NeurIPS 2024 Papers — Page 12
Conference on Neural Information Processing Systems · 4035 papers
Efficient multi-prompt evaluation of LLMs
Felipe Maia Polo (University of Michigan), Mikhail Yurochkin (IBM Research)
Large Language ModelPrompt EngineeringTextBenchmark
🎯 What it does: Proposes the PromptEval method, which achieves efficient estimation of LLM performance distribution under a large number of prompt templates.
Efficient Multi-task LLM Quantization and Serving for Multiple LoRA Adapters
Yifei Xia (Peking University), Bin CUI
TransformerLarge Language ModelText
🎯 What it does: This paper proposes LoRA-Inlaid, a service system for large language models (LLMs) that supports multi-tasking, quantization, and the dynamic addition of LoRA adapters.
Efficient Multi-task Reinforcement Learning with Cross-Task Policy Guidance
Jinmin He (Institute of Automation, Chinese Academy of Sciences), Jian Cheng (Tencent AI Lab)
Robotic IntelligenceReinforcement LearningSequential
🎯 What it does: This paper proposes the Cross-Task Policy Guidance (CTPG) framework, which generates better training trajectories in multi-task reinforcement learning by guiding the policy to select control strategies from other tasks for each task, and incorporates two gating mechanisms: policy filtering gate and guidance blocking gate to enhance learning efficiency.
Efficient Policy Evaluation Across Multiple Different Experimental Datasets
Yonghan Jung (Purdue University), Alexis Bellot
Tabular
🎯 What it does: This paper studies the inference of policy effects using multi-source experimental data in situations where the distributions of the target domain and source domain are inconsistent, and provides identifiable graphical criteria and multi-robust estimators.
Efficient Prompt Optimization Through the Lens of Best Arm Identification
Chengshuai Shi (University of Virginia), Cong Shen (University of Virginia)
OptimizationTransformerLarge Language ModelPrompt EngineeringText
🎯 What it does: Automatically optimize prompt words for large language models under a limited budget, proposing the TRIPLE framework that can be directly applied to prompt selection.
Efficient Recurrent Off-Policy RL Requires a Context-Encoder-Specific Learning Rate
Fan-Ming Luo (Nanjing University), Yang Yu (Nanjing University)
Recurrent Neural NetworkReinforcement LearningSequential
🎯 What it does: A recursive offline reinforcement learning algorithm RESeL is proposed, which uses a context encoder-specific learning rate to enhance the training stability and performance of RNNs in POMDP tasks.
Efficient Reinforcement Learning by Discovering Neural Pathways
Samin Yeasar Arnob (McGill University), Doina Precup (McGill University)
Reinforcement Learning
🎯 What it does: This paper proposes a method for single-task and multi-task training in reinforcement learning by discovering sparse neural pathways, achieving a network sparsity of up to 95% while maintaining performance.
Efficient Sign-Based Optimization: Accelerating Convergence via Variance Reduction
Wei Jiang (Nanjing University), Lijun Zhang (Nanjing University)
OptimizationConvolutional Neural NetworkImage
🎯 What it does: This paper studies a symbol-based stochastic variance reduction algorithm (SSVR) and provides convergence analysis in single-machine, finite summation, and distributed environments.
Efficient Sketches for Training Data Attribution and Studying the Loss Landscape
Andrea Schioppa (Google DeepMind)
OptimizationComputational EfficiencyTransformerLarge Language ModelSupervised Fine-TuningText
🎯 What it does: This paper proposes and validates three novel gradient/HVP Sketch algorithms (AFFD, AFJL, QK) for efficiently storing gradients, Hessian-vector products, and computing intrinsic dimensions and Hessian spectra on GPU/TPU. These algorithms are applied to training data attribution, feature dimension estimation, and Hessian spectral analysis of pre-trained language models.
Efficient Streaming Algorithms for Graphlet Sampling
Yann Bourreau (Cispa Helmholtz Center for Information Security), Mauro Sozio (Institut Polytechnique de Paris)
OptimizationComputational EfficiencyGraph Neural NetworkGraph
🎯 What it does: An efficient graphlet sampling algorithm STREAM-UGS has been developed under the semi-streaming model, capable of preprocessing the graph in O(log n) passes, and subsequently generating Θ(M/k^{O(k)}) independent uniform k-graphlets in parallel every O(k) passes; the algorithm requires only Ω(n log n) words (≈ n log n bits) of storage, and can complete sampling on actual large graphs in 30-40 passes with memory usage below the size of the edge list.
Efficient Temporal Action Segmentation via Boundary-aware Query Voting
Peiyao Wang (Stony Brook University), Haibin Ling (Stony Brook University)
SegmentationComputational EfficiencyTransformerVideo
🎯 What it does: This paper presents BaFormer, a boundary-aware query network based on Transformer, which transforms dense frame-by-frame action segmentation into sparse query classification, achieving efficient temporal action segmentation.
EfficientCAPER: An End-to-End Framework for Fast and Robust Category-Level Articulated Object Pose Estimation
Xinyi Yu (Zhejiang University of Technology), Liu Liu (Hefei University of Technology)
Pose EstimationPoint Cloud
🎯 What it does: An end-to-end EfficientCAPER framework has been developed for 6D joint pose estimation of category-level articulated objects, eliminating post-optimization and solving steps.
Efficiently Learning Significant Fourier Feature Pairs for Statistical Independence Testing
Yixin Ren (Fudan University), Shuigeng Zhou (Fudan University)
Tabular
🎯 What it does: Proposes a method for efficiently learning significant Fourier features to maximize the power of HSIC independence tests.
EffiLearner: Enhancing Efficiency of Generated Code via Self-Optimization
Dong HUANG, Jie Zhang
OptimizationComputational EfficiencyAI Code AssistantTransformerLarge Language ModelPrompt EngineeringTextBenchmark
🎯 What it does: The EFFI-LEARNER framework is proposed, which utilizes execution time and memory usage profiles for self-optimization after generating code with LLM, thereby improving code efficiency.
EgoChoir: Capturing 3D Human-Object Interaction Regions from Egocentric Views
Yuhang Yang (University of Science and Technology of China), Zheng-Jun Zha (University of Science and Technology of China)
Object DetectionPose EstimationGraph Neural NetworkVideoPoint Cloud
🎯 What it does: This paper proposes a method to simultaneously predict 3D human contact points and object usable areas from first-person perspective videos.
EGODE: An Event-attended Graph ODE Framework for Modeling Rigid Dynamics
Jingyang Yuan (Peking University), Ming Zhang (Peking University)
Graph Neural NetworkGraphPhysics RelatedOrdinary Differential Equation
🎯 What it does: This study investigates rigid body dynamics modeling and proposes the Event-attend Graph ODE (EGODE) framework to jointly simulate rigid body collisions and evolution using continuous ODE and event modules.
EGonc : Energy-based Open-Set Node Classification with substitute Unknowns
Qin Zhang (Shenzhen University), Xiaojun Chen (Shenzhen University)
ClassificationGraph Neural NetworkGenerative Adversarial NetworkGraph
🎯 What it does: A method for open set node classification based on energy models and generative unknown samples, called EGonc, is proposed.
EGSST: Event-based Graph Spatiotemporal Sensitive Transformer for Object Detection
Sheng Wu (Fudan University), Bo Hu (Fudan University)
Object DetectionAutonomous DrivingGraph Neural NetworkTransformerImage
🎯 What it does: A target detection framework for event cameras, EGSST, based on graph neural networks and a lightweight linear Vision Transformer, is proposed, achieving efficient low-latency detection through a spatiotemporal sensitive module and an adaptive temporal activation controller.
EigenVI: score-based variational inference with orthogonal function expansions
Diana Cai (Flatiron Institute), Lawrence K. Saul (Flatiron Institute)
Score-based ModelMultimodality
🎯 What it does: This paper proposes EigenVI, a black-box variational inference method that utilizes orthogonal function expansion for score matching, avoiding the gradient optimization and hyperparameter dependence of traditional BBVI.
einspace: Searching for Neural Architectures from Fundamental Operations
Linus Ericsson (University of Edinburgh), Elliot J. Crowley (University of Edinburgh)
Neural Architecture SearchConvolutional Neural NetworkTransformerImageTextAudio
🎯 What it does: This paper proposes einspace, a neural architecture search space based on parameterized probabilistic context-free grammar (PCFG), capable of representing various networks from ResNet, ViT, MLP-Mixer to custom mixed structures;
ElasTST: Towards Robust Varied-Horizon Forecasting with Elastic Time-Series Transformer
Jiawen Zhang (Hong Kong University of Science and Technology), Jia Li (Hong Kong University of Science and Technology)
TransformerTime Series
🎯 What it does: A non-autoregressive time series Transformer model, ElasTST, has been designed and implemented to maintain robustness under different prediction durations.
Elliptical Attention
Stefan Nielsen, Tan Minh Nguyen
TransformerImageText
🎯 What it does: An elliptical attention mechanism based on Mahalanobis distance is proposed, improving the traditional self-attention spherical kernel to an elliptical kernel, enhancing representation diversity and robustness.
Elo Uncovered: Robustness and Best Practices in Language Model Evaluation
Meriem Boubdir (Cohere For AI), Marzieh Fadaee (Cohere For AI)
Large Language ModelText
🎯 What it does: This paper conducts a systematic reliability and transitivity analysis of the Elo rating system, exploring its applicability and limitations in the evaluation of large language models (LLMs);
Elucidating the Design Space of Dataset Condensation
Shitong Shao (Mohamed bin Zayed University of AI), Zhiqiang Shen (Mohamed bin Zayed University of AI)
Knowledge DistillationData-Centric LearningConvolutional Neural NetworkImage
🎯 What it does: This paper explores the design space of dataset distillation and proposes the EDC framework to efficiently generate small synthetic datasets.
EM Distillation for One-step Diffusion Models
Sirui Xie (Google Research), Ruiqi Gao (Google Research)
GenerationKnowledge DistillationDiffusion modelScore-based ModelImage
🎯 What it does: One-step distillation of pre-trained diffusion models to learn a one-step generator.
Embedding Dimension of Contrastive Learning and $k$-Nearest Neighbors
Dmitrii Avdiukhin (Northwestern University), Grigory Yaroslavtsev (George Mason University)
OptimizationRepresentation LearningContrastive LearningImage
🎯 What it does: This paper studies the minimum embedding dimension required to satisfy distance comparison constraints in contrastive learning and k-nearest neighbors (k-NN) data, providing both upper and lower bounds.
Embedding Trajectory for Out-of-Distribution Detection in Mathematical Reasoning
Yiming Wang (Shanghai Jiao Tong University), Rui Wang (Shanghai Jiao Tong University)
GenerationAnomaly DetectionTransformerLarge Language ModelText
🎯 What it does: This study proposes an OOD detection method based on embedded trajectory fluctuations—TV Score—specifically designed for generative language models in mathematical reasoning scenarios.
Embedding-Aligned Language Models
Guy Tennenholtz (Google Research), Craig Boutilier (Google Research)
Recommendation SystemTransformerLarge Language ModelReinforcement LearningText
🎯 What it does: The EAGLE framework is proposed, utilizing reinforcement learning to treat pre-trained large language models (LLMs) as environments, guiding LLMs to generate text that meets domain requirements through embedding space objectives repeatedly.
Emergence of heavy tails in homogenized stochastic gradient descent
Zhe Jiao (Northwestern Polytechnical University), Martin Keller-Ressel (Technische Universität Dresden)
OptimizationImageTabularStochastic Differential Equation
🎯 What it does: Analyzed the heavy-tailed phenomenon in the parameter distribution obtained from homogenized stochastic gradient descent (hSGD) in regularized linear regression, and provided explicit upper and lower bounds for the tail index;
Emergence of Hidden Capabilities: Exploring Learning Dynamics in Concept Space
Core Francisco Park (Harvard University), Hidenori Tanaka (Harvard University)
GenerationData SynthesisDiffusion modelImage
🎯 What it does: This paper proposes a conceptual space framework to analyze the dynamic evolution of generative models' ability to recognize, decouple, and manipulate concepts during the learning process. By synthesizing 2D object data and conducting experiments with CelebA, it explores the impact of conceptual signals on learning speed and order, revealing that hidden capabilities can suddenly emerge during training but are often masked by ordinary prompts.
Emotion-LLaMA: Multimodal Emotion Recognition and Reasoning with Instruction Tuning
Zebang Cheng (Shenzhen Technology University), Alexander G Hauptmann
RecognitionTransformerLarge Language ModelSupervised Fine-TuningVideoTextMultimodalityAudio
🎯 What it does: Designed and trained Emotion-LLaMA to achieve emotion recognition and reasoning through multi-modal inputs of audio, visual, and text.
Empowering Active Learning for 3D Molecular Graphs with Geometric Graph Isomorphism
Ronast Subedi (Florida State University), Yi Liu (Stony Brook University)
OptimizationDrug DiscoveryGraph Neural NetworkReinforcement LearningGraph
🎯 What it does: This paper proposes an active learning framework for 3D molecular graphs, utilizing diversity and uncertainty assessment to achieve sample selection through quadratic programming, combined with new geometric isometric transformations and Bayesian graph neural networks for 3D molecular property prediction.
Empowering Visible-Infrared Person Re-Identification with Large Foundation Models
Zhangyi Hu (Wuhan University), Mang Ye (Wuhan University)
RecognitionRetrievalConvolutional Neural NetworkLarge Language ModelSupervised Fine-TuningVision Language ModelContrastive LearningImageTextMultimodality
🎯 What it does: A visible-infrared person re-identification framework based on large foundational models is proposed, utilizing generative visual language models and large language models to automatically generate text descriptions, enriching infrared modality information and enhancing cross-modal retrieval performance.
EMR-Merging: Tuning-Free High-Performance Model Merging
Chenyu Huang (Fudan University), Wanli Ouyang (Shanghai AI Laboratory)
TransformerTextMultimodality
🎯 What it does: A new model merging method called EMR-MERGING is proposed, aimed at merging models from different tasks by selecting a unified model and a lightweight task-specific regulator, without the need for additional tuning or data.
EMVP: Embracing Visual Foundation Model for Visual Place Recognition with Centroid-Free Probing
Qibo Qiu (Zhejiang University), Xiaofei He (Zhejiang University)
RecognitionRetrievalTransformerSupervised Fine-TuningContrastive LearningImage
🎯 What it does: This paper proposes an efficient fine-tuning pipeline EMVP for Visual Place Recognition (VPR) based on a Visual Foundation Model (VFM), incorporating a Centroid-Free Probing (CFP) and Dynamic Power Normalization (DPN) module, achieving parameter-efficient and high-performance fine-tuning.
Enabling Adaptive Agent Training in Open-Ended Simulators by Targeting Diversity
Robby Costales (University of Southern California), Stefanos Nikolaidis (University of Southern California)
Robotic IntelligenceMeta LearningRecurrent Neural NetworkReinforcement LearningAgentic AISequential
🎯 What it does: DIVA is proposed, a semi-supervised environment design method based on Quality Diversity (QD), used to generate diverse training tasks in an initially unstructured simulator, thereby training adaptable Meta-RL agents.
ENAT: Rethinking Spatial-temporal Interactions in Token-based Image Synthesis
Zanlin Ni (Tsinghua University), Gao Huang (Tsinghua University)
GenerationData SynthesisTransformerDiffusion modelImage
🎯 What it does: This paper studies the internal mechanisms of Token-based Non-Autoregressive Transformers (NAT) for image synthesis and proposes the EfficientNAT (ENAT) model, which significantly improves generation quality and efficiency through spatial decoupling and temporal feature reuse.
End-To-End Causal Effect Estimation from Unstructured Natural Language Data
Nikita Dhawan (University of Toronto), Chris J. Maddison (University of Toronto)
TransformerLarge Language ModelText
🎯 What it does: An end-to-end causal effect estimator named NATURAL was constructed by utilizing large language models to automatically infer the conditional distributions of treatment, outcomes, and covariates from free text, combined with classical causal estimation methods.
End-to-end Learnable Clustering for Intent Learning in Recommendation
Yue Liu (Ant Group), Wenliang Zhong (Ant Group)
Recommendation SystemTransformerContrastive LearningTabularSequential
🎯 What it does: An end-to-end learnable clustering intent learning framework ELCRec is proposed to enhance the joint optimization of user intent modeling and behavior representation in recommendation systems.
End-to-End Ontology Learning with Large Language Models
Andy Lo (University of Cambridge), Mateja Jamnik (University of Cambridge)
TransformerLarge Language ModelSupervised Fine-TuningTextGraph
🎯 What it does: This paper proposes an end-to-end framework called OLLM for constructing ontologies using large language models (LLM), which directly generates ontology subgraphs from documents and aggregates them to form a complete ontology.
End-to-End Video Semantic Segmentation in Adverse Weather using Fusion Blocks and Temporal-Spatial Teacher-Student Learning
Xin Yang (National University of Singapore), Robby T. Tan (National University of Singapore)
SegmentationDomain AdaptationKnowledge DistillationConvolutional Neural NetworkOptical FlowVideo
🎯 What it does: This paper proposes an end-to-end, optical flow-free domain adaptation method for video semantic segmentation, utilizing a fusion block, a spatiotemporal teacher-student learning framework, and temporal weather degradation augmentation to enhance segmentation performance under adverse weather conditions.
Energy-based Epistemic Uncertainty for Graph Neural Networks
Dominik Fuchsgruber (Technical University of Munich), Stephan Günnemann (Technical University of Munich)
ClassificationDomain AdaptationGraph Neural NetworkGaussian SplattingGraph
🎯 What it does: A post-hoc graph energy model (GEBM) is proposed to estimate empirical uncertainty in graph neural networks.
Energy-based Hopfield Boosting for Out-of-Distribution Detection
Claus Hofmann (Johannes Kepler University), Sepp Hochreiter (Johannes Kepler University)
Anomaly DetectionConvolutional Neural NetworkImage
🎯 What it does: A method for OOD detection named Hopfield Boosting is designed and implemented, utilizing the energy function of modern Hopfield networks and auxiliary anomalous data to enhance the model's ability to recognize discrete distribution samples through weighted sampling of weak learners.
Energy-Based Modelling for Discrete and Mixed Data via Heat Equations on Structured Spaces
Tobias Schröder (Imperial College London), Andrew B. Duncan
GenerationData SynthesisComputational EfficiencyImageTabular
🎯 What it does: An energy-based model (EBM) training method is proposed in discrete and mixed state spaces that does not require MCMC, using energy discrepancy loss and constructing perturbations through discrete heat equations.
Energy-Guided Continuous Entropic Barycenter Estimation for General Costs
Alexander Kolesov (Skolkovo Institute of Science and Technology), Alexander Korotin (Skolkovo Institute of Science and Technology)
GenerationOptimizationGenerative Adversarial NetworkImage
🎯 What it does: An energy-guided continuous entropy regularization OT centroid estimation algorithm is proposed, which can handle arbitrary OT costs and supports constraining centroids on image manifolds generated by pre-trained generative models.
Enhancing Chess Reinforcement Learning with Graph Representation
Tomas Rigaux (Kyoto University), Hisashi Kashima (Kyoto University)
Graph Neural NetworkReinforcement LearningGraph
🎯 What it does: This paper presents AlphaGateau, a reinforcement learning framework based on graph neural networks for playing international chess, which replaces the CNN architecture of AlphaZero and introduces a new GATEAU layer.
Enhancing Consistency-Based Image Generation via Adversarialy-Trained Classification and Energy-Based Discrimination
Shelly Golan (Technion), Michael Elad (Technion)
GenerationOptimizationConvolutional Neural NetworkGenerative Adversarial NetworkImage
🎯 What it does: Post-processing images generated by the consistency model using a joint robust classifier-discriminator to enhance image perceptual quality.
Enhancing Diversity in Bayesian Deep Learning via Hyperspherical Energy Minimization of CKA
David Smerkous (Oregon State University), Li Fuxin (Oregon State University)
Anomaly DetectionImage
🎯 What it does: This paper constructs a new regularization objective by combining centralized kernel alignment (CKA) with spherical energy (HE) at the network function layer to enhance model diversity in Bayesian deep learning, thereby improving uncertainty estimation and anomaly detection performance.
Enhancing Domain Adaptation through Prompt Gradient Alignment
Hoang Phan (New York University), Trung Le (Monash University)
Domain AdaptationPrompt EngineeringImage
🎯 What it does: A prompt learning-based unsupervised domain adaptation framework is proposed, which unifies the losses of the source and target domains into different objectives through multi-objective optimization, and optimizes shared prompts via gradient alignment and gradient norm regularization to enhance cross-domain generalization ability.
Enhancing Efficiency of Safe Reinforcement Learning via Sample Manipulation
Shangding Gu (University of California), Ming Jin (Virginia Tech)
Computational EfficiencyReinforcement LearningTabular
🎯 What it does: A safe reinforcement learning algorithm ESPO based on dynamic adjustment of sampling size with gradient conflict is proposed, significantly improving sample efficiency while maintaining safety constraints.
Enhancing Feature Diversity Boosts Channel-Adaptive Vision Transformers
Chau Pham (Boston University), Bryan A. Plummer (Boston University)
ClassificationDomain AdaptationTransformerImage
🎯 What it does: A multi-channel visual Transformer named DiChaViT is proposed, aimed at enhancing feature diversity and robustness, particularly for multi-channel image data;
Enhancing Graph Transformers with Hierarchical Distance Structural Encoding
Yuankai Luo (Beihang University), Xiao-Ming Wu (Hong Kong Polytechnic University)
ClassificationComputational EfficiencyRepresentation LearningGraph Neural NetworkTransformerGraph
🎯 What it does: A Hierarchical Distance Structure Encoding (HDSE) is proposed and seamlessly integrated into the attention mechanism of existing graph Transformer models to enhance the modeling capability of multi-level hierarchies and long-range dependencies in graphs.
Enhancing In-Context Learning Performance with just SVD-Based Weight Pruning: A Theoretical Perspective
Xinhao Yao (Renmin University of China), Yong Liu (Renmin University of China)
TransformerLarge Language ModelText
🎯 What it does: This study investigates the impact of SVD-based weight pruning on the performance of large language models (LLMs) in in-context learning (ICL) and proposes theoretical explanations and practical algorithms.
Enhancing Large Language Models through Adaptive Tokenizers
Mengyu Zheng (University of Sydney), Yunhe Wang (Huawei Noah's Ark Lab)
TransformerLarge Language ModelText
🎯 What it does: An adaptive tokenizer ADAT based on LLM feedback is proposed, which gradually prunes the vocabulary by monitoring the model's perplexity and token-level cross-entropy loss, achieving dynamic synchronization with the model.
Enhancing Large Vision Language Models with Self-Training on Image Comprehension
Yihe Deng (University of California), Wei Wang (Stanford University)
GenerationData SynthesisOptimizationTransformerSupervised Fine-TuningVision Language ModelImageTextMultimodality
🎯 What it does: A two-stage self-training framework called STIC is proposed, which generates unlabeled image description data through the model itself for preference learning, and subsequently performs description injection fine-tuning to enhance the image understanding and reasoning capabilities of large visual language models.
Enhancing LLM Reasoning via Vision-Augmented Prompting
Ziyang Xiao (Zhejiang University), Gang Chen (Singapore University of Social Sciences)
TransformerLarge Language ModelPrompt EngineeringVision Language ModelMultimodalityTime SeriesChain-of-Thought
🎯 What it does: Proposes the Vision-Augmented Prompting (VAP) framework, which utilizes large language models to automatically generate and iteratively draw images, and then completes tasks in conjunction with text reasoning, thereby enhancing the model's reasoning capabilities.
Enhancing LLM’s Cognition via Structurization
Kai Liu (Zhejiang University), Jieping Ye (Alibaba Cloud)
Knowledge DistillationTransformerLarge Language ModelPrompt EngineeringTextBenchmark
🎯 What it does: Proposes to enhance the cognitive and reasoning abilities of large language models by converting long texts into a three-layer Scope–Aspect–Description structured representation.
Enhancing Motion in Text-to-Video Generation with Decomposed Encoding and Conditioning
Penghui Ruan (Hong Kong Polytechnic University), Yuhui Shi (Southern University of Science and Technology)
GenerationData SynthesisTransformerDiffusion modelContrastive LearningOptical FlowVideoText
🎯 What it does: The DEcomposed MOtion (DEMO) framework is proposed, which significantly improves the text-driven video motion generation effect by decomposing text encoding and conditions into content and motion parts.
Enhancing Multiple Dimensions of Trustworthiness in LLMs via Sparse Activation Control
Yuxin Xiao (Zhejiang University), Jieping Ye (Alibaba Cloud)
TransformerLarge Language ModelSupervised Fine-TuningText
🎯 What it does: Utilizing Sparse Activation Control (SAC) for untrained regulation of multi-dimensional credibility (such as factuality, bias, and safety) at the attention head layer of LLMs.
Enhancing Preference-based Linear Bandits via Human Response Time
Shen Li (Massachusetts Institute of Technology), Julie Shah
Recommendation SystemOptimizationReinforcement LearningDiffusion modelTime Series
🎯 What it does: This paper proposes a method that utilizes human response time to enhance the learning efficiency of preference-based linear bandits, embedding this method into a fixed-budget optimal arm identification algorithm.
Enhancing Protein Mutation Effect Prediction through a Retrieval-Augmented Framework
Ruihan Guo (Helixon Research), Jianzhu Ma (Tsinghua University)
Protein Structure PredictionBiomedical DataRetrieval-Augmented Generation
🎯 What it does: A Structural Motif Embedding Database (SMEDB) was constructed, which retrieves local structural fragments through ESM-IF based vector search (MSMA) and aggregates retrieval information using Multi-Structural Motif Invariant Point Attention (MSM-IPA) to predict the ΔΔG of protein mutations.
Enhancing Reasoning Capabilities of LLMs via Principled Synthetic Logic Corpus
Terufumi Morishita (Advanced AI Innovation Center Hitachi), Yasuhiro Sogawa (Advanced AI Innovation Center Hitachi)
TransformerLarge Language ModelSupervised Fine-TuningText
🎯 What it does: Enhancing reasoning capabilities in large language models through Additional Logic Training (ALT).
Enhancing Robustness in Deep Reinforcement Learning: A Lyapunov Exponent Approach
Rory Young (University of Glasgow), Nicolas Pugeault (University of Glasgow)
Reinforcement Learning
🎯 What it does: This paper studies the sensitivity of deep reinforcement learning (RL) policies to small state perturbations in continuous control tasks and proposes improving policy robustness through maximum Lyapunov exponent (MLE) regularization.
Enhancing Robustness of Graph Neural Networks on Social Media with Explainable Inverse Reinforcement Learning
Yuefei Lyu (Beijing University of Posts and Telecommunications), Xi Zhang (Beijing University of Posts and Telecommunications)
Explainability and InterpretabilityAdversarial AttackGraph Neural NetworkReinforcement LearningMixture of ExpertsGraph
🎯 What it does: This paper proposes a framework based on maximum entropy inverse reinforcement learning and mixture of experts (MoE) to reconstruct multi-source graph attack strategies in social networks, and enhances the robustness of GNNs through data augmentation or adversarial training using the generated attack samples.
Enhancing Robustness of Last Layer Two-Stage Fair Model Corrections
Nathan Stromberg (Arizona State University), Lalitha Sankar (Arizona State University)
ClassificationSupervised Fine-TuningImage
🎯 What it does: This paper proposes the use of kNN label propagation preprocessing before fine-tuning the last layer, significantly improving the worst group accuracy in the presence of symmetric label noise.
Enhancing Semi-Supervised Learning via Representative and Diverse Sample Selection
Qian Shao (Zhejiang University), Jian Wu (Zhejiang University)
ClassificationOptimizationConvolutional Neural NetworkContrastive LearningImage
🎯 What it does: A method for unsupervised sample selection based on α-MMD metric (RDSS) is proposed for sample labeling under limited annotation budgets in semi-supervised learning.
Enhancing Zero-Shot Vision Models by Label-Free Prompt Distribution Learning and Bias Correcting
Xingyu Zhu (University of Science and Technology of China), Hanwang Zhang (Nanyang Technological University)
ClassificationDomain AdaptationTransformerContrastive LearningImage
🎯 What it does: A completely label-free, training-free, and hyperparameter-free framework called Frolic is proposed to enhance the performance of zero-shot visual models like CLIP.
EnOF-SNN: Training Accurate Spiking Neural Networks via Enhancing the Output Feature
Yufei Guo (Intelligent Science and Technology Academy of CASIC), Zhe Ma (Intelligent Science and Technology Academy of CASIC)
Knowledge DistillationRepresentation LearningSpiking Neural NetworkImage
🎯 What it does: Proposed L AF loss and RepAct method to enhance the output feature expressiveness of SNN.
Enriching Disentanglement: From Logical Definitions to Quantitative Metrics
Yivan Zhang (University of Tokyo), Masashi Sugiyama (RIKEN AIP)
Representation Learning
🎯 What it does: This paper proposes a systematic method to transform logical properties (such as decomposability and informativeness) in representation learning into quantifiable metrics, and provides various differentiable metrics that can serve as learning objectives through this method.
Ensemble Learning for Heterogeneous Large Language Models with Deep Parallel Collaboration
Yichong Huang (Harbin Institute of Technology), Bing Qin (Harbin Institute of Technology)
TransformerLarge Language ModelTextMultimodalityBenchmark
🎯 What it does: An untrained multimodal large language model ensemble framework called DEEPEN is proposed, which aligns the probability distributions of different vocabularies using relative representations and fuses them, mapping the aggregated results back to the main model space through inverse transformation search.
Ensemble sampling for linear bandits: small ensembles suffice
David Janz (University of Oxford), Csaba Szepesvari
Reinforcement Learning
🎯 What it does: Proposed and rigorously analyzed the ensemble sampling algorithm under the linear Bandit problem, proving that in the case of dimension d and time window T, using a small ensemble of size approximately d log T can achieve an approximate O(√T) regret upper bound;
EnsIR: An Ensemble Algorithm for Image Restoration via Gaussian Mixture Models
Shangquan Sun (Institute of Information Engineering Chinese Academy of Sciences), Xiaochun Cao (Shenzhen Campus of Sun Yat-sen University)
RestorationSuper ResolutionImage
🎯 What it does: A post-training image restoration ensemble algorithm called EnsIR based on Gaussian Mixture Model is proposed, which utilizes pixel value partitioning and EM learning interval weights stored in a LUT for fast inference.
Entity Alignment with Noisy Annotations from Large Language Models
Shengyuan Chen (Hong Kong Polytechnic University), Xiao Huang (Hong Kong Polytechnic University)
Graph Neural NetworkLarge Language ModelGraphBenchmark
🎯 What it does: This paper proposes a framework called LLM4EA, which utilizes large language models (LLM) to generate pseudo-labels under a limited query budget, actively selects entity pairs, and refines unsupervised labels, thereby achieving entity alignment across domains and languages in knowledge graphs.
Entropy testing and its application to testing Bayesian networks
Clement Louis Canonne, Qiping Yang
🎯 What it does: Proposed and solved the entropy identity testing problem: given samples from a known distribution q and an unknown distribution p, determine whether p = q or the difference between H(p) and H(q) is at least ε. The algorithm was then applied to identity testing in Bayesian networks with a maximum in-degree of d.
Entropy-regularized Diffusion Policy with Q-Ensembles for Offline Reinforcement Learning
Ruoqi Zhang (Uppsala University), Per Mattsson (Uppsala University)
Reinforcement LearningDiffusion modelTabularBenchmarkStochastic Differential Equation
🎯 What it does: A diffusion policy based on entropy regularization is proposed in offline reinforcement learning, incorporating Q-ensemble to enhance the handling of uncertainty and exploration capabilities with offline data.
Entrywise error bounds for low-rank approximations of kernel matrices
Alexander Modell (Imperial College London)
Tabular
🎯 What it does: This study investigates the element-wise error upper bound of the low-rank approximation of kernel matrices obtained through truncated feature decomposition and proves the delocalization property of the corresponding eigenvectors.
EPIC: Effective Prompting for Imbalanced-Class Data Synthesis in Tabular Data Classification via Large Language Models
Jinhee Kim, Jaegul Choo
ClassificationData SynthesisTransformerLarge Language ModelPrompt EngineeringTabular
🎯 What it does: Utilizing large language models (LLM) to generate high-quality, balanced tabular data through carefully designed prompts without additional parameter training, thereby alleviating the issue of class imbalance.
Epipolar-Free 3D Gaussian Splatting for Generalizable Novel View Synthesis
Zhiyuan Min (Zhejiang University), Yi Yang (Zhejiang University)
GenerationData SynthesisTransformerGaussian SplattingImage
🎯 What it does: A general 3D Gaussian drawing model is proposed that can directly generate new scene perspective images from a sparse perspective.
Episodic Future Thinking Mechanism for Multi-agent Reinforcement Learning
Dongsu Lee (Soongsil University), Minhae Kwon (Soongsil University)
Autonomous DrivingReinforcement Learning
🎯 What it does: A multi-agent reinforcement learning mechanism based on episodic future thinking (EFT) is proposed, which infers the roles of other agents through multi-role strategies, predicts their future actions, and simulates future observations to achieve forward-looking decision-making.
Equivariant Blurring Diffusion for Hierarchical Molecular Conformer Generation
Jiwoong Park (Texas A&M University), Yang Shen (Texas A&M University)
GenerationDrug DiscoveryGraph Neural NetworkDiffusion modelGraph
🎯 What it does: In molecular conformation generation, a hierarchical equivariant fuzzy diffusion (EBD) model is proposed, which first generates a coarse structure at the fragment level from the molecular graph and then refines it into a complete atomic-level conformation through an equivariant network.
Equivariant Machine Learning on Graphs with Nonlinear Spectral Filters
Ya-Wei Eileen Lin (Technion), Ron Levie (Technion)
ClassificationGraph Neural NetworkGraph
🎯 What it does: This paper proposes a nonlinear spectral filter (NLSF) that can be transferred between different graph structures and is invariant to graph functional translation symmetry, and constructs the corresponding graph neural network model.
Equivariant Neural Diffusion for Molecule Generation
François R J Cornet (Technical University of Denmark), Christian A. Naesseth (University of Amsterdam)
GenerationDrug DiscoveryGraph Neural NetworkDiffusion modelGraph
🎯 What it does: A new 3D molecular generation diffusion model called Equivariant Neural Diffusion (END) is proposed, achieving equivariance under Euclidean transformations and enhancing generation quality through a learnable forward process.
Equivariant spatio-hemispherical networks for diffusion MRI deconvolution
Axel Elaldi (New York University), Neel Dey (Massachusetts Institute of Technology)
RestorationComputational EfficiencyConvolutional Neural NetworkBiomedical DataMagnetic Resonance ImagingDiffusion Tensor Imaging
🎯 What it does: An efficient U-Net based on E(3)×SO(3) equivariant convolution has been developed for diffusion MRI fiber deconvolution, significantly reducing computation time and memory consumption.
Era3D: High-Resolution Multiview Diffusion using Efficient Row-wise Attention
Peng Li (Hong Kong University of Science and Technology), Yike Guo (Hong Kong University of Science and Technology)
GenerationData SynthesisDiffusion modelImageMesh
🎯 What it does: This paper proposes and implements Era3D, a multi-view diffusion model capable of generating high-resolution (512×512) multi-view images from a single-view image and performing 3D reconstruction.
Erasing Undesirable Concepts in Diffusion Models with Adversarial Preservation
Anh Tuan Bui, Dinh Phung (Monash University)
GenerationData SynthesisDiffusion modelGenerative Adversarial NetworkImage
🎯 What it does: A method for concept elimination in diffusion models based on adversarial concept protection is proposed, which can eliminate specified unacceptable concepts while maintaining the generation quality of other concepts.
Error Analysis of Spherically Constrained Least Squares Reformulation in Solving the Stackelberg Prediction Game
Xiyuan Li (Wuhan University), Weiwei Liu (Wuhan University)
OptimizationTabular
🎯 What it does: This paper studies the estimation error of the spherical constrained least squares reconstruction method (SCLS) in solving Stackelberg prediction games and proposes a theoretical error analysis based on auxiliary optimization problems.
Error Correction Output Codes for Robust Neural Networks against Weight-errors: A Neural Tangent Kernel Point of View
Anlan Yu (Lehigh University), Zhiyuan Yan (Lehigh University)
ClassificationOptimizationConvolutional Neural NetworkImage
🎯 What it does: This paper studies the impact of Error-Correcting Output Codes (ECOC) on the robustness of deep networks in the presence of weight errors under the NTK framework, providing a theoretical analysis.
ESPACE: Dimensionality Reduction of Activations for Model Compression
Charbel Sakr (NVIDIA Research), Brucek Khailany (NVIDIA Research)
CompressionLarge Language ModelText
🎯 What it does: LLM model compression is achieved by projecting the activation tensor onto pre-calibrated principal components.
Estimating Ego-Body Pose from Doubly Sparse Egocentric Video Data
Seunggeun Chi (Purdue University), Kwonjoon Lee (Honda Research Institute USA)
Pose EstimationDiffusion modelAuto EncoderVideo
🎯 What it does: Estimate the full-body posture of HMD users by utilizing head tracking signals and temporally completing and generating based on sparsely appearing hand detection results in the field of view.
Estimating Epistemic and Aleatoric Uncertainty with a Single Model
Matthew Albert Chan, Christopher Metzler
RestorationAnomaly DetectionDiffusion modelTime SeriesBiomedical DataComputed Tomography
🎯 What it does: A single model framework (HyperDM) is proposed that utilizes Bayesian hypernetworks and diffusion models to estimate and separate the epistemic uncertainty and aleatoric uncertainty in machine learning predictions.
Estimating Generalization Performance Along the Trajectory of Proximal SGD in Robust Regression
Kai Tan (Rutgers University), Pierre C Bellec
OptimizationTabular
🎯 What it does: This study investigates how to estimate and track generalization error during the iterative process when using Gradient Descent (GD), Stochastic Gradient Descent (SGD), and their proximal variants in high-dimensional robust regression problems, providing observable and consistent risk estimators.
Estimating Heterogeneous Treatment Effects by Combining Weak Instruments and Observational Data
Miruna Oprescu (Cornell University), Nathan Kallus (Cornell University)
TabularFinance Related
🎯 What it does: A two-stage framework is developed to estimate heterogeneous treatment effects and correct biases by combining weak IV with observational data.
Estimating the Hallucination Rate of Generative AI
Andrew Jesson (Columbia University), David Blei (Columbia University)
GenerationLarge Language ModelText
🎯 What it does: A method for estimating the posterior hallucination rate from a Bayesian perspective is proposed to assess the probability of hallucinations occurring in generative AI during contextual learning.
ET-Flow: Equivariant Flow-Matching for Molecular Conformer Generation
Majdi Hassan (Mila), Dominique Beaini (Valence Labs)
GenerationDrug DiscoveryTransformerFlow-based ModelGraph
🎯 What it does: A generative model based on equivariant flow matching is proposed for generating low-energy three-dimensional conformations of given molecular graphs.
ETO:Efficient Transformer-based Local Feature Matching by Organizing Multiple Homography Hypotheses
Junjie Ni (Zhejiang University), Hujun Bao (Zhejiang University)
Object DetectionSegmentationComputational EfficiencyTransformerSupervised Fine-TuningImage
🎯 What it does: This paper proposes an efficient Transformer local feature matching framework called ETO, which utilizes the Homography Hypothesis of multi-plane units to aggregate a large number of patches, significantly reducing the tokens required by the Transformer; during the fine matching stage, it employs uni-directional cross-attention and removes self-attention to further lower the computational load; it models the reallocation problem as a segmentation task and uses indirect supervision to achieve precise matching.
Euclidean distance compression via deep random features
Brett Leroux (University of California), Luis Rademacher (University of California)
RetrievalCompressionConvolutional Neural NetworkTextPoint Cloud
🎯 What it does: This paper proposes a sketch that compresses a point set into discrete cubes through deep random feature mapping φℓ, which can recover the squared distances between points while maintaining a (1±ε) multiplicative error.
Evaluate then Cooperate: Shapley-based View Cooperation Enhancement for Multi-view Clustering
Fangdi Wang (National University of Defence Technology), En Zhu (National University of Defence Technology)
Auto EncoderImage
🎯 What it does: Proposes a Shapley value-based perspective contribution assessment and collaboration enhancement method for multi-view clustering, achieving unsupervised perspective collaborative optimization.
Evaluating alignment between humans and neural network representations in image-based learning tasks
Can Demircan (Max Planck Institute for Biological Cybernetics), Eric Schulz (Max Planck Institute for Biological Cybernetics)
Representation LearningReinforcement Learning from Human FeedbackTransformerContrastive LearningImageMultimodality
🎯 What it does: Evaluate the alignment of representations from pre-trained neural networks with human learning trajectories in two natural image-based learning tasks (category learning and reward learning).
Evaluating the design space of diffusion-based generative models
Yuqing Wang (University of California Berkeley), Molei Tao (Georgia Institute of Technology)
GenerationData SynthesisDiffusion modelStochastic Differential Equation
🎯 What it does: The paper conducts a unified non-asymptotic error analysis of the training and sampling processes of diffusion models, providing the convergence of gradient descent during training and an upper bound on the VE SDE error during sampling.
Evaluating the World Model Implicit in a Generative Model
Keyon Vafa (Harvard University), Sendhil Mullainathan (MIT)
GenerationData SynthesisCompressionTransformerWorld ModelTime SeriesSequential
🎯 What it does: This study investigates whether generative models can implicitly learn and accurately recover the world model of the domain they are trained on (within the framework of deterministic finite automata), proposing new evaluation metrics and conducting experiments on maps, board games, and logic puzzles.