NeurIPS 2025 Papers — Page 37
Conference on Neural Information Processing Systems · 5275 papers
PUATE: Efficient ATE Estimation from Treated (Positive) and Unlabeled Units
Masahiro Kato (Mizuho-DL Financial Technology Co., Ltd.), Ryo Inokuchi
Tabular
🎯 What it does: In the scenario where the missing treatment indicator is absent (only the treated group and the unknown group are observed), a semi-parametric efficient estimator for estimating the Average Treatment Effect (ATE) is proposed, along with the corresponding efficiency lower bound.
PubSub-VFL: Towards Efficient Two-Party Split Learning in Heterogeneous Environments via Publisher/Subscriber Architecture
Yi Liu (City University of Hong Kong), Cong Wang (ByteDance Inc.)
Federated LearningComputational EfficiencyTabular
🎯 What it does: A two-party vertical federated learning framework named PubSub-VFL is proposed, utilizing a publish/subscribe architecture to achieve asynchronous and efficient training, addressing computational bottlenecks caused by resource and data heterogeneity;
Puppeteer: Rig and Animate Your 3D Models
Chaoyue Song (Nanyang Technological University), Jianfeng Zhang (ByteDance Seed)
GenerationPose EstimationOptimizationTransformerReinforcement LearningVideoMesh
🎯 What it does: A Puppeteer framework is proposed, achieving a fully automated pipeline from static 3D models to complete animated assets, including automatic skeleton generation, skin weight prediction, and video-guided animation.
Purest Quantum State Identification
Yingqi Yu (University of Science and Technology of China), Xiangyang Li
RecognitionOptimizationPhysics Related
🎯 What it does: A method for pure quantum state identification (PQSI) is proposed, aimed at identifying quantum states that are least affected by noise to improve the accuracy of quantum computing and communication.
Purifying Approximate Differential Privacy with Randomized Post-processing
Yingyu Lin (University of California, San Diego), Yu-Xiang Wang (University of California, San Diego)
OptimizationSafty and PrivacyTabular
🎯 What it does: A 'purification' framework is proposed, which uses random post-processing to convert (ε,δ)-approximate differential privacy mechanisms into pure differential privacy (ε,0) mechanisms.
Purifying Shampoo: Investigating Shampoo's Heuristics by Decomposing its Preconditioner
Runa Eschenhagen (University of Cambridge), Hao-Jun Michael Shi (Meta Superintelligence Labs)
OptimizationTransformerImageMagnetic Resonance Imaging
🎯 What it does: The study removes the learning rate grafting and old preprocessor heuristics in the Shampoo optimizer, proposing an improved method based on eigenvalue correction and adaptive feature basis updates.
Purity Law for Neural Routing Problem Solvers with Enhanced Generalizability
Wenzhao Liu (University of Chinese Academy of Sciences), Tiande Guo (University of Chinese Academy of Sciences)
OptimizationReinforcement LearningTabular
🎯 What it does: Proposes the Purity Law and designs the PUPO training framework based on this law to enhance the generalization ability of neural routing solvers.
PurpCode: Reasoning for Safer Code Generation
Jiawei Liu (University of Illinois Urbana-Champaign), Gang Wang (University of Illinois Urbana-Champaign)
Safty and PrivacyTransformerLarge Language ModelSupervised Fine-TuningReinforcement LearningText
🎯 What it does: PurpCode is proposed, a post-training method that enables code LLMs to perform network security reasoning through rule learning and reinforcement learning, capable of generating vulnerability-free code and rejecting malicious requests.
Put CASH on Bandits: A Max K-Armed Problem for Automated Machine Learning
Amir Rezaei Balef (University of Tübingen), Katharina Eggensperger (University of Tübingen)
OptimizationHyperparameter SearchTabular
🎯 What it does: This paper proposes a maximum k-armed bandit method—MaxUCB—for the joint algorithm selection and hyperparameter optimization (CASH) problem in automated machine learning (AutoML). The method decides on the model and its hyperparameters by calling a single HPO iteration during each round of budget allocation, aiming to maximize the observed best performance.
Puzzles: Unbounded Video-Depth Augmentation for Scalable End-to-End 3D Reconstruction
Jiahao Ma (Australian National University), Chuong Nguyen (Data61/CSIRO)
Data SynthesisDepth EstimationImageVideo
🎯 What it does: A data augmentation framework called Puzzles is proposed, which slices and rearranges a single image or short video, simulating camera motion to generate unbounded virtual videos with pose and depth for training end-to-end 3D reconstruction models.
PyraMotion: Attentional Pyramid-Structured Motion Integration for Co-Speech 3D Gesture Synthesis
Zhizhuo Yin (Hong Kong University of Science and Technology), Pan Hui (Hong Kong University of Science and Technology)
GenerationData SynthesisTransformerAuto EncoderMultimodalityAudio
🎯 What it does: A framework for audio-driven full-body gesture synthesis based on multi-scale discrete encoding, called PyraMotion, is proposed, and an Attentional Pyramidal VQ-VAE (APVQ-VAE) is designed for multi-scale gesture discretization and reconstruction.
Q-Insight: Understanding Image Quality via Visual Reinforcement Learning
Weiqi Li (Peking University), Jian Zhang (Peking University)
RecognitionOptimizationExplainability and InterpretabilityReinforcement LearningVision Language ModelImage
🎯 What it does: Q-Insight is proposed, a multi-task visual language model based on GRPO for score regression of image quality and degradation perception, capable of generating interpretable reasoning steps.
Q-Palette: Fractional-Bit Quantizers Toward Optimal Bit Allocation for Efficient LLM Deployment
Deokjae Lee (Seoul National University), Hyun Oh Song (Seoul National University)
CompressionOptimizationComputational EfficiencyTransformerLarge Language ModelText
🎯 What it does: This paper addresses the problem of post-training quantization (PTQ) for large language models (LLMs) and proposes a customizable set of score bit quantizers called Q-Palette, which is embedded in the mixed scheme quantization (MSQ) framework to achieve better compression and inference speed under memory or latency constraints.
Q3R: Quadratic Reweighted Rank Regularizer for Effective Low-Rank Training
Ipsita Ghosh (University of Central Florida), Christian Kümmerle (University of Central Florida)
OptimizationTransformerSupervised Fine-TuningImageText
🎯 What it does: This paper proposes a quadratic reweighted rank regularization (Q3R) based on the Iteratively Reweighted Least Squares (IRLS) framework, which induces a low-rank structure in the network weight matrix through regularization during training. It can be used for low-rank pre-training of large-scale models as well as for parameter-efficient fine-tuning.
QBasicVSR: Temporal Awareness Adaptation Quantization for Video Super-Resolution
Zhenwei Zhang (Tianjin University), Yanming Hui (Tianjin University)
RestorationSuper ResolutionComputational EfficiencySupervised Fine-TuningVideo
🎯 What it does: This paper proposes the first post-training quantization (PTQ) framework for video super-resolution (VSR) called QBasicVSR. It utilizes two modules: flow gradient video bit adaptive (FG-VBA) and temporal shared layer bit adaptive (TS-LBA), combined with supervised fine-tuning of the full-precision model. It achieves 4-bit quantization using only a small amount of unlabeled video calibration data, significantly improving inference speed and memory usage while maintaining or even surpassing the reconstruction quality of FP32 models.
QFFT, Question-Free Fine-Tuning for Adaptive Reasoning
Wanlong Liu (University of Electronic Science and Technology of China), Benyou Wang (Huawei Noah's Ark Lab)
TransformerLarge Language ModelSupervised Fine-TuningTextChain-of-Thought
🎯 What it does: A question-free fine-tuning (QFFT) method is proposed, enabling large language models to use concise short chain thinking (Short CoT) for simple questions and automatically switch to reflective long chain thinking (Long CoT) for difficult questions, thus achieving efficient and accurate reasoning.
QiMeng-CodeV-R1: Reasoning-Enhanced Verilog Generation
Yaoyu Zhu (Institute of Computing Technology, Chinese Academy of Sciences), Yunji Chen (Institute of Computing Technology, Chinese Academy of Sciences)
AI Code AssistantTransformerLarge Language ModelReinforcement LearningText
🎯 What it does: Developed the CodeV-R1 framework, which first obtains a reasoning-capable Verilog generation LLM through knowledge distillation, and then conducts reinforcement learning on high-quality equivalence detection data.
QiMeng-MuPa: Mutual-Supervised Learning for Sequential-to-Parallel Code Translation
Changxin Ke (State Key Laboratory of Processors, Institute of Computing Technology, Chinese Academy of Sciences), Yunji Chen (State Key Laboratory of Processors, Institute of Computing Technology, Chinese Academy of Sciences)
AI Code AssistantTransformerLarge Language ModelSequential
🎯 What it does: A mutual supervision learning framework named QiMeng-MuPa is proposed to automatically translate sequential code into parallel code and ensure functional equivalence through unit testing.
QiMeng-NeuComBack: Self-Evolving Translation from IR to Assembly Code
Hainan Fang (Chinese Academy of Sciences), Yunji Chen (Chinese Academy of Sciences)
OptimizationAI Code AssistantLarge Language ModelPrompt EngineeringTextBenchmark
🎯 What it does: Proposed the NeuComBack benchmark and self-evolving prompt optimization method, achieving neural compilation from LLVM IR to assembly;
QiMeng-SALV: Signal-Aware Learning for Verilog Code Generation
Yang Zhang (Institute of Computing Technology), Yunji Chen (Institute of Computing Technology)
GenerationAI Code AssistantReinforcement Learning from Human FeedbackLarge Language ModelSupervised Fine-TuningReinforcement LearningText
🎯 What it does: A Verilog code generation method based on signal-level reinforcement learning, QiMeng-SALV, is proposed, which utilizes signal implementations in partially correct modules to provide functional rewards, improving RL training.
QoQ-Med: Building Multimodal Clinical Foundation Models with Domain-Aware GRPO Training
Wei Dai (Massachusetts Institute of Technology), Paul Pu Liang (Massachusetts Institute of Technology)
OptimizationExplainability and InterpretabilityReinforcement Learning from Human FeedbackTransformerLarge Language ModelReinforcement LearningImageMultimodalityTime SeriesBiomedical DataMagnetic Resonance ImagingComputed TomographyUltrasoundElectrocardiogram
🎯 What it does: A multimodal clinical foundation model QoQ-Med has been constructed, capable of reasoning across medical images, time series signals, and text, and a Domain-Aware Relative Policy Optimization (DRPO) method has been proposed to balance the data imbalance issues across different modalities and domains.
QSCA: Quantization with Self-Compensating Auxiliary for Monocular Depth Estimation
Jincheol Yang (Sogang University), Suk-Ju Kang (Sogang University)
Depth EstimationKnowledge DistillationTransformerImage
🎯 What it does: A QSCA framework is proposed, embedding a lightweight Self-Compensating Auxiliary (SCA) module into the Transformer encoder and DPT decoder, performing 4-bit pure post-training quantization on large monocular depth estimation models, and recovering quantization loss on unlabeled data through self-supervised distillation.
QSVD: Efficient Low-rank Approximation for Unified Query-Key-Value Weight Compression in Low-Precision Vision-Language Models
Yutong Wang (New York University), Sai Qian Zhang (New York University)
CompressionComputational EfficiencyTransformerVision Language ModelMultimodality
🎯 What it does: Designed and implemented the QSVD method, which jointly compresses the query-key-value (QKV) weights of the visual-language model (VLM) using SVD, and combines low-precision quantization to significantly reduce model parameters, KV cache, and computational load while maintaining or even improving inference accuracy.
QuadEnhancer: Leveraging Quadratic Transformations to Enhance Deep Neural Networks
Qian Chen (Chinese University of Hong Kong), Yin Zhang (Chinese University of Hong Kong)
ClassificationOptimizationComputational EfficiencyTransformerLarge Language ModelSupervised Fine-TuningImageText
🎯 What it does: Insert a lightweight quadratic enhancement module into each linear layer of the neural network to capture the quadratic interactions between features.
Quadratic Coreset Selection: Certifying and Reconciling Sequence and Token Mining for Efficient Instruction Tuning
Ziliang Chen (Peng Cheng Laboratory), Liang Lin (Sun Yat-sen University)
OptimizationTransformerSupervised Fine-TuningText
🎯 What it does: This paper proposes a new Quadratic Coreset Selection (QCS) framework for jointly selecting instruction-response sequences and key tokens within them during instruction tuning to enhance data efficiency.
QuadricFormer: Scene as Superquadrics for 3D Semantic Occupancy Prediction
Sicheng Zuo (Tsinghua University), Jiwen Lu (Tsinghua University)
SegmentationAutonomous DrivingComputational EfficiencyTransformerMixture of ExpertsPoint Cloud
🎯 What it does: Proposes a sparse object center representation based on superquadrics, achieving efficient 3D semantic occupancy prediction.
Quality-Driven Curation of Remote Sensing Vision-Language Data via Learned Scoring Models
Dilxat Muhtar (Nanjing University), Xueliang Zhang (Nanjing University)
RetrievalTransformerLarge Language ModelReinforcement LearningVision Language ModelContrastive LearningImageTextMultimodality
🎯 What it does: This paper proposes ScoreRS, a learning-based scoring model for remote sensing visual-language data quality assessment, and constructs a large-scale RS visual-language preference dataset. ScoreRS is used for data filtering, reinforcement learning, and best N selection to enhance the performance of VLM in RS tasks.
QuanDA: Quantile-Based Discriminant Analysis for High-Dimensional Imbalanced Classification
Qian Tang (University of Minnesota), Boxiang Wang (University of Iowa)
ClassificationSupervised Fine-TuningTabular
🎯 What it does: A high-dimensional imbalanced binary classification method called QuanDA based on quantile regression is proposed.
Quantifying and Alleviating Co-Adaptation in Sparse-View 3D Gaussian Splatting
Kangjie Chen (Tsinghua University), Haoqian Wang (Tsinghua University)
RestorationGenerationGaussian SplattingImage
🎯 What it does: Analyze the source of appearance artifacts generated by 3D Gaussian Spraying (3DGS) from a sparse perspective, and propose to improve new view rendering quality by suppressing the co-adaptation between Gaussians.
Quantifying Cross-Modality Memorization in Vision-Language Models
Yuxin Wen (University of Maryland), Chiyuan Zhang (Google)
Data SynthesisRepresentation LearningTransformerSupervised Fine-TuningVision Language ModelImageTextMultimodality
🎯 What it does: This study investigates the characteristics of visual-language models in cross-modal memory, quantifying the transfer and gap of factual knowledge between images and text.
Quantifying Distributional Invariance in Causal Subgraph for IRM-Free Graph Generalization
Yang Qiu (Huazhong University of Science and Technology), Ruixuan Li (Huazhong University of Science and Technology)
Domain AdaptationDrug DiscoveryGraph Neural NetworkGraph
🎯 What it does: A graph neural network method without IRM, called IDG, has been developed, which utilizes subgraph representation norm optimization to identify causal subgraphs and achieve OOD generalization.
Quantifying Elicitation of Latent Capabilities in Language Models
Elizabeth Donoway (University of California), Jan Leike (Anthropic)
ClassificationGenerationTransformerLarge Language ModelSupervised Fine-TuningText
🎯 What it does: The study investigates the minimum number of trainable parameters required to activate the potential capabilities of large language models and constructs an 'activation frontier' to describe the relationship between parameters and performance; fine-tuning experiments are conducted using a randomly selected small number of parameters.
Quantifying Statistical Significance of Deep Nearest Neighbor Anomaly Detection via Selective Inference
Mizuki Niihori (Nagoya University), Ichiro Takeuchi (Nagoya University)
Anomaly DetectionConvolutional Neural NetworkImageTabular
🎯 What it does: A statistical significance quantification method for deep k-nearest neighbor anomaly detection (Deep k-NN AD) based on selective inference (SI) is proposed, providing p-values for detection results and enabling controllable false positive rates.
Quantifying Task-relevant Similarities in Representations Using Decision Variable Correlations
Yu Eric Qian, Xue-Xin Wei (University of Texas at Austin)
ClassificationRepresentation LearningConvolutional Neural NetworkImageMultimodality
🎯 What it does: A new metric method called Decision Variable Correlation (DVC) is proposed to evaluate the consistency of two neural representations in decision-making strategies for classification tasks.
Quantifying Uncertainty in Error Consistency: Towards Reliable Behavioral Comparison of Classifiers
Thomas Klein (Max Planck Institute for Intelligent Systems), Kristof Meding (CSZ Institute for Artificial Intelligence and Law)
ClassificationGenerative Adversarial NetworkImage
🎯 What it does: In the experiment, uncertainty quantification of error consistency (EC) is performed, proposing the use of bootstrap methods to calculate confidence intervals and conduct significance tests; simultaneously, a generative model is constructed to decompose EC into replication probability and accuracy mismatch, which is used for experimental design and sample size planning.
Quantifying Uncertainty in the Presence of Distribution Shifts
Yuli Slavutsky (Columbia University), David Blei
ClassificationDomain AdaptationTabular
🎯 What it does: A Bayesian framework based on adaptive priors (VIDS) is proposed to estimate predictive uncertainty when there is covariate distribution shift between the training and testing sets.
Quantile Reward Policy Optimization: Alignment with Pointwise Regression and Exact Partition Functions
Simon Matrenok (École Polytechnique Fédérale de Lausanne), Caglar Gulcehre (École Polytechnique Fédérale de Lausanne)
OptimizationReinforcement LearningText
🎯 What it does: This paper proposes the Quantile Reward Policy Optimization (QRPO) method, which utilizes quantile rewards to analytically derive the partition function of the KL regularized RL objective, allowing for the direct use of absolute point rewards in offline or offline + heterogeneous data RL fine-tuning within the policy fitting framework.
Quantitative convergence of trained neural networks to Gaussian processes
Andrea Agazzi (University of Bern), Dario Trevisan (University of Pisa)
Tabular
🎯 What it does: The study quantifies the convergence between single hidden layer neural networks with limited width and their corresponding Gaussian processes under gradient descent training (measured using the quadratic Wasserstein distance), providing explicit upper bounds regarding network width, training time, and the properties of activation functions.
Quantization Error Propagation: Revisiting Layer-Wise Post-Training Quantization
Yamato Arai (Fujitsu Limited), Yuma Ichikawa (Fujitsu Limited)
Text
🎯 What it does: Proposes the QEP framework to improve layer-wise PTQ, addressing the performance degradation caused by the accumulation of quantization errors in multi-layer networks;
Quantization-Free Autoregressive Action Transformer
Ziyad Sheebaelhamd (University of Tuebingen), Claire Vernade (University of Tuebingen)
Autonomous DrivingRobotic IntelligenceTransformerReinforcement LearningMultimodality
🎯 What it does: A Quantization-Free Autoregressive Action Transformer (Q-FAT) is proposed, which directly models actions in continuous action spaces and predicts action distributions using GMM, achieving behavior cloning for robotic simulation tasks.
Quantum Doubly Stochastic Transformers
Jannis Born (IBM Research), Aleksandros Sobczyk (IBM Research)
TransformerImage
🎯 What it does: This paper proposes and implements the Quantum Doubly Stochastic Transformer (QDSFormer), which replaces the Softmax of the Transformer with a parameterized quantum circuit QontOT to generate a Doubly Stochastic Matrix (DSM) attention matrix, and conducts experimental validation on Vision Transformer.
Quantum speedup of non-linear Monte Carlo problems
Jose Blanchet (Stanford University), Guanyang Wang (Rutgers University)
OptimizationComputational EfficiencyDrug DiscoveryPhysics Related
🎯 What it does: A multi-layer Monte Carlo (QM-MLMC) algorithm for quantum inner-layer quantum acceleration is proposed to solve nested expectation nonlinear problems, achieving a near-optimal computational complexity of ˜O(1/ε);
Quantum Speedups for Minimax Optimization and Beyond
Chengchang Liu (Chinese University of Hong Kong), John C.S. Lui (Chinese University of Hong Kong)
Optimization
🎯 What it does: This paper studies convex-concave minimization optimization problems that only allow function value access, and proposes a class of Hessian-aware quantum zero-order methods that can find ε-saddle points in O(d^(2/3)/ε^(2/3)) function value oracle calls.
Quantum Visual Fields with Neural Amplitude Encoding
Shuteng Wang (Max Planck Institute for Informatics), Vladislav Golyanik (Max Planck Institute for Informatics)
RestorationRepresentation LearningImagePoint CloudPhysics Related
🎯 What it does: A new Quantum Implicit Neural Representation (QINR) architecture called QVF is proposed for coordinate-based continuous representation learning of 2D images and 3D shapes, supporting interpolation, inpainting, and completion of images/shapes.
QuARI: Query Adaptive Retrieval Improvement
Eric Xing (Washington University in St. Louis), Nathan Jacobs (Washington University in St. Louis)
RetrievalTransformerVision Language ModelContrastive LearningImageTextMultimodality
🎯 What it does: A query adaptive retrieval framework QuARI is proposed, which utilizes transformers to generate query-specific linear projections to improve image-to-image and text-to-image retrieval.
Quartet: Native FP4 Training Can Be Optimal for Large Language Models
Roberto L. Castro, Dan Alistarh (IST Austria)
OptimizationComputational EfficiencyTransformerLarge Language ModelText
🎯 What it does: This paper proposes and implements a fully low-precision 4-bit floating-point training framework called Quartet, which can complete the full training of large language models on NVIDIA Blackwell GPUs while maintaining accuracy comparable to or even better than FP16.
Quasi-Self-Concordant Optimization with $\ell_{\infty}$ Lewis Weights
Alina Ene (Boston University), Adrian Vladu (Université Paris Cité)
OptimizationTabular
🎯 What it does: A fast algorithm for solving the quantum self-covariance (QSC) optimization problem based on trust regions and IRLS is proposed, where the objective function is a summation form composed of polynomial QSC components.
Query-Efficient Locally Private Hypothesis Selection via the Scheffe Graph
Gautam Kamath (University of Waterloo and Vector Institute), David Woodruff
OptimizationSafty and Privacy
🎯 What it does: An algorithm for non-interactive hypothesis selection under local differential privacy (LDP) constraints is proposed, which can achieve approximate matching to an unknown distribution p with only about O(k^{3/2}) times polylog(k) queries under the premise of near-optimality.
R-KV: Redundancy-aware KV Cache Compression for Reasoning Models
Zefan Cai (University of Wisconsin - Madison), Junjie Hu (University of Wisconsin - Madison)
CompressionComputational EfficiencyTransformerLarge Language ModelText
🎯 What it does: A KV cache compression method for inference models, R-KV, is proposed, which dynamically eliminates useless tokens during the inference process using a dual evaluation of importance and redundancy, significantly reducing KV cache usage.
R$^2$ec: Towards Large Recommender Models with Reasoning
Runyang You (Hong Kong Polytechnic University), Liqiang Nie (Harbin Institute of Technology)
Recommendation SystemTransformerLarge Language ModelReinforcement LearningContrastive LearningTabular
🎯 What it does: This paper proposes R2ec, a unified large-scale recommendation model that integrates inference and recommendation functions within the same architecture, and designs the RecPO RL training framework to achieve end-to-end optimization.
R1-ShareVL: Incentivizing Reasoning Capabilities of Multimodal Large Language Models via Share-GRPO
Huanjin Yao (Tsinghua University), Jiaxing Huang (Nanyang Technological University)
OptimizationTransformerLarge Language ModelReinforcement LearningVision Language ModelTextMultimodality
🎯 What it does: The Share-GRPO method, based on reinforcement learning, enhances the performance of multimodal large language models in long-chain reasoning tasks.
R2R: Efficiently Navigating Divergent Reasoning Paths with Small-Large Model Token Routing
Tianyu Fu (Tsinghua University), Yu Wang (Tsinghua University)
GenerationComputational EfficiencyTransformerLarge Language ModelMixture of ExpertsText
🎯 What it does: A token-level router R2R is designed, allowing small language models (SLM) to operate during most generation steps, only invoking the large language model (LLM) on critical 'path divergence' tokens, significantly reducing inference costs while maintaining high-quality reasoning.
RAD: Towards Trustworthy Retrieval-Augmented Multi-modal Clinical Diagnosis
Haolin Li (Fudan University), Yanfeng Wang (Shanghai Jiao Tong University)
RetrievalExplainability and InterpretabilityTransformerLarge Language ModelContrastive LearningMultimodalityBiomedical DataRetrieval-Augmented Generation
🎯 What it does: A retrieval-augmented diagnostic framework called RAD is proposed, which improves multimodal diagnostic models using external medical knowledge.
RAD: Training an End-to-End Driving Policy via Large-Scale 3DGS-based Reinforcement Learning
Hao Gao (Huazhong University of Science and Technology), Xinggang Wang (Huazhong University of Science and Technology)
Autonomous DrivingTransformerReinforcement LearningGaussian SplattingPoint Cloud
🎯 What it does: Proposes the RAD framework, utilizing 3D Gaussian Splatting (3DGS) digital twin environments for closed-loop reinforcement learning to train end-to-end autonomous driving strategies.
RadarQA: Multi-modal Quality Analysis of Weather Radar Forecasts
Xuming He (ZheJiang University), LEI BAI
TransformerLarge Language ModelSupervised Fine-TuningReinforcement LearningTextMultimodality
🎯 What it does: RadarQA is proposed, a weather radar forecast quality analysis framework based on a multimodal large language model, which includes four tasks (scoring and evaluation of frames/sequences).
Radial Attention: $\mathcal O(n \log n)$ Sparse Attention for Long Video Generation
Xingyang Li (Massachusetts Institute of Technology), Song Han (Massachusetts Institute of Technology)
GenerationData SynthesisComputational EfficiencyTransformerSupervised Fine-TuningVideo
🎯 What it does: Proposes Radial Attention, a sparse attention mechanism with a complexity of O(n log n), designed for efficient generation of long videos;
RadZero: Similarity-Based Cross-Attention for Explainable Vision-Language Alignment in Chest X-ray with Zero-Shot Multi-Task Capability
Jonggwon Park (DEEPNOID Inc), Kyoyun Choi (DEEPNOID Inc)
ClassificationObject DetectionSegmentationExplainability and InterpretabilityTransformerLarge Language ModelContrastive LearningImageTextMultimodalityBiomedical DataComputed Tomography
🎯 What it does: A visual-language alignment framework called RadZero has been developed for zero-shot classification, localization, and segmentation on chest X-ray images, along with a similarity-based cross-attention mechanism (VL-CABS) and multi-positive contrastive learning.
RAG4GFM: Bridging Knowledge Gaps in Graph Foundation Models through Graph Retrieval Augmented Generation
Xingliang Wang (Zhejiang University), Shuiguang Deng (Zhejiang University)
GenerationRetrievalGraph Neural NetworkGraphRetrieval-Augmented Generation
🎯 What it does: The RAG4GFM framework is proposed, which enhances retrieval-augmented generation for graph foundational models through multi-level graph indexing, task-aware retrieval, and graph fusion, improving knowledge update efficiency and inference credibility.
RAGRouter: Learning to Route Queries to Multiple Retrieval-Augmented Language Models
Jiarui Zhang (Shanghai Jiao Tong University), Guihai Chen (Shanghai Jiao Tong University)
RetrievalTransformerLarge Language ModelContrastive LearningTextRetrieval-Augmented Generation
🎯 What it does: This paper proposes a multi-model routing method called RAGRouter for retrieval-augmented generation (RAG) scenarios, which can dynamically select the most suitable large language model (LLM) to generate answers based on queries and retrieved documents.
Rainbow Delay Compensation: A Multi-Agent Reinforcement Learning Framework for Mitigating Observation Delays
Songchen Fu (Institute of Acoustics, Chinese Academy of Sciences), Yonghong Yan (Institute of Acoustics, Chinese Academy of Sciences)
Knowledge DistillationRecurrent Neural NetworkTransformerReinforcement LearningSequential
🎯 What it does: To address the common issue of observation delay in multi-agent reinforcement learning, the Rainbow Delay Compensation (RDC) framework is proposed, which achieves adaptive compensation for delays through modules such as a compensator that reconstructs delay-free observations, delay-aligned critics, curriculum learning strategies, and knowledge distillation.
Random Forest Autoencoders for Guided Representation Learning
Adrien Aumon (Université de Montréal), Jake Slater Rhodes
Representation LearningAuto EncoderImageTabularBiomedical Data
🎯 What it does: This paper proposes Random Forest Autoencoders (RF‑AE), a supervised visualization method that combines neighborhood information generated by random forests with the structure of autoencoders, capable of efficiently generating low-dimensional embeddings for unlabeled new samples.
Random Search Neural Networks for Efficient and Expressive Graph Learning
Michael Ito (University of Michigan), Jenna Wiens (University of Michigan)
Drug DiscoveryRecurrent Neural NetworkGraph Neural NetworkGraphBiomedical Data
🎯 What it does: This paper proposes a Random Search Neural Network (RSNN), which replaces random walks with depth-first search (DFS) to address the insufficient expressiveness of RWNN when sampling is limited. It is proven that RSNN can achieve full graph coverage with logarithmic search times on sparse graphs, making it a universal approximator.
Randomized-MLP Regularization Improves Domain Adaptation and Interpretability in DINOv2
Joel Valdivia Ortega (Helmholtz Munich), Tingying Peng (Helmholtz Munich)
SegmentationDomain AdaptationExplainability and InterpretabilityTransformerContrastive LearningImageBiomedical DataComputed Tomography
🎯 What it does: This paper proposes Randomized-MLP (RMLP) regularization, which replaces the trainable MLP head in ViT (such as DINOv2) to better align features with semantics during the refinement process and enhance the interpretability of attention and feature maps.
RANK++LETR: Learn to Rank and Optimize Candidates for Line Segment Detection
Xin Tong (Intelligent Science and Technology Academy of CASIC), Zhe Ma (Intelligent Science and Technology Academy of CASIC)
Object DetectionOptimizationTransformerImage
🎯 What it does: An end-to-end line segment detection framework RANK++LETR is designed, using a Transformer encoder to generate line segment candidates, and a decoder to reorder candidates by confidence and refine their positions.
Ranking-based Preference Optimization for Diffusion Models from Implicit User Feedback
Yi-Lun Wu (National Yang Ming Chiao Tung University), Hong-Han Shuai (National Yang Ming Chiao Tung University)
GenerationRecommendation SystemOptimizationReinforcement LearningDiffusion modelImage
🎯 What it does: This paper proposes a preference optimization framework for text-to-image diffusion models based on inverse reinforcement learning, called Diffusion-DRO.
RankMatch: A Novel Approach to Semi-Supervised Label Distribution Learning Leveraging Rank Correlation between Labels
Zhiqiang Kou (Southeast University), Xin Geng (Southeast University)
ClassificationContrastive LearningImage
🎯 What it does: A semi-supervised label distribution learning framework called RankMatch is proposed, which utilizes ranking-related loss to enhance the model's understanding of the relative importance relationships between labels.
RankSEG-RMA: An Efficient Segmentation Algorithm via Reciprocal Moment Approximation
Zixun Wang (Chinese University of Hong Kong), Ben Dai (Chinese University of Hong Kong)
SegmentationComputational EfficiencyImage
🎯 What it does: This paper proposes RankSEG-RMA, an algorithm that achieves efficient and compatible non-overlapping multi-class semantic segmentation through recursive matrix approximation.
Rao-Blackwell Gradient Estimators for Equivariant Denoising Diffusion
Vinh Tong (University of Stuttgart), Mathias Niepert
GenerationData SynthesisProtein Structure PredictionDiffusion modelTabularSequential
🎯 What it does: This paper proposes Orbit Diffusion, which utilizes a Rao-Blackwellized gradient estimator to train diffusion models under symmetry constraints, significantly reducing gradient variance and accelerating convergence.
Rao-Blackwellised Reparameterisation Gradients
Kevin H. Lam (University of Oxford), Yee Whye Teh (University of Oxford)
OptimizationAuto EncoderImage
🎯 What it does: The R2-G2 gradient estimator is proposed, achieving low-variance gradients from a single sample through Rao-Blackwellization of the reparameterized gradient.
RAPID Hand: Robust, Affordable, Perception-Integrated, Dexterous Manipulation Platfrom for Embodied Intelligence
Zhaoliang Wan, Hui Cheng (Sun Yat-sen University)
Robotic IntelligenceDiffusion modelMultimodality
🎯 What it does: A low-cost, 20-DoF RAPID Hand platform has been designed and implemented, integrating multimodal perception at the hardware level (wrist RGBD camera, fingertip pressure patches, joint angles) and a high-degree-of-freedom remote control interface for collecting high-quality multi-finger robotic manipulation demonstration data.
RAPTR: Radar-based 3D Pose Estimation using Transformer
Sorachi Kato (Mitsubishi Electric Research Laboratories), Petros Boufounos (Mitsubishi Electric Research Laboratories)
Pose EstimationTransformerPoint Cloud
🎯 What it does: A radar-based 3D human pose estimation framework called RAPTR is proposed, which can directly estimate complete 3D poses from multi-view radar heatmaps under weak supervision conditions using only 3D bounding boxes (BBox) and 2D keypoint labels.
Rare Text Semantics Were Always There in Your Diffusion Transformer
Seil Kang (Yonsei University), Seong Jae Hwang (Yonsei University)
GenerationData SynthesisTransformerDiffusion modelImageVideoTextBenchmark
🎯 What it does: This paper proposes TORA (Token Spacing and Residual Alignment) intervention, which enhances the performance of rare text semantics in visual generation within the MM-DiT generative model through lightweight text embedding variance diffusion and residual alignment, without the need for additional training, optimization, or external modules.
RAST: Reasoning Activation in LLMs via Small-model Transfer
Siru Ouyang (University of Illinois Urbana-Champaign), Jiawei Han (University of Illinois Urbana-Champaign)
OptimizationReinforcement Learning from Human FeedbackTransformerLarge Language ModelReinforcement LearningTextBenchmark
🎯 What it does: Developed the RAST method, which injects the logit differences obtained from RL training of a small model into a large model during inference, thereby activating the inference capability of the large model and avoiding the high-cost RL training on the large model.
RAT: Bridging RNN Efficiency and Attention Accuracy via Chunk-based Sequence Modeling
Xiuying Wei (École Polytechnique Fédérale de Lausanne), Caglar Gulcehre
OptimizationComputational EfficiencyRecurrent Neural NetworkSupervised Fine-TuningTextSequential
🎯 What it does: This paper proposes a block cyclic attention architecture named RAT, which divides long sequences into several blocks. Within each block, a lightweight RNN is used to handle short-range dependencies, while global information interaction between blocks is achieved through softmax attention, thus balancing efficiency and accuracy.
Rationalized All-Atom Protein Design with Unified Multi-Modal Bayesian Flow
Hanlin Wu (Tsinghua University), Jingjing Liu (Tsinghua University)
Drug DiscoveryProtein Structure PredictionFlow-based ModelMultimodalityBiomedical Data
🎯 What it does: A unified all-atom protein design model called ProBayes is proposed, capable of end-to-end generation of sequences, backbone, and side chains.
Ravan: Multi-Head Low-Rank Adaptation for Federated Fine-Tuning
Arian Raje (Carnegie Mellon University), Gauri Joshi (Carnegie Mellon University)
Federated LearningTransformerLarge Language ModelSupervised Fine-TuningImageText
🎯 What it does: In a federated learning environment, a parameter-efficient fine-tuning method for large language models is proposed, called the multi-head low-rank adaptive method RAVAN;
Raw2Drive: Reinforcement Learning with Aligned World Models for End-to-End Autonomous Driving (in CARLA v2)
Zhenjie Yang (Shanghai Jiao Tong University), Junchi Yan (Shanghai Jiao Tong University)
Autonomous DrivingReinforcement LearningWorld ModelVideoBenchmark
🎯 What it does: This paper presents Raw2Drive, a dual-stream reinforcement learning framework based on a world model that achieves end-to-end autonomous driving using raw sensor inputs.
RayFusion: Ray Fusion Enhanced Collaborative Visual Perception
Shaohong Wang (Zhejiang University), Eryun Liu (Zhejiang University)
Object DetectionAutonomous DrivingImage
🎯 What it does: Proposes RayFusion, a framework for collaborative visual perception using multi-vehicle perspective ray occupancy information;
RaySt3R: Predicting Novel Depth Maps for Zero-Shot Object Completion
Bardienus Pieter Duisterhof, Jeffrey Ichnowski
GenerationDepth EstimationTransformerContrastive LearningImagePoint Cloud
🎯 What it does: We propose RaySt3R, a 3D shape completion method that treats a single RGB-D image and a foreground mask as a new perspective synthesis task, capable of generating complete multi-object 3D geometry from zero-shot samples.
RCCDA: Adaptive Model Updates in the Presence of Concept Drift under a Constrained Resource Budget
Adam Piaseczny (Purdue University), Christopher Brinton
Domain AdaptationOptimizationReinforcement LearningTabular
🎯 What it does: Adaptive update strategy for deep learning models under constrained resource budgets in the context of concept drift.
RDD: Retrieval-Based Demonstration Decomposer for Planner Alignment in Long-Horizon Tasks
Mingxuan Yan (University of California), Jiachen Li (University of California)
Robotic IntelligenceLarge Language ModelReinforcement LearningVideoBenchmark
🎯 What it does: This paper proposes a retrieval-based demonstration decomposer (RDD) that automatically segments demonstration videos into subtasks that match low-level visual motion strategy training data.
Re-coding for Uncertainties: Edge-awareness Semantic Concordance for Resilient Event-RGB Segmentation
Nan Bao (Beihang University), Jia Li (Beihang University)
SegmentationOptimizationAuto EncoderImageMultimodality
🎯 What it does: This paper proposes an Edge-awareness Semantic Concordance (ESC) framework that utilizes edge information to unify event and RGB features, achieving semantic segmentation under extreme conditions.
Re-ttention: Ultra Sparse Visual Generation via Attention Statistical Reshape
Ruichen Chen (University of Alberta), Di Niu (University of Alberta)
GenerationTransformerDiffusion modelImageVideo
🎯 What it does: This paper proposes a training-independent sparse attention method called Re‑ttention for text-to-video (T2V) and text-to-image (T2I) generation tasks in the Diffusion Transformer (DiT).
Reaction Prediction via Interaction Modeling of Symmetric Difference Shingle Sets
Runhan Shi (Shanghai Jiao Tong University), Yang Yang (Shanghai Jiao Tong University)
Drug DiscoveryTransformerGraph
🎯 What it does: This paper proposes a chemical reaction prediction model called ReaDISH, based on molecular shingle symmetric difference and interactive attention, aimed at addressing the issues of traditional models being sensitive to input order and insufficient in substructure interaction modeling.
Reading Recognition in the Wild
Charig Yang (VGG University of Oxford), Hyo Jin Kim (Meta Reality Labs Research)
RecognitionTransformerSimultaneous Localization and MappingMultimodality
🎯 What it does: This paper proposes the task of reading recognition using active smart glasses in real-world scenarios, constructing a 100-hour multimodal 'Reading in the Wild' dataset, and designing a lightweight Transformer model that combines eye movement, RGB, and IMU for real-time low-power reading detection.
ReAgent-V: A Reward-Driven Multi-Agent Framework for Video Understanding
Yiyang Zhou (University of North Carolina Chapel Hill), Huaxiu Yao (University of North Carolina Chapel Hill)
Large Language ModelReinforcement LearningVision Language ModelVideoMultimodality
🎯 What it does: Proposes the ReAgent-V framework, which enhances video understanding performance through multi-agent real-time rewards and multi-perspective reflection.
Real-DRL: Teach and Learn in Reality
Yanbing Mao (Wayne State University), Lui Sha (University of Illinois)
Robotic IntelligenceReinforcement Learning
🎯 What it does: Proposes the Real-DRL framework, which implements safety-critical deep reinforcement learning (DRL) on real physical systems, achieving real-time learning and safety assurance through three interactive components: DRL-Student, PHY-Teacher, and Trigger.
Real-Time Execution of Action Chunking Flow Policies
Kevin Black (Physical Intelligence), Sergey Levine (UC Berkeley)
OptimizationRobotic IntelligenceDiffusion modelFlow-based ModelVideo
🎯 What it does: A real-time block execution method RTC based on flow matching has been designed, capable of achieving asynchronous execution and continuous action streams without retraining.
Real-Time Scene-Adaptive Tone Mapping for High-Dynamic Range Object Detection
Gongzhe Li, Qilin Sun (Point Spread Technology)
Object DetectionAutonomous DrivingConvolutional Neural NetworkSupervised Fine-TuningImage
🎯 What it does: A real-time scene-adaptive HDR RAW to LDR color mapping method is proposed to enhance target detection performance in autonomous driving and achieve end-to-end joint training with the detection network.
Real-World Adverse Weather Image Restoration via Dual-Level Reinforcement Learning with High-Quality Cold Start
Fuyang Liu (Nanjing University of Science and Technology), Xiaowei Hu (South China University of Technology)
RestorationReinforcement LearningImage
🎯 What it does: A high-precision large-scale weather dataset HFLS-Weather has been constructed, and a dual-layer reinforcement learning framework has been proposed to achieve real-time recovery of adverse weather images through local model refinement and global meta-controller dynamic scheduling.
Real-World Reinforcement Learning of Active Perception Behaviors
Edward S. Hu (University of Pennsylvania), Dinesh Jayaraman (University of Pennsylvania)
Robotic IntelligenceReinforcement Learning
🎯 What it does: This paper proposes an algorithm that utilizes privileged perception (additional sensors) during training to efficiently learn robotic active perception behaviors—AAWR (Asymmetric Advantage Weighted Regression)—and implements various active/interactive perception tasks on real robots.
REArtGS: Reconstructing and Generating Articulated Objects via 3D Gaussian Splatting with Geometric and Motion Constraints
Di Wu (University of Science and Technology of China), Cewu Lu (Shanghai Jiao Tong University)
GenerationData SynthesisGaussian SplattingImageMesh
🎯 What it does: Under the premise of only providing two multi-view RGB images of arbitrary poses, high-quality textured mesh reconstruction of articulated objects and mesh generation in arbitrary unseen states are achieved using 3D Gaussian Splatting (3DGS).
Reason-RFT: Reinforcement Fine-Tuning for Visual Reasoning of Vision Language Models
Huajie Tan (Peking University), Shanghang Zhang (Peking University)
Domain AdaptationComputational EfficiencyTransformerSupervised Fine-TuningReinforcement LearningVision Language ModelMultimodalityChain-of-Thought
🎯 What it does: A two-stage reinforcement refinement framework, Reason-RFT, is proposed to enhance the generalization and efficiency of visual language models in visual reasoning tasks.
ReasonFlux-PRM: Trajectory-Aware PRMs for Long Chain-of-Thought Reasoning in LLMs
Jiaru Zou (University of Illinois Urbana-Champaign), Mengdi Wang (Princeton University)
Reinforcement Learning from Human FeedbackTransformerLarge Language ModelReinforcement LearningContrastive LearningTextChain-of-Thought
🎯 What it does: Designed and trained a trajectory-aware process reward model, ReasonFlux-PRM, for fine-grained evaluation of intermediate thinking trajectories and final answers in long-chain reasoning, which is then used for offline data selection, reinforcement learning, and optimal answer selection during reasoning.
Reasoning as an Adaptive Defense for Safety
Taeyoun Kim (Carnegie Mellon University), Aviral Kumar (Carnegie Mellon University)
Safty and PrivacyReinforcement Learning from Human FeedbackSupervised Fine-TuningReinforcement LearningTextChain-of-Thought
🎯 What it does: A training process named TARS is designed and implemented, which begins with lightweight supervised fine-tuning (SFT) for preheating, followed by reinforcement learning (RL) training on prompts that include chain-of-thought (CoT) reasoning. This allows the model to dynamically allocate reasoning computation based on the safety of the prompts, achieving adaptive safe reasoning.
Reasoning Beyond Points: A Visual Introspective Approach for Few-Shot 3D Segmentation
Changshuo Wang (University College London), Prayag Tiwari (Halmstad University)
SegmentationDomain AdaptationContrastive LearningPoint Cloud
🎯 What it does: A pre-training-free point cloud few-shot semantic segmentation framework VIP-Seg is proposed.
Reasoning by Superposition: A Theoretical Perspective on Chain of Continuous Thought
Hanlin Zhu (University of California Berkeley), Yuandong Tian (Meta AI)
TransformerLarge Language ModelGraphChain-of-Thought
🎯 What it does: This paper studies and demonstrates that using a continuous chain of thought (COC) allows a two-layer Transformer to efficiently solve the directed graph reachability problem, and the experimental results validate the superior performance of this method on real datasets.
REASONING COMPILER: LLM-Guided Optimizations for Efficient Model Serving
Sujun Tang (University of California San Diego), Hadi Esmaeilzadeh (University of California San Diego)
OptimizationComputational EfficiencyLarge Language ModelMixture of ExpertsText
🎯 What it does: A compiler optimization framework called REASONING COMPILER is proposed, which combines large language models with Monte Carlo tree search to generate efficient transformation sequences for neural network code.
Reasoning Is Not a Race: When Stopping Early Beats Going Deeper
Mohan Zhang (Tsinghua University), Yi Wu (Tsinghua University)
Reinforcement LearningTextChain-of-Thought
🎯 What it does: This paper studies the effect of using a Process Reward Model (PRM) to guide Beam Search in Long Chain Thinking (Long CoT), discovering a phenomenon of step quality degradation, and proposes an early stopping strategy based on z-scores called ZGES.
Reasoning is Periodicity? Improving Large Language Models Through Effective Periodicity Modeling
Yihong Dong (Peking University), Hong Mei (Advanced Institute of Big Data)
TransformerLarge Language ModelText
🎯 What it does: The FANformer structure is proposed, embedding a Fourier Analysis Network into the attention mechanism of the Transformer to enhance periodic modeling, thereby improving the learning efficiency and performance of large language models.