NeurIPS 2023 Papers — Page 10
Conference on Neural Information Processing Systems · 3218 papers
Dynamo-Depth: Fixing Unsupervised Depth Estimation for Dynamical Scenes
Yihong Sun (Cornell University), Bharath Hariharan (Cornell University)
Depth EstimationAutonomous DrivingOptical FlowVideo
🎯 What it does: We propose Dynamo-Depth, an unsupervised method for jointly learning monocular depth, camera motion, 3D independent motion, and motion segmentation in unlabeled videos.
DynGFN: Towards Bayesian Inference of Gene Regulatory Networks with GFlowNets
Lazar Atanackovic (University of Toronto), Jason Hartford (Mila - Quebec AI Institute)
Generative Adversarial NetworkBiomedical DataOrdinary Differential Equation
🎯 What it does: A Bayesian dynamic structure learning framework named DynGFN is proposed for inferring gene regulatory networks with cyclic feedback.
DynPoint: Dynamic Neural Point For View Synthesis
Kaichen Zhou (University of Oxford), Niki Trigoni (University of Oxford)
GenerationData SynthesisDepth EstimationNeural Radiance FieldOptical FlowVideoPoint Cloud
🎯 What it does: This paper presents DynPoint, a fast perspective synthesis algorithm for monocular video;
E2PNet: Event to Point Cloud Registration with Spatio-Temporal Representation Learning
Xiuhong Lin (Xiamen University), Cheng Wang (Xiamen University)
Object DetectionPose EstimationAutonomous DrivingRepresentation LearningConvolutional Neural NetworkSimultaneous Localization and MappingMultimodalityPoint Cloud
🎯 What it does: The first learning-based cross-modal registration method E2PNet for event cameras and 3D point clouds is proposed, and the EP2T network is designed to encode event data into a two-dimensional grid feature tensor.
Easy Learning from Label Proportions
Robert Istvan Busa-Fekete, Andres Munoz medina
ClassificationDomain AdaptationConvolutional Neural NetworkSupervised Fine-TuningTabular
🎯 What it does: This paper studies the problem of Learning with Label Proportions (LLP) and proposes a general unbiased estimation method called EASYLLP, along with a theoretical analysis of the traditional proportion matching algorithm PROPMATCH.
Echoes Beyond Points: Unleashing the Power of Raw Radar Data in Multi-modality Fusion
Yang Liu (Chinese Academy of Sciences Institute of Automation), Zhaoxiang Zhang (Chinese Academy of Sciences Institute of Automation and University of Chinese Academy of Sciences)
Object DetectionAutonomous DrivingTransformerMultimodalityPoint Cloud
🎯 What it does: The EchoFusion method is proposed, which directly fuses raw radar time-domain/frequency-domain data with images to achieve efficient cross-modal fusion in the BEV space, thereby enhancing object detection performance.
Ecosystem-level Analysis of Deployed Machine Learning Reveals Homogeneous Outcomes
Connor Toups (Stanford University), Percy Liang (Stanford University)
ImageMultimodality
🎯 What it does: This paper proposes and applies an ecosystem-level analysis method to assess the comprehensive impact of multi-model ensembles deployed in different scenarios on individuals, revealing systemic failures and homogenized outcomes.
EDGI: Equivariant Diffusion for Planning with Embodied Agents
Johann Brehmer (Qualcomm AI Research), Taco Cohen (Qualcomm AI Research)
Robotic IntelligenceReinforcement LearningDiffusion modelSequential
🎯 What it does: Designed and implemented a diffusion model equivalent to SE(3)×Z×Sₙ for model-based reinforcement learning and planning, achieving higher performance in navigation and object manipulation tasks.
Effective Bayesian Heteroscedastic Regression with Deep Neural Networks
Alexander Immer (ETH Zurich), Julia E Vogt
Auto EncoderGaussian SplattingTabular
🎯 What it does: A deep heteroscedastic regression model based on natural parameterization is proposed, and a scalable Laplace approximation is introduced to achieve automatic Bayesian regularization and model uncertainty estimation.
Effective Human-AI Teams via Learned Natural Language Rules and Onboarding
Hussein Mozannar (Massachusetts Institute of Technology), David Sontag (Massachusetts Institute of Technology)
Object DetectionRecommendation SystemAutonomous DrivingTransformerLarge Language ModelContrastive LearningImageText
🎯 What it does: A framework called IntegrAI is proposed for automatically learning and mastering natural language rules, aimed at guiding humans in making three types of decisions—ignore, rely, or collaborate—when working with AI.
Effective Robustness against Natural Distribution Shifts for Models with Different Training Data
Zhouxing Shi (University of California, Los Angeles), Yao Qin (University of California, Santa Barbara)
ClassificationDomain AdaptationImage
🎯 What it does: This paper proposes a new method for evaluating 'multi-ID effective robustness' to more accurately measure the robustness of models under natural distribution shifts (OOD) when the training data distribution differs.
Effective Targeted Attacks for Adversarial Self-Supervised Learning
Minseon Kim (Korea Advanced Institute of Science and Technology), Sung Ju Hwang (Korea Advanced Institute of Science and Technology)
Representation LearningAdversarial AttackConvolutional Neural NetworkContrastive LearningImage
🎯 What it does: A targeted adversarial attack for self-supervised learning is proposed, which selects the most confusing samples by combining entropy and similarity, allowing the model to achieve better robustness under unlabeled conditions.
Effectively Learning Initiation Sets in Hierarchical Reinforcement Learning
Akhil Bagaria (Brown University), George Konidaris (Brown University)
Reinforcement Learning
🎯 What it does: By constructing the Initial Value Function (IVF) and combining it with a weighted classifier, a method is proposed that can simultaneously learn option policies and their initiation sets, using experience rewards or count-based rewards to alleviate the pessimistic shrinkage phenomenon of the initiation sets.
Efficient Activation Function Optimization through Surrogate Modeling
Garrett Bingham (University of Texas at Austin), Risto Miikkulainen (University of Texas at Austin)
OptimizationConvolutional Neural NetworkTransformerImageBenchmark
🎯 What it does: By constructing three major benchmark datasets for activation functions and utilizing the Fisher information matrix features along with activation function outputs to create a low-dimensional proxy space, efficient optimization of activation functions was achieved through simple regression search, discovering various activation functions that surpass ReLU (including sigmoidal forms).
Efficient Adaptation of Large Vision Transformer via Adapter Re-Composing
Wei Dong (Xi'an University of Architecture and Technology), Peng Wang (University of Electronic Science and Technology of China)
ClassificationComputational EfficiencyTransformerImage
🎯 What it does: The Adapter Re-Composing (ARC) method is proposed, which achieves efficient fine-tuning of Vision Transformers by sharing low-rank adapter parameters.
Efficient Adversarial Attacks on Online Multi-agent Reinforcement Learning
Guanlin Liu (University of California), Lifeng Lai (University of California)
Adversarial AttackReinforcement LearningTabular
🎯 What it does: This paper studies adversarial attack models in online multi-agent reinforcement learning and provides attack strategies and theoretical analysis in white-box, gray-box, and black-box scenarios.
Efficient Adversarial Contrastive Learning via Robustness-Aware Coreset Selection
Xilie Xu (National University of Singapore), Mohan Kankanhalli (National University of Singapore)
Computational EfficiencyRepresentation LearningAdversarial AttackContrastive LearningImage
🎯 What it does: A robust sensitivity-aware core subset selection (RCS) framework is proposed, which utilizes an unlabeled representative difference (RD) metric to select a subset that enables adversarial contrastive learning (ACL) to produce robust representations, thereby significantly accelerating ACL training.
Efficient Algorithms for Generalized Linear Bandits with Heavy-tailed Rewards
Bo Xue (City University of Hong Kong), Lijun Zhang (Nanjing University)
Tabular
🎯 What it does: This paper addresses the generalized linear bandit (GLB) problem with heavy-tailed rewards and proposes two new algorithms—CRTM (truncated strategy) and CRMM (mean-median strategy)—and proves that both can achieve an approximately optimal ˜O(dT^{1/(1+ϵ)}) low scheduling;
Efficient Batched Algorithm for Contextual Linear Bandits with Large Action Space via Soft Elimination
Osama Hanna, Christina Fragouli (University of California Los Angeles)
OptimizationReinforcement Learning from Human Feedback
🎯 What it does: The first efficient batch contextual linear bandit algorithm for large action spaces is proposed, utilizing a 'soft elimination' method to explore only a limited number of benchmark actions in each batch and estimate the parameter vector.
Efficient Bayesian Learning Curve Extrapolation using Prior-Data Fitted Networks
Steven Adriaensen (University of Freiburg), Frank Hutter (University of Freiburg)
TransformerTime SeriesBenchmark
🎯 What it does: This study investigates the use of Prior-Data Fitting Networks (PFN) for Bayesian learning curve extrapolation, proposing and implementing the LC-PFN model, which is further applied to online early stopping decisions.
Efficient Beam Tree Recursion
Jishnu Ray Chowdhury (University of Illinois Chicago), Cornelia Caragea (University of Illinois Chicago)
Recurrent Neural NetworkText
🎯 What it does: An efficient Beam Tree Recursive Neural Network (EBT-RvNN) is proposed and extended into a structured model (EBT-GAU) that can contextualize tokens.
Efficient Data Subset Selection to Generalize Training Across Models: Transductive and Inductive Networks
Eeshaan Jain (Indian Institute of Technology Bombay), Abir De (Google)
OptimizationData-Centric LearningNeural Architecture SearchGraph Neural NetworkTransformerImage
🎯 What it does: This paper presents SUBSELNET, a trainable data subset selection framework that can quickly generate optimal training subsets without explicitly training the model when given a new model architecture. It implements two variants (transductive and inductive) and their hybrid version.
Efficient Diffusion Policies For Offline Reinforcement Learning
Bingyi Kang (Sea AI Lab), Shuicheng YAN
Reinforcement LearningDiffusion modelTabularBenchmark
🎯 What it does: An efficient diffusion strategy (EDP) is proposed for offline reinforcement learning, significantly accelerating training.
Efficient Equivariant Transfer Learning from Pretrained Models
Sourya Basu (University of Illinois at Urbana-Champaign), Lav R. Varshney (University of Illinois at Urbana-Champaign)
ClassificationOptimizationTransformerLarge Language ModelReinforcement LearningImageText
🎯 What it does: Proposes λ-equitune and its special case equizero, utilizing pre-trained models to achieve weighted group equivariant transfer learning across various tasks.
Efficient Exploration in Continuous-time Model-based Reinforcement Learning
Lenart Treven (ETH Zürich), Andreas Krause (ETH Zürich)
OptimizationRobotic IntelligenceReinforcement LearningTime SeriesBiomedical DataOrdinary Differential Equation
🎯 What it does: A model-based reinforcement learning algorithm for continuous time systems, OCORL, is proposed, which utilizes ODE to describe dynamics, Gaussian Process (GP) to estimate the model, and employs an optimistic principle for policy optimization. Various measurement selection strategies (MSS) are designed and proven to achieve asymptotic no-regret under the GP model.
Efficient Hyper-parameter Optimization with Cubic Regularization
Zhenqian Shen (Tsinghua University), quanming yao
OptimizationHyperparameter SearchTabular
🎯 What it does: A hyperparameter optimization algorithm that does not rely on hypergradients is proposed, utilizing stochastic relaxation and cubic regularization to update the hyperparameter distribution parameters.
Efficient Learning of Linear Graph Neural Networks via Node Subsampling
Seiyun Shin (University of Illinois), Han Zhao (University of Illinois)
Graph Neural NetworkGraph
🎯 What it does: A training algorithm for linear graph convolutional networks based on node subsampling and leverage score sampling is designed, requiring only O(nd log n) entries of A for training, avoiding the O(nd²) computational cost.
Efficient Low-rank Backpropagation for Vision Transformer Adaptation
Yuedong Yang (University of Texas at Austin), Radu Marculescu (University of Texas at Austin)
ClassificationSegmentationDomain AdaptationComputational EfficiencyTransformerSupervised Fine-TuningImage
🎯 What it does: A low-rank backpropagation method called LBP-WHT is proposed, which utilizes the Walsh-Hadamard transform to perform gradient matrix multiplication in a low-rank space, significantly reducing the computational load of Vision Transformers during the fine-tuning process.
Efficient Meta Neural Heuristic for Multi-Objective Combinatorial Optimization
Jinbiao Chen (Sun Yat-sen University), Siyuan Chen (Singapore Management University)
OptimizationMeta LearningReinforcement Learning
🎯 What it does: An efficient Evolutionary Meta-Neuro Heuristic (EMNH) framework is proposed for Multi-Objective Combinatorial Optimization Problems (MOCOP), which generates Pareto approximate solutions by first training a meta-model and then quickly fine-tuning it for different weight vectors.
Efficient Model-Free Exploration in Low-Rank MDPs
Zakaria Mhammedi (Massachusetts Institute of Technology), Alexander Rakhlin (Massachusetts Institute of Technology)
Reinforcement Learning
🎯 What it does: A new algorithm called SpanRL is proposed for efficient model-free exploration in Low-Rank Markov Decision Processes (MDPs).
Efficient Neural Music Generation
Max W. Y. Lam (ByteDance), Yuxuan Wang (ByteDance)
GenerationData SynthesisComputational EfficiencyTransformerLarge Language ModelDiffusion modelGenerative Adversarial NetworkTextAudio
🎯 What it does: This paper proposes an efficient text-to-music generation model called MeLoDy, which combines language models with diffusion models to achieve audio quality comparable to MusicLM while significantly accelerating the sampling process.
Efficient Online Clustering with Moving Costs
Dimitris Christou (University of Texas at Austin), Volkan Cevher (École Polytechnique Fédérale de Lausanne)
OptimizationImage
🎯 What it does: A polynomial-time online k-clustering algorithm is proposed, which can only decide the locations of k facilities after observing the last request in each round, while simultaneously considering connection costs and facility migration costs.
Efficient Policy Adaptation with Contrastive Prompt Ensemble for Embodied Agents
Wonje Choi (Sungkyunkwan University), Honguk Woo (Sungkyunkwan University)
Domain AdaptationAutonomous DrivingRobotic IntelligenceReinforcement LearningVision Language ModelContrastive LearningMultimodality
🎯 What it does: Proposes the CONPE framework, which utilizes the CLIP visual-language model and contrastive learning to generate visual prompts, and achieves zero-shot domain adaptation in embodied RL through guided attention integration.
Efficient Potential-based Exploration in Reinforcement Learning using Inverse Dynamic Bisimulation Metric
YIMING WANG, Leong Hou U (University of Macau)
Reinforcement LearningSequential
🎯 What it does: A potential function based on the inverse dynamics bisimulation metric is proposed for potential reward shaping without human intervention, promoting efficient exploration in deep reinforcement learning.
Efficient RL with Impaired Observability: Learning to Act with Delayed and Missing State Observations
Minshuo Chen (Princeton University), Mengdi Wang (Princeton University)
Reinforcement Learning
🎯 What it does: This paper studies how to efficiently perform reinforcement learning in Markov Decision Processes (MDPs) with delayed or missing observations, and provides an approximate optimal regret upper bound.
Efficient Robust Bayesian Optimization for Arbitrary Uncertain inputs
Lin Yang (Huawei Noah's Ark Lab), Zhitang Chen (Huawei Noah's Ark Lab)
OptimizationRobotic IntelligenceTabular
🎯 What it does: This paper proposes a robust Bayesian optimization algorithm called AIRBO, which addresses the uncertainty of arbitrary distribution inputs and aims to find optimal solutions that perform stably under uncertain inputs.
Efficient Sampling of Stochastic Differential Equations with Positive Semi-Definite Models
Anant Raj (University of Illinois Urbana-Champaign), Alessandro Rudi (Inria)
OptimizationComputational EfficiencyStochastic Differential Equation
🎯 What it does: This paper proposes a method for approximating the Fokker-Planck equation (including fractional Fokker-Planck) solutions using positive semi-definite (PSD) models, and provides an efficient sampling theoretical framework for SDE based on this approximate solution, giving upper bounds on error, model dimension, and sampling cost.
Efficient Subgame Refinement for Extensive-form Games
Zhenxing Ge (Nanjing University), Yang Gao (Nanjing University)
OptimizationComputational EfficiencyReinforcement Learning
🎯 What it does: The Generative Subgame Solving (GS2) method is proposed, which achieves real-time strategy improvement for large information set games by constructing a subset of subgames using a generative function.
Efficient Symbolic Policy Learning with Differentiable Symbolic Expression
Jiaming Guo (Institute of Computing Technology, Chinese Academy of Sciences), Yunji Chen (Institute of Software, Chinese Academy of Sciences)
Meta LearningReinforcement LearningSequential
🎯 What it does: This paper proposes an efficient gradient-based symbolic policy learning method (ESPL) that can learn interpretable symbolic policies from scratch and generate symbolic policies (CSP) for unknown tasks in meta reinforcement learning.
Efficient Test-Time Adaptation for Super-Resolution with Second-Order Degradation and Reconstruction
Zeshuai Deng (South China University of Technology), Mingkui Tan (South China University of Technology)
RestorationSuper ResolutionImage
🎯 What it does: A super-resolution test-time adaptation framework (SRTTA) is proposed, which can quickly adjust a pre-trained model to generate high-quality HR images for unknown degradation of test images without training data.
Efficient Testable Learning of Halfspaces with Adversarial Label Noise
Ilias Diakonikolas (University of Wisconsin Madison), Nikos Zarifis (University of Wisconsin Madison)
OptimizationAdversarial Attack
🎯 What it does: Developed a polynomial-time verifiable learning algorithm that can learn half-spaces under Gaussian distribution in the presence of adversarial label noise.
Efficient Training of Energy-Based Models Using Jarzynski Equality
Davide Carbone (Politecnico di Torino), Eric Vanden-Eijnden (New York University)
GenerationOptimizationContrastive LearningImage
🎯 What it does: Using the Jarzynski equality combined with Sequential Monte Carlo (SMC) techniques, a weighted Unadjusted Langevin Algorithm (ULA) is proposed to unbiasedly estimate the cross-entropy gradient when training energy-based models (EBMs), thereby achieving accurate learning of the weights of multimodal distributions.
Efficient Uncertainty Quantification and Reduction for Over-Parameterized Neural Networks
Ziyi Huang (Columbia University), Haofeng Zhang (Columbia University)
OptimizationComputational EfficiencyTabular
🎯 What it does: A framework for constructing confidence intervals and eliminating uncertainty for over-parameterized neural networks is proposed, which is centered on training the network only twice (the base network and the artificial label network). The Procedural-Noise-Correcting (PNC) predictor is used to achieve procedural noise removal; combined with lightweight resampling (batching and cheap bootstrap), it results in confidence intervals with controllable statistical coverage.
Efficiently incorporating quintuple interactions into geometric deep learning force fields
Zun Wang (Microsoft Research AI4Science), Bin Shao (Microsoft Research AI4Science)
Graph Neural NetworkGraphPhysics Related
🎯 What it does: This paper studies a graph neural network named QuinNet, which can efficiently and explicitly incorporate five-body interaction force field models.
Egocentric Planning for Scalable Embodied Task Achievement
Xiaotian Liu (ServiceNow Research), Christian Muise (Queen's University)
Robotic IntelligenceConvolutional Neural NetworkReinforcement LearningVision Language ModelMultimodalityBenchmark
🎯 What it does: The Egocentric Planning method is proposed in the ALFRED environment, combining symbolic planning with Object-oriented POMDP. It constructs a semantic graph using existing visual and language models for iterative exploration and replanning, achieving scalable task completion.
EgoDistill: Egocentric Head Motion Distillation for Efficient Video Understanding
Shuhan Tan (University of Texas at Austin), Kristen Grauman (University of Texas at Austin)
RecognitionComputational EfficiencyKnowledge DistillationConvolutional Neural NetworkRecurrent Neural NetworkContrastive LearningVideoMultimodality
🎯 What it does: EgoDistill is proposed, a distillation-based framework that utilizes sparse RGB images and IMU head motion to reconstruct heavy egocentric video features, achieving efficient video understanding.
EgoEnv: Human-centric environment representations from egocentric video
Tushar Nagarajan (University of Texas at Austin), Kristen Grauman (University of Texas at Austin)
RecognitionDomain AdaptationTransformerVideo
🎯 What it does: The EgoEnv method is proposed, which pre-trains a Transformer in a simulation environment to predict local environmental states, thereby generating transferable environment perception features for human-centered tasks such as room recognition and natural language query localization in first-person videos.
EICIL: Joint Excitatory Inhibitory Cycle Iteration Learning for Deep Spiking Neural Networks
Zihang Shao (Dalian University of Technology), Qi Xu (Dalian University of Technology)
Spiking Neural NetworkSupervised Fine-TuningImage
🎯 What it does: A method called EICIL (Excitation-Inhibition Cycle Iterative Learning) is proposed, which integrates STDP and Surrogate Gradient to train deep SNNs.
Elastic Decision Transformer
Yueh-Hua Wu (University of California San Diego), Masashi Hamaya (Omron Sinic)
TransformerReinforcement LearningSequential
🎯 What it does: This paper proposes the Elastic Decision Transformer (EDT) for offline reinforcement learning, addressing the shortcomings of traditional Decision Transformers in trajectory stitching.
ELDEN: Exploration via Local Dependencies
Zizhao Wang (University of Texas at Austin), Roberto Martín-Martín (University of Texas at Austin)
Reinforcement Learning
🎯 What it does: To address the efficient exploration problem in sparse reward reinforcement learning, an intrinsic reward method based on local dependencies, ELDEN, is proposed and validated in various simulated environments with complex chain interactions.
Eliciting User Preferences for Personalized Multi-Objective Decision Making through Comparative Feedback
Han Shao (Toyota Technological Institute at Chicago), Matthew Walter
OptimizationReinforcement Learning from Human FeedbackReinforcement Learning
🎯 What it does: This paper proposes a multi-objective decision-making framework that utilizes user feedback on the comparison of two strategies to learn their implicit linear preference vector, and outputs an approximately optimal personalized strategy under the assumption of a known MDP.
Eliminating Catastrophic Overfitting Via Abnormal Adversarial Examples Regularization
Runqi Lin (Sydney AI Centre, University of Sydney), Tongliang Liu (Sydney AI Centre, University of Sydney)
OptimizationAdversarial AttackConvolutional Neural NetworkImage
🎯 What it does: By introducing the Anomaly Adversarial Example Regularization (AAER) method, catastrophic overfitting (CO) in single-step adversarial training is eliminated, and model robustness is enhanced.
Eliminating Domain Bias for Federated Learning in Representation Space
Jianqing Zhang (Shanghai Jiao Tong University), Haibing Guan (Shanghai Jiao Tong University)
Federated LearningRepresentation LearningImageText
🎯 What it does: This paper proposes the Domain Bias Eliminator (DBE) framework, which eliminates representation bias and degradation issues caused by uneven data domains in federated learning by introducing two modules: Personalized Representation Bias Memory (PRBM) and Mean Regularization (MR) into the local model.
Embedding Space Interpolation Beyond Mini-Batch, Beyond Pairs and Beyond Examples
Shashanka Venkataramanan (Inria), Yannis Avrithis (Institute of Advanced Research on Artificial Intelligence)
ClassificationObject DetectionGenerative Adversarial NetworkImage
🎯 What it does: Proposes two data augmentation methods, MultiMix and Dense MultiMix, for large-scale linear interpolation in the embedding space.
EmbodiedGPT: Vision-Language Pre-Training via Embodied Chain of Thought
Yao Mu (University of Hong Kong), Ping Luo (Shanghai AI Laboratory)
Robotic IntelligenceTransformerLarge Language ModelPrompt EngineeringVision Language ModelVision-Language-Action ModelVideoTextMultimodalityChain-of-Thought
🎯 What it does: This paper proposes EmbodiedGPT, an end-to-end multimodal foundational model capable of generating executable chain-of-thought planning from video input and executing low-level actions through closed-loop control. It also constructs large-scale EgoCOT and EgoVQA datasets.
Embracing the chaos: analysis and diagnosis of numerical instability in variational flows
Zuheng Xu (University of British Columbia), Trevor Campbell (University of British Columbia)
Flow-based ModelTabular
🎯 What it does: This study investigates the errors in sampling, density estimation, and ELBO estimation of variational flows under numerically unstable conditions, and proposes diagnostic methods.
Embroid: Unsupervised Prediction Smoothing Can Improve Few-Shot Classification
Neel Guha (Stanford University), Christopher Re
ClassificationTransformerLarge Language ModelPrompt EngineeringText
🎯 What it does: The EMBROID method is proposed, which utilizes unlabeled data and embedding space to smooth the prediction of prompts in language models, enhancing few-shot classification performance.
Emergence of Shape Bias in Convolutional Neural Networks through Activation Sparsity
Tianqin Li (Carnegie Mellon University), Tai Sing Lee (Carnegie Mellon University)
GenerationData SynthesisConvolutional Neural NetworkTransformerGenerative Adversarial NetworkImage
🎯 What it does: This study investigates how introducing sparse coding (Top-K method) in convolutional neural networks can significantly enhance the model's shape bias and improve robustness against texture interference, while generating images with greater structural coherence in few-shot generation tasks.
Emergent and Predictable Memorization in Large Language Models
Stella Biderman (Booz Allen Hamilton), Edward Raff (Stability AI)
TransformerLarge Language ModelText
🎯 What it does: This paper studies how to predict whether a model will remember specific training data by observing the memory performance of smaller models or intermediate checkpoints before training large language models.
Emergent Communication for Rules Reasoning
Yuxuan Guo (University of Science and Technology of China), Yunji Chen (University of Chinese Academy of Sciences)
Recurrent Neural NetworkReinforcement LearningText
🎯 What it does: This paper proposes a cognitive-oriented 'Reasoning Game' that allows two intelligent agents to solve the Raven Progressive Matrices problem by reasoning and communicating high-level rules. Based on this, an unbiased rule-RAVEN dataset and a two-stage curriculum learning training framework are designed.
Emergent Communication in Interactive Sketch Question Answering
Zixing Lei (Shanghai Jiao Tong University), Siheng Chen (Cornell University)
Object DetectionGenerationExplainability and InterpretabilityConvolutional Neural NetworkVision Language ModelImageMultimodality
🎯 What it does: An interactive sketch question answering (ISQA) task is proposed, allowing two agents to answer image questions through multi-round sketch communication.
Emergent Correspondence from Image Diffusion
Luming Tang (Cornell University), Bharath Hariharan (Cornell University)
RecognitionRetrievalDiffusion modelImage
🎯 What it does: This paper proposes using intermediate features of pre-trained diffusion models (DIFT) for unsupervised visual correspondence matching, achieving high-precision correspondences in semantic, geometric, and temporal dimensions.
EMMA-X: An EM-like Multilingual Pre-training Algorithm for Cross-lingual Representation Learning
Ping Guo (Institute of Information Engineering, Chinese Academy of Sciences), jun xie
RetrievalRepresentation LearningTransformerContrastive LearningText
🎯 What it does: EMMA-X is proposed, a cross-language pre-training method based on the EM framework, which learns universal sentence representations by utilizing a large amount of non-parallel multilingual data through bidirectional supervision of a GMM classifier and a cross-language encoder.
Empowering Collaborative Filtering with Principled Adversarial Contrastive Loss
An Zhang (National University of Singapore), Tat-Seng Chua (National University of Singapore)
Recommendation SystemGraph Neural NetworkContrastive LearningTabular
🎯 What it does: This paper proposes and implements an adversarial InfoNCE loss (AdvInfoNCE) for collaborative filtering (CF), enhancing the robustness and generalization ability of recommendation models through fine-grained hardness learning of difficult negative samples.
Empowering Convolutional Neural Nets with MetaSin Activation
Farnood Salehi (Disney Research), Yuxuan Wang (ETH Zürich)
RestorationKnowledge DistillationConvolutional Neural NetworkImage
🎯 What it does: A trainable activation function METASIN is proposed to replace the traditional ReLU, aiming to enhance the performance of convolutional networks in image prediction tasks such as image resampling and Monte-Carlo denoising.
Encoding Human Behavior in Information Design through Deep Learning
Guanghui Yu (Washington University in St. Louis), Chien-Ju Ho (Washington University in St. Louis)
OptimizationTabular
🎯 What it does: Proposed and implemented the HAIDNet framework, which encodes human behavior in information design using deep learning and optimizes information disclosure strategies;
Encoding Time-Series Explanations through Self-Supervised Model Behavior Consistency
Owen Queen (Harvard University), Marinka Zitnik (Harvard University)
Anomaly DetectionExplainability and InterpretabilityTransformerTime SeriesSequentialBiomedical DataElectrocardiogram
🎯 What it does: This paper proposes TIMEX, an interpretable proxy model for explaining pre-trained time series models, which enhances interpretability by generating interpretable attribution graphs while maintaining consistency with the behavior of the original model.
End-To-End Latent Variational Diffusion Models for Inverse Problems in High Energy Physics
Alexander Shmakov (University of California Irvine), Daniel Whiteson (University of California Irvine)
TransformerDiffusion modelTabularPhysics Related
🎯 What it does: A new end-to-end Variational Latent Diffusion model (VLD) has been researched and implemented to reverse-engineer LHC detector observation data to theoretical levels, thereby addressing the unfolding problem in high-energy physics.
End-to-End Meta-Bayesian Optimisation with Transformer Neural Processes
Alexandre Max Maraval, Haitham Bou Ammar (Huawei Noah's Ark Lab)
OptimizationHyperparameter SearchTransformerReinforcement LearningTabularSequential
🎯 What it does: The first end-to-end differentiable meta Bayesian optimization framework is proposed, which directly predicts the acquisition function using a Transformer neural process and implements sampling decisions through reinforcement learning.
Energy Discrepancies: A Score-Independent Loss for Energy-Based Models
Tobias Schröder (Imperial College London), Andrew Duncan (Imperial College London)
GenerationAnomaly DetectionDiffusion modelContrastive LearningImageTabular
🎯 What it does: An Energy Discrepancy (ED) loss is proposed for training energy models without the need for MCMC or gradient information, allowing training in both Euclidean and discrete spaces.
Energy Guided Diffusion for Generating Neurally Exciting Images
Paweł A. Pierzchlewicz (University of Göttingen), Fabian H. Sinz (Baylor College of Medicine)
GenerationConvolutional Neural NetworkTransformerDiffusion modelImage
🎯 What it does: This paper proposes a readout layer based on a visual attention mechanism, combined with Energy-Guided Diffusion (EGG) technology, to generate more natural and cross-model generalizable Most Exciting Images (MEI) and image reconstructions.
Energy Transformer
Benjamin Hoover (IBM Research), Dmitry Krotov (IBM Research)
ClassificationImage TranslationAnomaly DetectionGraph Neural NetworkTransformerImageGraph
🎯 What it does: The Energy Transformer (ET) architecture is proposed, which combines attention mechanisms, energy models, and associative memory. By designing an energy function, the Transformer block is transformed into a recursive energy descending process for tasks such as image completion, graph anomaly detection, and graph classification.
Energy-Based Cross Attention for Bayesian Context Update in Text-to-Image Diffusion Models
Geon Yeong Park (Korea Advanced Institute of Science and Technology), Jong Chul Ye (Korea Advanced Institute of Science and Technology)
GenerationData SynthesisTransformerDiffusion modelImageText
🎯 What it does: Proposes an energy-based cross-attention framework that enhances the semantic consistency of text-to-image diffusion models through untrained adaptive context updates.
Energy-based learning algorithms for analog computing: a comparative study
Benjamin Scellier (Rain AI), Suhas Kumar (Rain AI)
ClassificationOptimizationComputational EfficiencyConvolutional Neural NetworkImage
🎯 What it does: A systematic comparison of seven energy-based learning algorithms (CL, P-EP, N-EP, C-EP, P-CpL, N-CpL, C-CpL) on deep convolutional Hopfield networks is conducted, proposing asynchronous energy minimization and low-precision solving, achieving faster and more accurate results than existing methods.
Energy-Based Models for Anomaly Detection: A Manifold Diffusion Recovery Approach
Sangwoong Yoon (Korea Institute for Advanced Study), Frank C. Park (Seoul National University)
Anomaly DetectionAuto EncoderImageTabularAudio
🎯 What it does: An anomaly detection method based on an energy model, MPDR, is proposed. By projecting diffusion perturbations on a low-dimensional manifold learned by an autoencoder and training the EBM using recovery likelihood maximization, the anomaly detection performance for various data types is significantly improved.
Energy-Based Sliced Wasserstein Distance
Khai Nguyen (University of Texas at Austin), Nhat Ho (University of Texas at Austin)
Data SynthesisOptimizationImagePoint Cloud
🎯 What it does: An energy-based slice Wasserstein distance (EBSW) is proposed, which adaptively selects the projection direction without optimization by setting the slice distribution to be a non-parametric distribution proportional to the energy of the one-dimensional Wasserstein distance.
Energy-Efficient Scheduling with Predictions
Eric Balkanski (Columbia University), Hao-Ting Wei (Columbia University)
OptimizationComputational EfficiencyTabular
🎯 What it does: This paper proposes a two-phase scheduling framework TPE based on learning enhancement, aimed at achieving energy-efficient scheduling under the premise of machine learning predictions, addressing both deadline and flow time issues.
Enhancing Adaptive History Reserving by Spiking Convolutional Block Attention Module in Recurrent Neural Networks
Qi Xu (Dalian University of Technology), Gang Pan (Zhejiang University)
Recurrent Neural NetworkSpiking Neural NetworkTime SeriesSequential
🎯 What it does: Combining convolutional LSTM and attention mechanisms, the SRNN-SCBAM model is proposed to adapt to spatiotemporal event stream data.
Enhancing Adversarial Contrastive Learning via Adversarial Invariant Regularization
Xilie Xu (National University of Singapore), Mohan Kankanhalli (National University of Singapore)
ClassificationRepresentation LearningAdversarial AttackContrastive LearningImage
🎯 What it does: Proposed and implemented Adversarial Invariant Regularization based on Causal Inference (AIR), which forces the model to be insensitive to style factors in adversarial contrastive learning, thereby obtaining more robust and transferable representations.
Enhancing Adversarial Robustness via Score-Based Optimization
Boya Zhang (Peking University), Zhihua Zhang (Peking University)
OptimizationComputational EfficiencyAdversarial AttackDiffusion modelScore-based ModelImage
🎯 What it does: A novel adversarial defense framework called ScoreOpt is proposed, which is based on a pre-trained score diffusion model. By performing gradient optimization on adversarial samples during testing, it approaches the maximum a posteriori point of the prior distribution, thereby achieving rapid purification of adversarial samples.
Enhancing CLIP with CLIP: Exploring Pseudolabeling for Limited-Label Prompt Tuning
Cristina Menghini (Brown University), Stephen Bach
ClassificationRecognitionTransformerPrompt EngineeringVision Language ModelContrastive LearningImage
🎯 What it does: The study utilizes CLIP to generate pseudo-labels and combines iterative prompt tuning to enhance the image classification performance of VLM in scenarios with few labels, transfer zero-shot, and unsupervised learning.
Enhancing Knowledge Transfer for Task Incremental Learning with Data-free Subnetwork
Qiang Gao (Southwestern University of Finance and Economics), Fan Zhou (University of Electronic Science and Technology of China)
ClassificationKnowledge DistillationImage
🎯 What it does: This paper proposes a data-independent subnetwork (DSN) method for task incremental learning, achieving forward and backward knowledge transfer through neuron-level masking and data-independent replay, thereby avoiding catastrophic forgetting.
Enhancing Minority Classes by Mixing: An Adaptative Optimal Transport Approach for Long-tailed Classification
Jintong Gao (Jilin University), Dan dan Guo
ClassificationContrastive LearningImage
🎯 What it does: An adaptive image mixing method based on optimal transport, OTmix, is proposed to enhance the performance of minority classes in long-tail classification.
Enhancing Motion Deblurring in High-Speed Scenes with Spike Streams
Shiyan Chen (Peking University), Zhaofei Yu (Peking University)
RestorationTransformerImageVideo
🎯 What it does: A dual-modal motion deblurring framework called SpkDeblurNet is proposed, which utilizes a high temporal resolution pulsed camera working in conjunction with a traditional RGB camera to achieve motion deblurring in high-speed scenes.
Enhancing Robot Program Synthesis Through Environmental Context
Tianyi Chen (Fudan University), Xin Peng (Fudan University)
Robotic IntelligenceConvolutional Neural NetworkGraph Neural NetworkSequential
🎯 What it does: This study proposes the EVAPS framework for robot program synthesis under conditions of partial environmental observation, enabling precise generation and generalization of programs through environmental context correction and syntax alignment.
Enhancing Sharpness-Aware Optimization Through Variance Suppression
Bingcong Li (University of Minnesota), Georgios B. Giannakis (University of Minnesota)
OptimizationAdversarial AttackConvolutional Neural NetworkTransformerImageText
🎯 What it does: Proposed the VaSSO method to stabilize the adversarial perturbations of the Sharpness-Aware Optimizer (SAM) by suppressing variance;
Enhancing User Intent Capture in Session-Based Recommendation with Attribute Patterns
Xin Liu (Hong Kong University of Science and Technology), Yangqiu Song (Hong Kong University of Science and Technology)
Recommendation SystemGraph Neural NetworkTransformerSequential
🎯 What it does: The FAPAT framework is designed to enhance anonymous session sequence encoding using frequent attribute graph patterns, thereby improving user intent capture and next item prediction in session recommendations.
Ensemble-based Deep Reinforcement Learning for Vehicle Routing Problems under Distribution Shift
Yuan Jiang (Nanyang Technological University), Jie Zhang (Nanyang Technological University)
OptimizationReinforcement LearningTabular
🎯 What it does: A set-based deep reinforcement learning framework (EL-DRL) is proposed to solve vehicle routing problems (TSP, CVRP) under distribution drift.
Entropic Neural Optimal Transport via Diffusion Processes
Nikita Gushchin (Skolkovo Institute of Science and Technology), Evgeny Burnaev (Skolkovo Institute of Science and Technology)
GenerationData SynthesisOptimizationDiffusion modelImageStochastic Differential Equation
🎯 What it does: An end-to-end neural network algorithm based on the Schrödinger bridge is proposed to solve the entropy-regularized optimal transport (EOT) plan between continuous probability distributions, supporting small entropy coefficients and capable of being trained in one go.
Entropy-based Training Methods for Scalable Neural Implicit Samplers
Weijian Luo (Peking University), Zhihua Zhang (Peking University)
Score-based ModelImageTabular
🎯 What it does: A trainable neural implicit sampler is proposed, capable of sampling directly from an unnormalized target distribution through a single forward pass.
Entropy-dissipation Informed Neural Network for McKean-Vlasov Type PDEs
Zebang Shen (ETH Zurich), Zhenfu Wang (Peking University)
Physics RelatedStochastic Differential Equation
🎯 What it does: A framework based on entropy dissipation neural networks (EINN) is proposed to solve McKean-Vlasov partial differential equations (MVE) with singular interactions.
Environment-Aware Dynamic Graph Learning for Out-of-Distribution Generalization
Haonan Yuan (Beihang University), Jianxin Li (Beihang University)
Domain AdaptationRecommendation SystemGraph Neural NetworkAuto EncoderGraphTime Series
🎯 What it does: Proposes the EAGLE framework to achieve adaptive generalization to distribution drift on dynamic graphs, targeting node-level future link prediction tasks.
Epidemic Learning: Boosting Decentralized Learning with Randomized Communication
Martijn De Vos, Rishi Sharma (École Polytechnique Fédérale de Lausanne)
OptimizationFederated LearningImage
🎯 What it does: A new decentralized learning algorithm called Epidemic Learning (EL) is proposed, which achieves dynamic topology through random sampling and communication with several nodes in each round;
Episodic Multi-Task Learning with Heterogeneous Neural Processes
Jiayi Shen (University of Amsterdam), Marcel Worring (University of Amsterdam)
Meta LearningTransformerTabular
🎯 What it does: This paper proposes a heterogeneous neural process (HNPs) aimed at combining meta-learning and multi-task learning, which simultaneously handles multiple related and heterogeneous tasks in each meta-training/testing round to address the issue of insufficient data.
Epistemic Neural Networks
Ian Osband (Google DeepMind), Benjamin Van Roy (Google DeepMind)
ClassificationExplainability and InterpretabilityComputational EfficiencyImage
🎯 What it does: Epinet is proposed as a structure that can supplement any conventional neural network to improve the joint prediction quality of the model, thereby enhancing decision-making performance.
Equal Opportunity of Coverage in Fair Regression
Fangxin Wang (University of Illinois Chicago), Philip S. Yu (University of Illinois Chicago)
Tabular
🎯 What it does: This paper proposes a new fairness metric—Equal Opportunity Coverage (EOC) in the context of uncertain regression, and presents a post-processing method BFQR, which aims to achieve consistent coverage within bins while maintaining overall coverage and minimizing the width of prediction intervals.
Equivariant Adaptation of Large Pretrained Models
Arnab Kumar Mondal (Mila), Siamak Ravanbakhsh (Mila)
Object DetectionSegmentationConvolutional Neural NetworkImagePoint Cloud
🎯 What it does: By adding a learnable normalization network in front of a pre-trained large model, it achieves equivariance to specific transformations (such as rotation) while maintaining original performance;
Equivariant flow matching
Leon Klein (Freie Universität Berlin), Frank Noe
GenerationOptimizationGraph Neural NetworkFlow-based ModelGraphPhysics RelatedStochastic Differential Equation
🎯 What it does: This paper proposes a covariant optimal transport flow matching objective for symmetric energy density and trains a covariant continuous normalizing flow (CNF) with it, achieving shorter integration paths, higher sampling efficiency, and better generation quality.
Equivariant Flow Matching with Hybrid Probability Transport for 3D Molecule Generation
Yuxuan Song (Institute of AI Industry Research), Wei-Ying Ma (Institute of AI Industry Research)
GenerationDrug DiscoveryGraph Neural NetworkDiffusion modelFlow-based ModelPoint CloudOrdinary Differential Equation
🎯 What it does: A geometric generative model based on flow matching, EquiFM, is proposed to simultaneously generate molecular atom types and three-dimensional coordinates, addressing the issues of probabilistic dynamics instability and slow sampling speed in traditional diffusion models.
Equivariant Neural Operator Learning with Graphon Convolution
Chaoran Cheng (University of Illinois Urbana-Champaign), Jian Peng (University of Illinois Urbana-Champaign)
Graph Neural NetworkGraphPhysics Related
🎯 What it does: A covariant neural operator that combines coefficient learning and coordinate residual layers is proposed, capable of learning mappings between continuous functions in 3D Euclidean space while ensuring SE(3) covariance.