ICLR 2024 Papers — Page 17
International Conference on Learning Representations · 2260 papers
PROGRAM: PROtotype GRAph Model based Pseudo-Label Learning for Test-Time Adaptation
Haopeng Sun (Tsinghua University), Wentao Liu (SenseTime Research)
Domain AdaptationGraph Neural NetworkImage
🎯 What it does: Adapt the pre-trained model during testing to improve performance using unlabeled target domain samples.
Progressive Fourier Neural Representation for Sequential Video Compilation
Haeyong Kang (Korea Advanced Institute of Science and Technology), Chang D. Yoo (Korea Advanced Institute of Science and Technology)
Data SynthesisRepresentation LearningNeural Radiance FieldVideoSequential
🎯 What it does: Proposed the Progressive Fourier Neural Representation (PFNR), a sequential video compilation method that utilizes learnable sparse subnetworks to encode videos in Fourier space, avoiding catastrophic forgetting.
Progressive3D: Progressively Local Editing for Text-to-3D Content Creation with Complex Semantic Prompts
Xinhua Cheng (Peking University), Li Yuan (International Digital Economy Academy)
GenerationData SynthesisDiffusion modelScore-based ModelMesh
🎯 What it does: A Progressive3D framework is proposed, which generates high-quality 3D content that meets complex semantic prompts through step-by-step local editing.
Project and Probe: Sample-Efficient Adaptation by Interpolating Orthogonal Features
Annie S Chen, Chelsea Finn (Stanford University)
Domain AdaptationImage
🎯 What it does: A two-step lightweight few-shot adaptation framework PRO 2 is proposed, which first learns a set of orthogonal predictive features from source domain data, and then trains a linear classifier on these features using a small number of target domain samples.
Prometheus: Inducing Fine-Grained Evaluation Capability in Language Models
Seungone Kim (KAIST), Minjoon Seo (KAIST)
TransformerLarge Language ModelSupervised Fine-TuningTextChain-of-Thought
🎯 What it does: An open-source evaluation language model called PROMETHEUS is proposed, which can perform fine-grained evaluation of long texts based on user-defined scoring criteria.
Prompt Gradient Projection for Continual Learning
Jingyang Qiao (East China Normal University), Yuan Xie (East China Normal University)
SegmentationTransformerPrompt EngineeringImage
🎯 What it does: This paper proposes Prompt Gradient Projection (PGP), which combines prompt-tuning with gradient projection to suppress catastrophic forgetting in continual learning without task identifiers.
Prompt Learning with Quaternion Networks
Boya Shi (Shanghai Jiao Tong University), Chao Ma (Shanghai Jiao Tong University)
ClassificationDomain AdaptationTransformerPrompt EngineeringVision Language ModelImageTextMultimodality
🎯 What it does: A prompt learning framework utilizing Quaternion Networks (QNet) is proposed for fusing visual and textual modality features in zero-shot scenarios and generating high-quality prompts across multiple tasks.
Prompt Risk Control: A Rigorous Framework for Responsible Deployment of Large Language Models
Thomas P Zollo, Richard Zemel (Columbia University)
AI Code AssistantTransformerLarge Language ModelPrompt EngineeringText
🎯 What it does: Proposes a Prompt Risk Control framework that utilizes distribution-free uncertainty quantification methods to provide a high-probability risk upper bound for prompt selection in large language models.
PromptAgent: Strategic Planning with Language Models Enables Expert-level Prompt Optimization
Xinyuan Wang (University of California San Diego), Zhiting Hu
OptimizationDrug DiscoveryTransformerLarge Language ModelReinforcement LearningPrompt EngineeringTextBiomedical Data
🎯 What it does: This paper presents PromptAgent, a framework that views prompt engineering as a strategic planning problem, generating expert-level prompts through MCTS and LLM self-reflection, which can automatically improve performance across multiple tasks.
PromptTTS 2: Describing and Generating Voices with Text Prompt
Yichong Leng (University of Science and Technology of China), Jiang Bian (Microsoft)
GenerationTransformerLarge Language ModelPrompt EngineeringDiffusion modelTextAudio
🎯 What it does: Proposes PromptTTS 2, which uses a differential network and LLM to generate text prompts, addressing the one-to-many and data scale challenges of text prompt TTS;
Proper Laplacian Representation Learning
Diego Gomez (University of Alberta), Marlos C. Machado (University of Alberta)
OptimizationRepresentation LearningGraph
🎯 What it does: This paper proposes a new max-min objective function (ALLO) that achieves hyperparameter-free Laplacian representation learning through stop-gradient and augmented Lagrangian techniques, capable of simultaneously recovering the minimum eigenvector and corresponding eigenvalue of the Laplacian operator.
Protein Discovery with Discrete Walk-Jump Sampling
Nathan C. Frey (Genentech), Saeed Saremi (Genentech)
GenerationDrug DiscoveryDiffusion modelScore-based ModelSequentialBiomedical Data
🎯 What it does: A new discrete generative model framework is proposed—Discrete Walk-Jump Sampling (dWJS), which achieves high-quality and fast discrete sequence generation by training an energy model in a noise-smoothed space and returning to the original discrete data using least squares estimation.
Protein Multimer Structure Prediction via Prompt Learning
Ziqi Gao (Hong Kong University of Science and Technology), Jia Li (Hong Kong University of Science and Technology)
Meta LearningProtein Structure PredictionGraph Neural NetworkPrompt EngineeringGraph
🎯 What it does: A multi-assembly structure prediction framework based on prompt learning, PROMPTMSP, is proposed, which utilizes pre-training and prompt fine-tuning to achieve the stepwise assembly of multi-chain proteins.
Protein-ligand binding representation learning from fine-grained interactions
Shikun Feng (Tsinghua University), Yanyan Lan (Tsinghua University)
Representation LearningDrug DiscoveryTransformerBiomedical Data
🎯 What it does: This paper proposes the BindNet framework, which uses a Transformer interaction module to learn protein-ligand binding representations during the self-supervised pre-training phase, completing pre-training under multiple tasks, and then fine-tuning on various downstream tasks.
Protein-Ligand Interaction Prior for Binding-aware 3D Molecule Diffusion Models
Zhilin Huang (Tsinghua University), Wenming Yang
Drug DiscoveryDiffusion modelBiomedical Data
🎯 What it does: A 3D molecular diffusion model named IPDIFF is proposed, incorporating protein-ligand interaction priors in both forward and reverse diffusion processes to achieve molecule generation targeted at specific targets.
Prototypical Information Bottlenecking and Disentangling for Multimodal Cancer Survival Prediction
Yilan Zhang (Hong Kong University of Science and Technology), Hao Chen (Hong Kong University of Science and Technology)
ClassificationData-Centric LearningTransformerMultimodalityBiomedical Data
🎯 What it does: For multimodal cancer survival prediction, we propose the Prototypical Information Bottleneck and Prototypical Information Disentanglement (PIBD) framework, which first eliminates unimodal redundancy through the prototypical information bottleneck, and then utilizes prototype distribution to drive the decoupling of shared and specific information, thereby enhancing prediction performance.
Provable and Practical: Efficient Exploration in Reinforcement Learning via Langevin Monte Carlo
Haque Ishfaq (Mila, McGill University), Kamyar Azizzadenesheli (Nvidia)
Reinforcement LearningStochastic Differential Equation
🎯 What it does: This paper proposes a reinforcement learning exploration algorithm LMC-LSVI that directly samples the posterior distribution of the Q function using Langevin Monte Carlo (LMC), and extends it to deep Q networks (Adam LMCDQN), achieving scalable and theoretically provable efficient exploration.
Provable Benefits of Multi-task RL under Non-Markovian Decision Making Processes
Ruiquan Huang (Penn State University), Yingbin Liang (Ohio State University)
Reinforcement Learning
🎯 What it does: This study investigates the theoretical advantages of multi-task reinforcement learning in non-Markov low-rank decision problems, proposing the η-bracketing number as a unified characterization of the complexity of multi-task model spaces. It presents the upstream multi-task algorithm UMT-PSR based on Predictive State Representations (PSR) and a scheme for downstream transfer learning using upstream knowledge, proving that sample efficiency can be improved under low bracketing number conditions.
Provable Compositional Generalization for Object-Centric Learning
Thaddäus Wiedemer (University of Tuebingen), Wieland Brendel (Max Planck Institute for Intelligent Systems)
Data SynthesisRepresentation LearningTransformerAuto EncoderImage
🎯 What it does: A theoretical framework and empirical method are proposed, demonstrating that identifiable and generalizable object-level representations of unknown combinations can be achieved in autoencoders that satisfy specific structural assumptions.
Provable Memory Efficient Self-Play Algorithm for Model-free Reinforcement Learning
Na Li (Zhejiang University), Shefeng Yan (Chinese Academy of Sciences)
Reinforcement Learning
🎯 What it does: A new model-free self-play algorithm ME-Nash-QL is proposed for two-player zero-sum Markov games.
Provable Offline Preference-Based Reinforcement Learning
Wenhao Zhan (Princeton University), Wen Sun (Cornell University)
Reinforcement LearningSequential
🎯 What it does: A general function approximation algorithm for offline preference-based reinforcement learning (PbRL) called FREEHAND is proposed. It reconstructs implicit rewards from trajectory preference data through maximum likelihood estimation and performs distributionally robust planning on a confidence set, achieving competitive performance guarantees for the target policy. It is also extended to scenarios with unknown transitions and action comparisons.
Provable Reward-Agnostic Preference-Based Reinforcement Learning
Wenhao Zhan (Princeton University), Jason D. Lee (Princeton University)
Reinforcement Learning from Human FeedbackReinforcement Learning
🎯 What it does: A reward-agnostic experimental design method is proposed, which first collects exploration trajectories without human feedback, then gathers preference feedback to learn the reward function, and finally solves the optimal policy under the learned reward.
Provable Robust Watermarking for AI-Generated Text
Xuandong Zhao (University of California Santa Barbara), Yu-Xiang Wang (University of California Santa Barbara)
GenerationLarge Language ModelText
🎯 What it does: This paper studies the problem of adding watermarks to text generated by large language models (LLMs) and proposes a high-quality watermarking method called UNIGRAM-WATERMARK, establishing a rigorous theoretical framework to quantify the effectiveness and robustness of the watermark.
Provably Efficient CVaR RL in Low-rank MDPs
Yulai Zhao (Princeton University), Jason D. Lee (Princeton University)
Reinforcement Learning
🎯 What it does: This study investigates risk-sensitive reinforcement learning (RL), aiming to maximize the conditional value-at-risk (CVaR) with a fixed risk tolerance level τ.
Provably Efficient Iterated CVaR Reinforcement Learning with Function Approximation and Human Feedback
Yu Chen (Tsinghua University), Longbo Huang (Tsinghua University)
Reinforcement Learning from Human FeedbackReinforcement Learning
🎯 What it does: This paper studies and proposes an Iterative CVaR (ICVaR) reinforcement learning framework, providing sample-efficient algorithms for linear, general function approximation, and human feedback scenarios.
Provably Efficient UCB-type Algorithms For Learning Predictive State Representations
Ruiquan Huang (Penn State University), Jing Yang (Penn State University)
OptimizationReinforcement Learning
🎯 What it does: This paper proposes two algorithms based on UCB/LCB—PSR-UCB (online) and PSR-LCB (offline)—for learning low-rank predictive state representations (PSR) models, and derives near-optimal policies based on this.
Provably Robust Conformal Prediction with Improved Efficiency
Ge Yan (University of California San Diego), Tsui-Wei Weng (University of California San Diego)
ClassificationComputational EfficiencyImage
🎯 What it does: A provably robust framework for synthetic prediction, RSCP+, is proposed, along with two methods: PTT, which is lossless after training, and RCT, which is robust for synthetic training, addressing the robustness proof flaws and inefficiencies of the original RSCP.
Proving Test Set Contamination in Black-Box Language Models
Yonatan Oren (Stanford University), Tatsunori Hashimoto
TransformerLarge Language ModelText
🎯 What it does: This study investigates a statistical method that utilizes exchangeability in black-box language models to test whether the test set is contaminated by pre-training data through log probability ratio comparison.
Proximal Policy Gradient Arborescence for Quality Diversity Reinforcement Learning
Sumeet Batra (University of Southern California), Gaurav S. Sukhatme (University of Southern California)
Robotic IntelligenceReinforcement Learning
🎯 What it does: A quality diversity reinforcement learning algorithm based on PPO, called PPGA, is proposed for high-dimensional robotic motion tasks.
Pseudo-Generalized Dynamic View Synthesis from a Video
Xiaoming Zhao (University of Illinois Urbana-Champaign), Alex Schwing
GenerationData SynthesisDepth EstimationTransformerNeural Radiance FieldOptical FlowVideo
🎯 What it does: A pseudo-general method is proposed to achieve dynamic scene synthesis from a new viewpoint using a monocular video without scene-specific appearance optimization.
PTaRL: Prototype-based Tabular Representation Learning via Space Calibration
Hangting Ye (Jilin University), Yi Chang (Jilin University)
Representation LearningTransformerContrastive LearningTabular
🎯 What it does: Proposes the PTARL framework, which maps tabular data to the prototype projection space (P-Space) and achieves decoupled representation through prototype learning.
PubDef: Defending Against Transfer Attacks From Public Models
Chawin Sitawarin (University of California Berkeley), David Wagner (University of California Berkeley)
Adversarial AttackImage
🎯 What it does: A threat model for public model transfer attacks (TAPM) is proposed, and a defense method called PUBDEF is designed based on game theory;
Pushing Boundaries: Mixup's Influence on Neural Collapse
Quinn LeBlanc Fisher (University of Toronto), Vardan Papyan (University of Toronto)
ClassificationOptimizationConvolutional Neural NetworkTransformerImage
🎯 What it does: A geometric configuration analysis of the final layer features of deep networks trained with mixup reveals that activations of the same class form a simplex ETF, while activations of different classes form channels along the decision boundary, and an unconstrained feature model is used to explain its impact on calibration.
Pushing Mixture of Experts to the Limit: Extremely Parameter Efficient MoE for Instruction Tuning
Ted Zadouri (Cohere For AI), Sara Hooker (Cohere For AI)
TransformerLarge Language ModelSupervised Fine-TuningMixture of ExpertsText
🎯 What it does: An extremely parameter-efficient mixture of experts architecture (MoV, MoLORA) is proposed in the context of instruction fine-tuning, achieving performance comparable to full model fine-tuning with less than 1% of the parameters updated.
Q-Bench: A Benchmark for General-Purpose Foundation Models on Low-level Vision
Haoning Wu (Nanyang Technological University), Weisi Lin (Nanyang Technological University)
TransformerLarge Language ModelVision Language ModelImageTextMultimodalityBenchmark
🎯 What it does: A benchmark called Q-Bench has been constructed and released to systematically evaluate the performance of multimodal large language models (MLLMs) in three key capabilities: low-level visual perception, description, and quality assessment.
QA-LoRA: Quantization-Aware Low-Rank Adaptation of Large Language Models
Yuhui Xu (Huawei Inc.), Qi Tian (Huawei Inc.)
TransformerLarge Language ModelSupervised Fine-TuningText
🎯 What it does: A Quantized Perception Low-Rank Adaptation (QA-LoRA) method is proposed, balancing the parameter-efficient fine-tuning of large language models and low-bit quantization deployment.
QLLM: Accurate and Efficient Low-Bitwidth Quantization for Large Language Models
Jing Liu (Monash University), Bohan Zhuang (SenseTime Research)
CompressionComputational EfficiencyTransformerLarge Language ModelText
🎯 What it does: A low-bit LLM compression method based on post-training quantization, called QLLM, is proposed. It suppresses abnormal channels by splitting and merging activation values, and corrects quantization errors using low-rank parameter fine-tuning. It can complete 4-bit quantization of a 70B model in 10 hours on a single A100 80G GPU.
Quadratic models for understanding catapult dynamics of neural networks
Libin Zhu (University of California San Diego), Mikhail Belkin (University of California San Diego)
OptimizationImage
🎯 What it does: This paper studies Neural Quadratic Models (NQM) under high learning rates, analyzing their optimization dynamics and generalization properties, and comparing them with wide networks and linearized models.
Quality-Diversity through AI Feedback
Herbie Bradley (CarperAI), Joel Lehman
GenerationOptimizationTransformerLarge Language ModelPrompt EngineeringText
🎯 What it does: An AI feedback-based quality-diversity search framework QDAIF is proposed, which combines evolutionary algorithms with large language models to generate and evaluate text, aiming to produce creative writing content that is both high-quality and diverse.
Quantifying and Enhancing Multi-modal Robustness with Modality Preference
Zequn Yang (Renmin University of China), Di Hu (Renmin University of China)
Adversarial AttackVideoMultimodalityAudio
🎯 What it does: This paper studies the robustness of multi-modal models and proposes a new training method called Certifiable Robust Multi-modal Training (CRMT), aimed at improving the robustness of multi-modal models when facing single-modal attacks and missing conditions.
Quantifying Language Models' Sensitivity to Spurious Features in Prompt Design or: How I learned to start worrying about prompt formatting
Melanie Sclar (University of Washington), Alane Suhr (University of California)
TransformerLarge Language ModelPrompt EngineeringText
🎯 What it does: This study investigates the performance fluctuations of large language models under different equivalent prompt formats and proposes a rapid evaluation method called FORMATSPREAD.
Quantifying the Plausibility of Context Reliance in Neural Machine Translation
Gabriele Sarti (University of Groningen), Arianna Bisazza (University of Groningen)
GenerationExplainability and InterpretabilityTransformerText
🎯 What it does: Proposes the PECORE framework for identifying context-sensitive words in generated text from neural machine translation and attributing the contextual cues that lead to their generation, thereby quantifying the contextual dependency interpretability of the model.
Quantifying the Sensitivity of Inverse Reinforcement Learning to Misspecification
Joar Max Viktor Skalse (Oxford University), Alessandro Abate (Oxford University)
Reinforcement Learning
🎯 What it does: This study investigates the sensitivity of Inverse Reinforcement Learning (IRL) to mis-specification of behavioral models, providing necessary and sufficient conditions to quantify the impact of different forms of mis-specification on the inference of reward functions.
Quasi-Monte Carlo for 3D Sliced Wasserstein
Khai Nguyen (University of Texas at Austin), Nhat Ho (University of Texas at Austin)
ClassificationData SynthesisOptimizationAuto EncoderImagePoint Cloud
🎯 What it does: This paper proposes quasi-Monte Carlo (QMC) slice Wasserstein (QSW) and randomized quasi-Monte Carlo (RQSW) estimation methods based on spherical low-discrepancy point sets to improve the traditional Monte Carlo (MC) slice Wasserstein (SW) distance computation and gradient estimation.
Query-Dependent Prompt Evaluation and Optimization with Offline Inverse RL
Hao Sun (University of Cambridge), Mihaela van der Schaar (University of Cambridge)
OptimizationReinforcement Learning from Human FeedbackTransformerReinforcement LearningPrompt EngineeringText
🎯 What it does: A Prompt-OIRL method based on offline inverse reinforcement learning is proposed, which utilizes offline prompt demonstration data to train a reward model, achieving query-dependent prompt evaluation and optimization during inference, significantly enhancing arithmetic reasoning performance.
Query-Policy Misalignment in Preference-Based Reinforcement Learning
Xiao Hu (Tsinghua University), Ya-Qin Zhang (Tsinghua University)
Reinforcement LearningTime SeriesBenchmark
🎯 What it does: This paper studies the query-policy misalignment issue in preference-based reinforcement learning (PbRL) and proposes an improved framework based on policy alignment for query and mixed experience replay (QPA) to enhance feedback efficiency and sampling efficiency.
Querying Easily Flip-flopped Samples for Deep Active Learning
Seong Jin Cho (Korea Advanced Institute of Science and Technology), Chang D. Yoo (Korea Advanced Institute of Science and Technology)
OptimizationConvolutional Neural NetworkGaussian SplattingImage
🎯 What it does: The Least Disagree Metric (LDM) is proposed to measure the uncertainty of deep models on unlabeled samples, and based on this, an active learning algorithm LDM-S is designed.
Quick-Tune: Quickly Learning Which Pretrained Model to Finetune and How
Sebastian Pineda Arango (University of Freiburg), Josif Grabocka (University of Freiburg)
ClassificationHyperparameter SearchMeta LearningImage
🎯 What it does: A method named Quick-Tune has been designed and implemented, which can quickly select the best pre-trained model and its fine-tuning hyperparameters simultaneously from a given model hub.
R-EDL: Relaxing Nonessential Settings of Evidential Deep Learning
Mengyuan Chen (Chinese Academy of Sciences), Changsheng Xu (Chinese Academy of Sciences)
ClassificationAnomaly DetectionImageVideo
🎯 What it does: Improved Evidential Deep Learning and proposed the R-EDL method.
R-MAE: Regions Meet Masked Autoencoders
Duy Kien Nguyen, Xinlei Chen (Meta AI)
Object DetectionSegmentationRepresentation LearningTransformerAuto EncoderImage
🎯 What it does: This paper proposes a region-based masked autoencoding task (RAE) and combines it with MAE to form R-MAE for unsupervised image representation learning and detection/segmentation tasks.
R&B: Region and Boundary Aware Zero-shot Grounded Text-to-image Generation
Jiayu Xiao (Key Lab of Intelligent Information Processing, ICT, CAS), Qingming Huang (University of Chinese Academy of Sciences)
GenerationData SynthesisDiffusion modelImageText
🎯 What it does: A zero-shot grounded text-to-image generation method based on attention guidance is proposed, which simultaneously satisfies text prompts and bounding box constraints without training additional modules.
RA-DIT: Retrieval-Augmented Dual Instruction Tuning
Xi Victoria Lin (Meta), Wen-tau Yih
RetrievalTransformerLarge Language ModelSupervised Fine-TuningTextRetrieval-Augmented Generation
🎯 What it does: This paper proposes a lightweight retrieval-augmented dual instruction tuning (RA-DIT) method, which first fine-tunes a language model to better utilize retrieval information, and then performs LM-supervised retrieval fine-tuning on the retriever to transform any pre-trained LLM into a retrieval-augmented model.
Raidar: geneRative AI Detection viA Rewriting
Chengzhi Mao (Columbia University), Junfeng Yang (Columbia University)
ClassificationTransformerLarge Language ModelPrompt EngineeringText
🎯 What it does: Utilizing large language models (LLMs) to perform rewriting tasks on the same text, calculating the edit distance (Levenshtein distance) and bag-of-words similarity metrics between the original and rewritten texts as features for a binary classifier to determine whether the text is AI-generated.
RAIN: Your Language Models Can Align Themselves without Finetuning
Yuhui Li (Peking University), Hongyang Zhang (University of Waterloo)
GenerationTransformerLarge Language ModelReinforcement LearningText
🎯 What it does: By incorporating a self-assessment and backtracking mechanism during the inference phase, the RAIN method enables frozen large language models to achieve self-alignment without fine-tuning or additional data.
Random Sparse Lifts: Construction, Analysis and Convergence of finite sparse networks
David A. R. Robin (INRIA - ENS Paris PSL Research University), Marc Lelarge (INRIA - ENS Paris PSL Research University)
🎯 What it does: Proposes a random sparse boosting framework and proves that gradient flow can achieve low loss in large-scale sparse networks.
RAPPER: Reinforced Rationale-Prompted Paradigm for Natural Language Explanation in Visual Question Answering
Kai-Po Chang (National Taiwan University), Yu-Chiang Frank Wang (NVIDIA)
Explainability and InterpretabilityKnowledge DistillationReinforcement Learning from Human FeedbackTransformerLarge Language ModelReinforcement LearningVision Language ModelImageTextMultimodality
🎯 What it does: Proposes the RAPPER framework, which utilizes two-stage reinforcement learning and large model distillation to generate interpretable and trustworthy natural language explanations;
RAPTOR: Recursive Abstractive Processing for Tree-Organized Retrieval
Parth Sarthi (Stanford University), Christopher D Manning
RetrievalTransformerLarge Language ModelTextRetrieval-Augmented Generation
🎯 What it does: This paper proposes and implements a recursive abstract tree structure retrieval system called RAPTOR, which can first chunk, embed, and cluster long texts, and then recursively generate a summary tree. During retrieval, information can be obtained at different levels of abstraction, thereby better supporting multi-step reasoning and understanding of long texts.
Rayleigh Quotient Graph Neural Networks for Graph-level Anomaly Detection
Xiangyu Dong (Chinese University of Hong Kong), Sibo Wang (Chinese University of Hong Kong)
Anomaly DetectionGraph Neural NetworkGraph
🎯 What it does: This paper proposes a Rayleigh Quotient-based Graph Neural Network (RQGNN) for graph-level anomaly detection.
RDesign: Hierarchical Data-efficient Representation Learning for Tertiary Structure-based RNA Design
Cheng Tan (Zhejiang University), Stan Z. Li (AI Lab, Research Center for Industries of the Future, Westlake University)
Representation LearningData-Centric LearningGraph Neural NetworkContrastive LearningGraph
🎯 What it does: A three-level structure-based RNA design method called RDesign is proposed, which includes a complete design pipeline for hierarchical data efficient representation learning and secondary structure constraints.
Real-Fake: Effective Training Data Synthesis Through Distribution Matching
Jianhao Yuan (University of Oxford), Bo Zhao (Beijing Academy of Artificial Intelligence)
ClassificationData SynthesisDiffusion modelAuto EncoderImage
🎯 What it does: A theoretical framework based on distribution matching is proposed to guide the synthesis of training data, and under this framework, Stable Diffusion is fine-tuned to generate synthetic training sets suitable for image classification.
Real-time Photorealistic Dynamic Scene Representation and Rendering with 4D Gaussian Splatting
Zeyu Yang (Fudan University), Li Zhang (Fudan University)
GenerationComputational EfficiencyNeural Radiance FieldGaussian SplattingVideo
🎯 What it does: A dynamic scene real-time lighting rendering framework based on 4D Gaussian primitives is proposed, capable of generating high-quality new views at any point in time.
Real3D-Portrait: One-shot Realistic 3D Talking Portrait Synthesis
Zhenhui Ye (Zhejiang University), Zhou Zhao (Zhejiang University)
GenerationData SynthesisTransformerAuto EncoderGenerative Adversarial NetworkImageVideoAudio
🎯 What it does: Developed Real3D-Portrait, a one-shot 3D talking portrait generation framework that can reconstruct a 3D avatar from a single reference image and achieve natural lip-sync and expression animation driven by video or audio, ultimately synthesizing high-quality talking portrait videos that include a complete torso and switchable backgrounds.
Realistic Evaluation of Semi-supervised Learning Algorithms in Open Environments
Lin-Han Jia (Nanjing University), Yu-Feng Li (Nanjing University)
ClassificationData-Centric LearningImageTextTabular
🎯 What it does: This paper studies the robustness of semi-supervised learning in open environments (with inconsistencies in distribution, features, and labels), proposes a multidimensional robustness metric based on the Robustness Analysis Curve, and constructs a theoretical framework to re-implement and uniformly evaluate various statistical, deep, and robust SSL algorithms.
Reasoning on Graphs: Faithful and Interpretable Large Language Model Reasoning
LINHAO LUO, Shirui Pan (Griffith University)
Explainability and InterpretabilityKnowledge DistillationTransformerLarge Language ModelSupervised Fine-TuningTextGraphRetrieval-Augmented Generation
🎯 What it does: This paper proposes a reasoning framework called RoG that combines large language models with knowledge graphs, implementing trustworthy and interpretable KG question answering through a three-stage process of planning, retrieval, and reasoning.
Reasoning with Latent Diffusion in Offline Reinforcement Learning
Siddarth Venkatraman (Mila Universit e de Montr al e), Glen Berseth (Mila Universit e de Montr al e)
Reinforcement LearningDiffusion modelSequential
🎯 What it does: In offline reinforcement learning, a Q-learning method that utilizes a latent diffusion model for temporal abstraction of trajectories combined with batch constraints (LDCQ) is proposed.
REBAR: Retrieval-Based Reconstruction for Time-series Contrastive Learning
Maxwell Xu, James Matthew Rehg
ClassificationAnomaly DetectionRepresentation LearningConvolutional Neural NetworkContrastive LearningTime SeriesElectrocardiogram
🎯 What it does: A retrieval-reconstruction-based time series contrastive learning method called REBAR is proposed, which constructs positive and negative samples by learning the reconstruction error between sequences.
Reclaiming the Source of Programmatic Policies: Programmatic versus Latent Spaces
Tales Henrique Carvalho, Levi Lelis
OptimizationRobotic IntelligenceReinforcement Learning from Human FeedbackReinforcement LearningAuto EncoderSequential
🎯 What it does: This study investigates the effects of using the raw DSL space and the learned latent space for local search in programmatic strategy synthesis.
RECOMBINER: Robust and Enhanced Compression with Bayesian Implicit Neural Representations
Jiajun He (University of Cambridge), José Miguel Hernández-Lobato (University of Cambridge)
Data SynthesisCompressionImageVideoBiomedical DataAudio
🎯 What it does: This paper presents RECOMBINER, a compression method based on Bayesian Implicit Neural Representation (INR), which enhances compression quality through linear reparameterization, learnable positional information encoding, and hierarchical priors.
RECOMP: Improving Retrieval-Augmented LMs with Context Compression and Selective Augmentation
Fangyuan Xu (University of Texas at Austin), Eunsol Choi (University of Washington)
RetrievalCompressionComputational EfficiencyTransformerLarge Language ModelContrastive LearningTextRetrieval-Augmented Generation
🎯 What it does: The RECOMP method is proposed, which first compresses the retrieved long texts into brief summaries in the retrieval-augmented language model, and then uses the summary as a prefix input to the language model, thereby improving inference efficiency while maintaining or enhancing task performance.
Reconciling Spatial and Temporal Abstractions for Goal Representation
Mehdi Zadem (École Polytechnique), Sao Mai Nguyen (IMT Atlantique)
Robotic IntelligenceReinforcement Learning
🎯 What it does: This paper proposes a three-layer hierarchical reinforcement learning algorithm called STAR, which utilizes spatial and temporal abstraction to jointly construct goal representations, thereby improving the learning efficiency of continuous control tasks.
Recursive Generalization Transformer for Image Super-Resolution
Zheng Chen (Shanghai Jiao Tong University), Xiaokang Yang (Shanghai Jiao Tong University)
RestorationSuper ResolutionTransformerImage
🎯 What it does: This paper proposes a Recursive Generalized Transformer (RGT) for image super-resolution, aiming to efficiently capture global spatial information at high resolutions and improve reconstruction quality.
REFACTOR: Learning to Extract Theorems from Proofs
Jin Peng Zhou (Cornell University), Roger Baker Grosse
Graph Neural NetworkGraph
🎯 What it does: This paper proposes a neural network model named REFACTOR, which is used to automatically extract reusable theorems (lemmas) from formal mathematical proof trees, and implements extraction, reconstruction, and proof tasks in the Metamath library.
ReFusion: Improving Natural Language Understanding with Computation-Efficient Retrieval Representation Fusion
Shangyu Wu (City University of Hong Kong), Chun Jason Xue (Mohamed bin Zayed University of Artificial Intelligence)
RetrievalOptimizationComputational EfficiencyTransformerTextRetrieval-Augmented Generation
🎯 What it does: ReFusion is proposed, a retrieval-enhanced framework for non-knowledge-intensive tasks that directly integrates retrieval representations into the model's hidden layers, using a bi-level optimization approach to find the optimal fusion structure.
Reinforcement Symbolic Regression Machine
Yilong Xu (Renmin University of China), Hao Sun (Renmin University of China)
OptimizationReinforcement LearningTabular
🎯 What it does: This paper proposes a symbolic regression framework RSRM that combines Monte Carlo Tree Search, Double Q-Learning, and Modular Subtree Discovery to efficiently search for and discover complex mathematical equations with limited data.
Relaxing the Additivity Constraints in Decentralized No-Regret High-Dimensional Bayesian Optimization
Anthony Bardou (École Polytechnique Fédérale de Lausanne), Thomas Begin (École Normale Supérieure de Lyon)
Optimization
🎯 What it does: A distributed Bayesian optimization algorithm named DuMBO is proposed, which can infer and utilize arbitrarily complex additive decompositions to optimize high-dimensional noisy and expensive black-box functions without restricting the maximum factor size (MFS) of the additive structure.
Relay Diffusion: Unifying diffusion process across resolutions for image synthesis
Jiayan Teng (Tsinghua University), Jie Tang (Tsinghua University)
GenerationData SynthesisDiffusion modelImage
🎯 What it does: The Relay Diffusion Model (RDM) is proposed, which seamlessly extends low-resolution generation results to high resolution through block noise and block-level fuzzy diffusion, addressing the issue of noise scheduling and low-resolution conditions mismatch in traditional cascaded methods.
ReLoRA: High-Rank Training Through Low-Rank Updates
Vladislav Lialin (University of Massachusetts Lowell), Anna Rumshisky (Amazon)
OptimizationTransformerLarge Language ModelText
🎯 What it does: Introduces a parameter-efficient pre-training method called ReLoRA, which utilizes multiple low-rank updates to train high-rank networks;
ReLU Strikes Back: Exploiting Activation Sparsity in Large Language Models
Seyed Iman Mirzadeh (Apple), Mehrdad Farajtabar (Apple)
GenerationComputational EfficiencyTransformerLarge Language ModelSupervised Fine-TuningText
🎯 What it does: The study investigates the activation sparsity brought by the ReLU activation function in large language models and proposes a two-stage Relufication method to transform existing non-ReLU models into ReLU while significantly reducing inference computation costs while maintaining nearly the same performance. It further explores the combination of aggregated sparsity and speculative decoding to enhance generation speed.
ReMasker: Imputing Tabular Data with Masked Autoencoding
Tianyu Du (Zhejiang University), Ting Wang (Penn State)
TransformerAuto EncoderTabular
🎯 What it does: A table missing value imputation method based on masked autoencoders—REMASKER is proposed.
Remote Sensing Vision-Language Foundation Models without Annotations via Ground Remote Alignment
Utkarsh Mall (Cornell University), Kavita Bala (Cornell University)
ClassificationSegmentationRetrievalTransformerVision Language ModelContrastive LearningImageMultimodality
🎯 What it does: Utilizing ground internet images as a bridge, we train a remote sensing vision-language model (GRAFT) that does not require text annotations, achieving zero-shot classification, retrieval, segmentation, and question answering;
Removing Biases from Molecular Representations via Information Maximization
Chenyu Wang (Massachusetts Institute of Technology), Tommi S. Jaakkola
Representation LearningDrug DiscoveryContrastive LearningMultimodalityTabular
🎯 What it does: This paper proposes the InfoCORE method to eliminate batch effects and improve representation quality in multimodal molecular representation learning.
Repeated Random Sampling for Minimizing the Time-to-Accuracy of Learning
Patrik Okanovic (ETH Zürich), Theodoros Rekatsinas (ETH Zürich)
OptimizationKnowledge DistillationConvolutional Neural NetworkImage
🎯 What it does: This paper proposes a method called 'Repeated Random Sampling' (RS2), which randomly samples subsets multiple times during each training epoch, and demonstrates that it outperforms existing complex data pruning and data distillation techniques in terms of time-to-accuracy.
Repelling Random Walks
Isaac Reid (University of Cambridge), Adrian Weller (University of Cambridge)
Graph Neural NetworkGraph
🎯 What it does: A repelling random walks mechanism is proposed, which utilizes the mutual repulsion relationship between multiple walk trajectories to enhance the efficiency of random walks on graphs by reducing the estimator variance while keeping the edge transition probabilities unchanged.
Rephrase, Augment, Reason: Visual Grounding of Questions for Vision-Language Models
Archiki Prasad (University of North Carolina), Mohit Bansal (University of North Carolina)
RecognitionGenerationTransformerLarge Language ModelVision Language ModelImageTextMultimodality
🎯 What it does: A gradient-free, instance-level framework called REPARE is proposed, which rewrites and supplements visual question answering (VQA) problems to enhance their visual relevance and clarity, thereby improving zero-shot VQA performance.
Replay across Experiments: A Natural Extension of Off-Policy RL
Dhruva Tirumala (University College London), Markus Wulfmeier (Google DeepMind)
Robotic IntelligenceReinforcement LearningSequential
🎯 What it does: A simple framework called 'Replay across Experiments (RaE)' is proposed, which significantly improves the performance and efficiency of offline RL algorithms by sharing experience data across multiple experiments and using a fixed ratio of offline and online replay, with minimal changes to the existing RL workflow.
RepoBench: Benchmarking Repository-Level Code Auto-Completion Systems
Tianyang Liu (University of California San Diego), Julian McAuley (University of California San Diego)
RetrievalAI Code AssistantTransformerLarge Language ModelTextBenchmarkRetrieval-Augmented Generation
🎯 What it does: This paper proposes RepoBench, a benchmarking system for code autocompletion in multi-file projects, covering three main tasks: retrieval (RepoBench-R), completion (RepoBench-C), and the complete pipeline (RepoBench-P), supporting both Python and Java languages.
Representation Deficiency in Masked Language Modeling
Yu Meng (University of Illinois Urbana Champaign), Luke Zettlemoyer (Meta AI)
Representation LearningTransformerLarge Language ModelSupervised Fine-TuningAuto EncoderText
🎯 What it does: This study investigates how the MASK symbol in Masked Language Modeling pre-training leads to model dimensions being specifically occupied by MASK, resulting in defects in the representation of real words, and proposes MAE-LM to address this issue by excluding MASK in the encoder.
ResFields: Residual Neural Fields for Spatiotemporal Signals
Marko Mihajlovic (ETH Zurich), Siyu Tang (Microsoft)
Neural Radiance FieldVideo
🎯 What it does: The study introduces a temporal residual layer (ResFields) into the neural field model to enhance its ability to model long and complex sequences.
ReSimAD: Zero-Shot 3D Domain Transfer for Autonomous Driving with Source Reconstruction and Target Simulation
Bo Zhang (Shanghai Artificial Intelligence Laboratory), Yu Qiao (Shanghai Artificial Intelligence Laboratory)
Object DetectionDomain AdaptationAutonomous DrivingNeural Radiance FieldPoint CloudMesh
🎯 What it does: This paper proposes a method to generate 3D meshes from source domain point clouds through implicit reconstruction, and then render these meshes into point clouds according to the target domain LiDAR parameters. This allows for the generation of point clouds in the style of the target domain without using any target domain labels, and utilizes these zero-shot data to enhance cross-domain 3D detection performance.
Respect the model: Fine-grained and Robust Explanation with Sharing Ratio Decomposition
Sangyu Han (Seoul National University), Nojun Kwak (Seoul National University)
Explainability and InterpretabilityConvolutional Neural NetworkImage
🎯 What it does: This paper proposes an explanation method based on Shared Ratio Decomposition (SRD) that provides high-resolution and robust pixel-level explanations without altering the model.
ReTaSA: A Nonparametric Functional Estimation Approach for Addressing Continuous Target Shift
Hwanwoo Kim (University of Chicago), Qinglong Tian (University of Waterloo)
Domain AdaptationTabularTime Series
🎯 What it does: This study addresses the problem of continuous target shift and proposes the ReTaSA method, which solves ill-posed integral equations to non-parametrically estimate important weights, thereby achieving a label-free domain adaptation regression model.
Rethinking Adversarial Policies: A Generalized Attack Formulation and Provable Defense in RL
Xiangyu Liu (University of Maryland), Furong Huang (University of Maryland)
Adversarial AttackRobotic IntelligenceReinforcement LearningSequential
🎯 What it does: This paper proposes a general adversarial attack framework under a multi-agent reinforcement learning environment and presents a provably convergent adversarial training algorithm based on time scale separation.
Rethinking and Extending the Probabilistic Inference Capacity of GNNs
Tuo Xu (Peking University), Lei Zou (Peking University)
Graph Neural NetworkGraph
🎯 What it does: This paper reinterprets and extends the expressive power of Graph Neural Networks (GNNs) in graphical models from the perspective of probabilistic reasoning. It proposes a new hierarchical framework based on Markov Random Fields (MRF), evaluates the performance of various GNN variants in posterior inference on nodes and edges, and introduces two preprocessing methods, 'phantom nodes' and 'phantom edges', to enhance the high-order dependency modeling and link prediction capabilities of MPNN.
Rethinking Backdoor Attacks on Dataset Distillation: A Kernel Method Perspective
Ming-Yu Chung (National Taiwan University), Tsung-Yi Ho (Chinese University of Hong Kong)
Knowledge DistillationAdversarial AttackImage
🎯 What it does: This study investigates the feasibility of backdoor attacks during the Knowledge Induction Process (KIP) of dataset distillation, constructs a theoretical framework based on kernel methods, and designs two types of triggers (simple-trigger and relax-trigger) to implant backdoors into the distilled dataset.
Rethinking Branching on Exact Combinatorial Optimization Solver: The First Deep Symbolic Discovery Framework
Yufei Kuang (University of Science and Technology of China), Feng Wu (University of Science and Technology of China)
OptimizationExplainability and InterpretabilityComputational EfficiencyRecurrent Neural NetworkReinforcement LearningTabular
🎯 What it does: A deep symbolic discovery framework called Symb4CO is proposed for the branch selection task in Mixed Integer Linear Programming (MILP), aimed at automatically learning efficient and interpretable branching strategies, and compiling the learned symbolic expressions into rules that can be directly deployed on pure CPU.
Rethinking Channel Dependence for Multivariate Time Series Forecasting: Learning from Leading Indicators
Lifan Zhao (Shanghai Jiao Tong University), Yanyan Shen (Shanghai Jiao Tong University)
Time Series
🎯 What it does: The study focuses on multivariate time series forecasting and proposes a module called LIFT that utilizes locally stable lead-lag relationships.
Rethinking Channel Dimensions to Isolate Outliers for Low-bit Weight Quantization of Large Language Models
Jung Hwan Heo (University of Southern California), Dongsoo Lee (NAVER Cloud)
Anomaly DetectionOptimizationTransformerLarge Language ModelText
🎯 What it does: This paper proposes a novel per-IC quantization method and constructs the AdaDim adaptation framework to achieve low-bit weight quantization (3-bit/4-bit) while maintaining LLM performance.
Rethinking CNN’s Generalization to Backdoor Attack from Frequency Domain
Quanrui Rao (Ludong University), Wuying Liu (Ludong University)
Adversarial AttackConvolutional Neural NetworkImage
🎯 What it does: This study investigates the memory mechanism of CNN backdoor attacks in the frequency domain, explores the impact of triggers on different frequency components, and proposes methods to conceal visible triggers and attack based on low-frequency semantic information.
Rethinking Complex Queries on Knowledge Graphs with Neural Link Predictors
Hang Yin (Tsinghua University), Yangqiu Song (Hong Kong University of Science and Technology)
Graph Neural NetworkGraph
🎯 What it does: This paper evaluates and redefines the reasoning methods for complex query answers (EFO 1) in knowledge graphs, pointing out the limitations of existing query embedding techniques in terms of syntax and expressiveness. It extends to complete EFO 1 queries and proposes a neural-symbolic reasoning algorithm called FIT, which combines fuzzy logic to achieve end-to-end reasoning for any EFO 1 query.
Rethinking Information-theoretic Generalization: Loss Entropy Induced PAC Bounds
Yuxin Dong (Xi'an Jiaotong University), Chen Li (Huazhong Agriculture University)
ClassificationOptimizationConvolutional Neural NetworkImage
🎯 What it does: A high-probability information-theoretic generalization bound based on loss entropy is proposed, replacing the traditional mutual information bound, and providing two types of PAC confidence bounds: data-independent and data-dependent.