutional Neural NetworkReinforcement LearningGenerative Adversarial NetworkAudio

🎯 What it does: In unsupervised speech recognition, an iterative training framework called REBORN is proposed, which alternately trains a segmentation model and a phoneme prediction model.

Reciprocal Learning

Julian Rodemann (LMU Munich), Georg Schollmeyer (LMU Munich)

OptimizationReinforcement Learning from Human Feedback

🎯 What it does: The paper proposes a reciprocal learning framework that unifies algorithms such as self-training, active learning, and multi-armed bandits into a cyclic learning process where data and parameters influence each other.

Reciprocal Reward Influence Encourages Cooperation From Self-Interested Agents

John Luoyu Zhou, Jonathan Kao

Reinforcement Learning

🎯 What it does: This paper proposes a reinforcement learning agent named Reciprocator, which encourages self-interested agents to cooperate through intrinsic reciprocal rewards.

Recognize Any Regions

Haosen Yang (University of Surrey), Xiatian Zhu (University of Surrey)

RecognitionObject DetectionTransformerPrompt EngineeringImage

🎯 What it does: Proposes the RegionSpot model, which utilizes frozen SAM and CLIP for cross-model alignment of location awareness and semantic features, achieving efficient open-world region recognition;

Reconstruct and Match: Out-of-Distribution Robustness via Topological Homogeneity

Chaoqi Chen (Shenzhen University), Hui Huang (Shenzhen University)

ClassificationDomain AdaptationGraph Neural NetworkContrastive LearningImage

🎯 What it does: This paper proposes the REconstruct and Match (REMA) framework, which first extracts sparse main components of images through a self-supervised Slot-Attention reconstruction module (SSR), and then utilizes a hypergraph-based higher-order relationship reasoning module (HORR) to match the component topology across different domains, thereby enhancing the model's robustness in out-of-distribution (OOD) scenarios.

Reconstructing the Image Stitching Pipeline: Integrating Fusion and Rectangling into a Unified Inpainting Model

Xieziqi, Ning Jia (Tongji University)

Image TranslationRestorationDiffusion modelImage

🎯 What it does: This paper proposes SRStitcher, which unifies the two stages of image stitching—fusion and rectification—into a single filling task within a diffusion model, eliminating the issues of multi-stage training and error accumulation.

Reconstruction Attacks on Machine Unlearning: Simple Models are Vulnerable

Martin Andres Bertran, Steven Wu

ClassificationSafty and PrivacyImageTabular

🎯 What it does: This study investigates reconstruction attacks on simple models (such as linear regression, embedding + linear layer, arbitrary loss) after machine unlearning (data deletion), which can recover deleted samples from the differences in model parameters before and after deletion.

Reconstruction of Manipulated Garment with Guided Deformation Prior

Ren Li (École Polytechnique Fédérale de Lausanne), Pascal Fua (École Polytechnique Fédérale de Lausanne)

RestorationGenerationTransformerDiffusion modelVideoPoint CloudMesh

🎯 What it does: An algorithm is proposed to recover complete 3D meshes from incomplete 3D point clouds of non-worn and manipulated garments using the ISP model, an extended diffusion deformation prior, and a UV mapping network.

Recovering Complete Actions for Cross-dataset Skeleton Action Recognition

Hanchao Liu (Tsinghua University), Shi-min Hu

RecognitionDomain AdaptationGraph Neural NetworkContrastive LearningVideo

🎯 What it does: By restoring complete actions and resampling, cross-dataset skeletal action enhancement is generated to achieve single-domain generalization.

RectifID: Personalizing Rectified Flow with Anchored Classifier Guidance

Zhicheng Sun (Peking University), Yadong MU

GenerationData SynthesisDiffusion modelFlow-based ModelRectified FlowImage

🎯 What it does: This paper proposes a personalized image generation method that requires no training, utilizing Anchored Classifier Guidance to guide Rectified Flow, which can maintain identity consistency based on reference images provided by users.

Recurrent Complex-Weighted Autoencoders for Unsupervised Object Discovery

Anand Gopalakrishnan (Swiss AI Lab), Michael Curtis Mozer (Google DeepMind)

Object DetectionConvolutional Neural NetworkAuto EncoderImage

🎯 What it does: A fully convolutional autoencoder (SynCx) is proposed, utilizing complex weights and recursive phase updates to achieve unsupervised object detection.

Recurrent neural network dynamical systems for biological vision

Wayne WM Soo, Xiao-Jing Wang (New York University)

ClassificationRecognitionConvolutional Neural NetworkRecurrent Neural NetworkImageOrdinary Differential Equation

🎯 What it does: A hybrid architecture that integrates convolutional neural networks with continuous-time recurrent dynamic systems (CordsNet) is proposed, achieving a unification of image processing and biological dynamics.

Recurrent neural networks: vanishing and exploding gradients are not the end of the story

Nicolas Zucchet (ETH Zürich), Antonio Orvieto (MPI for Intelligent Systems)

Recurrent Neural NetworkTextSequential

🎯 What it does: This paper studies the 'memory curse' that occurs in recurrent neural networks (RNNs) when learning long-term sequential memories—where an increase in the network's memory length leads to extreme sensitivity of hidden states and gradients to small parameter changes, even in the absence of gradient explosion.

Recurrent Reinforcement Learning with Memoroids

Steven Morad (University of Macau), Amanda Prorok (University of Cambridge)

Recurrent Neural NetworkTransformerReinforcement LearningSequentialBenchmark

🎯 What it does: This paper proposes the memoroid framework, unifying various efficient recursive models, and based on this, designs Tape-Based Batching (TBB) to eliminate the zero-padding and gradient truncation issues of traditional Segment-Based Batching (SBB), thereby significantly improving sample efficiency.

Recursive Introspection: Teaching Language Model Agents How to Self-Improve

Yuxiao Qu (Carnegie Mellon University), Aviral Kumar (MultiOn)

Knowledge DistillationReinforcement Learning from Human FeedbackTransformerLarge Language ModelSupervised Fine-TuningReinforcement LearningAgentic AIText

🎯 What it does: This paper proposes a fine-tuning method called RISE, which trains large language models to self-correct and improve answers in multi-turn interactions.

Recursive PAC-Bayes: A Frequentist Approach to Sequential Prior Updates with No Information Loss

Yi-Shan Wu (University of South Denmark), Yevgeny Seldin (University of Copenhagen)

ClassificationOptimizationImage

🎯 What it does: A recursive PAC-Bayes framework is proposed, allowing for sequential prior updates without losing confidence information, and corresponding PAC-Bayes bounds are provided;

Reducing Transformer Key-Value Cache Size with Cross-Layer Attention

William Brandon (Massachusetts Institute of Technology), Jonathan Ragan-Kelley (Massachusetts Institute of Technology)

CompressionOptimizationTransformerLarge Language ModelText

🎯 What it does: Proposes a Cross-Layer Attention (CLA) mechanism that allows adjacent layers in the Transformer to share the same set of key/value activations, thereby halving the KV cache size without significantly affecting language modeling accuracy.

REDUCR: Robust Data Downsampling using Class Priority Reweighting

William Bankes (University College London), Zi Wang (Google DeepMind)

OptimizationData-Centric LearningConvolutional Neural NetworkTransformerSupervised Fine-TuningImage

🎯 What it does: This paper proposes an online batch selection method called REDUCR, which implements robust data downsampling using class priority re-weighting, significantly improving the generalization performance of the worst class while reducing the amount of training data.

ReEvo: Large Language Models as Hyper-Heuristics with Reflective Evolution

Haoran Ye (Peking University), Guojie Song (Peking University)

OptimizationTransformerLarge Language ModelPrompt EngineeringText

🎯 What it does: A language hyper-heuristic framework called ReEvo is proposed, which uses large language models to generate and evolve solutions for combinatorial optimization problems, capable of automatically designing efficient heuristic methods with limited sample sizes.

ReF-LDM: A Latent Diffusion Model for Reference-based Face Image Restoration

Chi-Wei Hsiao (MediaTek), Chia-Ping Chen (MediaTek)

RestorationDiffusion modelImage

🎯 What it does: A reference image-driven face image restoration method based on latent diffusion models, ReF-LDM, is proposed, which can generate high-definition restored images consistent with the original person's identity using multiple non-aligned high-quality reference images.

RefDrop: Controllable Consistency in Image or Video Generation via Reference Feature Guidance

Jiaojiao Fan (Georgia Institute of Technology), Yongxin Chen (Georgia Institute of Technology)

GenerationData SynthesisDiffusion modelImageVideo

🎯 What it does: By incorporating reference feature guidance into the self-attention of the diffusion model UNet, controllable consistency generation that is independent of training has been achieved.

Reference Trustable Decoding: A Training-Free Augmentation Paradigm for Large Language Models

Shi Luohe (Wuhan University), hai zhao

GenerationRetrievalTransformerLarge Language ModelTextRetrieval-Augmented Generation

🎯 What it does: A training-free, retrieval-based decoding method (Reference Trustable Decoding, RTD) is proposed, which constructs a reference dataset and adjusts the output distribution based on similarity during the decoding phase, allowing LLMs to quickly adapt to downstream tasks while keeping the original parameters unchanged.

Referencing Where to Focus: Improving Visual Grounding with Referential Query

Yabing Wang (Xi'an Jiaotong University), Le Wang (Xi'an Jiaotong University)

Object DetectionSegmentationTransformerVision Language ModelImageMultimodality

🎯 What it does: This paper proposes the RefFormer model, which utilizes a Query Adaptation (QA) module to generate target-related reference queries within the CLIP backbone, and uses these queries in the Transformer decoder to achieve visual localization.

Referring Human Pose and Mask Estimation In the Wild

Bo Miao (University of Western Australia), Ajmal Saeed Mian

Object DetectionSegmentationPose EstimationTransformerVision Language ModelImageTextMultimodality

🎯 What it does: Proposed and implemented the 'R-HPM' (Referring Human Pose and Mask Estimation) task, which can accurately predict the joint location information and segmentation mask of a specified person through text or location prompts (scribble/point).

ReFIR: Grounding Large Restoration Models with Retrieval Augmentation

Hang Guo (Tsinghua University), Shu-Tao Xia (Tsinghua University)

RestorationRetrievalDiffusion modelImage

🎯 What it does: A training-free retrieval-enhanced framework called ReFIR is proposed, which uses high-quality reference images obtained through retrieval to help large image restoration models avoid 'hallucinations' and generate more realistic details.

Reflective Multi-Agent Collaboration based on Large Language Models

Xiaohe Bo (Renmin University of China), Ji-Rong Wen (Renmin University of China)

Reinforcement Learning from Human FeedbackTransformerLarge Language ModelSupervised Fine-TuningReinforcement LearningText

🎯 What it does: The COPPER framework is proposed to enhance the collaborative performance of multi-agent LLMs through self-reflection, utilizing shared reflectors and counterfactual PPO for efficient training.

ReFT: Representation Finetuning for Language Models

Zhengxuan Wu (Stanford University), Christopher Potts (Stanford University)

Explainability and InterpretabilityComputational EfficiencyRepresentation LearningTransformerLarge Language ModelSupervised Fine-TuningText

🎯 What it does: This paper proposes and evaluates a parameter-efficient fine-tuning method through low-rank linear subspace intervention (ReFT) on the hidden representations of pre-trained language models, focusing on LoReFT and its efficient variant DiReFT.

Refusal in Language Models Is Mediated by a Single Direction

Andy Arditi (Independent), Neel Nanda

Large Language ModelText

🎯 What it does: The rejection mechanism in large language models was studied and found to be driven by a single direction.

RegExplainer: Generating Explanations for Graph Neural Networks in Regression Tasks

Jiaxing Zhang (New Jersey Institute of Technology), Hua Wei (Arizona State University)

Explainability and InterpretabilityGraph Neural NetworkContrastive LearningGraph

🎯 What it does: A post-hoc explanation method named RegExplainer is proposed for explaining graph neural networks in graph regression tasks.

Regression under demographic parity constraints via unlabeled post-processing

Gayane Taturyan (Universite Gustave Eiffel), Mohamed Hebiri (Universite Gustave Eiffel)

OptimizationTabular

🎯 What it does: This paper proposes a post-processing algorithm for a regression model that does not require inferring sensitive attributes, ensuring fair predictions under demographic balance constraints.

Regret Minimization in Stackelberg Games with Side Information

Keegan Harris (Carnegie Mellon University), Maria Florina Balcan

OptimizationReinforcement LearningTabular

🎯 What it does: This paper studies Stackelberg games with side information and proposes a new algorithmic framework that considers the interaction between a leader and a series of followers in an online environment, where the leader chooses a strategy after observing contextual information in each round.

ReGS: Reference-based Controllable Scene Stylization with Gaussian Splatting

Yiqun Mei (Johns Hopkins University), Vishal M. Patel (Johns Hopkins University)

Image TranslationGenerationGaussian SplattingImage

🎯 What it does: Utilizing a pre-trained 3D Gaussian Splatting (3DGS) model for reference image-driven controllable stylization of scenes, achieving real-time viewpoint synthesis.

Regularized Adaptive Momentum Dual Averaging with an Efficient Inexact Subproblem Solver for Training Structured Neural Network

Zih-Syuan Huang (National Taiwan University), Ching-pei Lee (Institute of Statistical Mathematics)

OptimizationConvolutional Neural NetworkTransformerImageTextAudio

🎯 What it does: A new Regularized Adaptive Dual Equilibrium (RAMDA) algorithm is proposed for training deep neural networks with structures (such as sparsity), and an implementable subproblem approximation solver is designed for this algorithm.

Regularized Conditional Diffusion Model for Multi-Task Preference Alignment

Xudong Yu (Harbin Institute of Technology), Xuelong Li (Institute of Artificial Intelligence China Telecom)

Recommendation SystemTransformerReinforcement LearningDiffusion modelSequential

🎯 What it does: In offline multi-task reinforcement learning, preference labels are used to learn trajectory representations, which are then employed as conditions to guide the diffusion model in generating trajectories consistent with preferences.

Regularized Q-Learning

Han-Dong Lim (Korea Advanced Institute of Science and Technology), Donghwan Lee (Korea Advanced Institute of Science and Technology)

Reinforcement LearningOrdinary Differential Equation

🎯 What it does: A regularized linear function approximation Q-learning algorithm, RegQ, is proposed, ensuring convergence.

Regularizing Hidden States Enables Learning Generalizable Reward Model for LLMs

Rui Yang (University of Illinois Urbana-Champaign), Tong Zhang (University of Illinois Urbana-Champaign)

Reinforcement Learning from Human FeedbackTransformerLarge Language ModelSupervised Fine-TuningReinforcement LearningText

🎯 What it does: This paper proposes enhancing the generalization of the reward model to unseen data by adding text generation regularization (GRM) in the hidden layers of the reward model, thereby alleviating the issue of reward over-optimization in RLHF.

Reimagining Mutual Information for Enhanced Defense against Data Leakage in Collaborative Inference

Lin Duan (Duke University), Maria Gorlatova (Duke University)

Federated LearningConvolutional Neural NetworkGenerative Adversarial NetworkImage

🎯 What it does: A defense method named InfoScissors is developed to prevent data leakage by minimizing the mutual information between intermediate results in collaborative inference and the input and predictions.

Reinforced Cross-Domain Knowledge Distillation on Time Series Data

QING XU, Zhenghua Chen (Institute for Infocomm Research A*STAR)

Domain AdaptationKnowledge DistillationReinforcement LearningTime Series

🎯 What it does: This paper proposes an end-to-end cross-domain knowledge distillation framework RCD-KD, which combines reinforcement learning to dynamically select target domain samples and an adversarial domain discriminator to achieve unsupervised domain adaptation for lightweight models in time series tasks.

Reinforcement Learning Gradients as Vitamin for Online Finetuning Decision Transformers

Kai Yan (University of Illinois Urbana-Champaign), Yu-Xiong Wang (University of Illinois Urbana-Champaign)

TransformerReinforcement Learning

🎯 What it does: This paper studies the issue of online fine-tuning of the Decision Transformer after offline pre-training and proposes combining TD3 reinforcement learning gradients with the original self-supervised training to enhance the online fine-tuning effect.

Reinforcement Learning Guided Semi-Supervised Learning

Marzi Heidari (Carleton University), Yuhong Guo (Carleton University)

ClassificationReinforcement LearningImage

🎯 What it does: A semi-supervised learning method based on reinforcement learning, RLGSSL, is proposed, which utilizes RL rewards to guide pseudo-label generation and model learning, significantly improving classification performance in scenarios with scarce labels.