NeurIPS 2024 Papers — Page 17
Conference on Neural Information Processing Systems · 4035 papers
HardCore Generation: Generating Hard UNSAT Problems for Data Augmentation
Joseph Cotnareanu (McGill University), Mark Coates (McGill University)
GenerationData SynthesisOptimizationGraph Neural NetworkGraph
🎯 What it does: A data augmentation method for quickly generating hard UNSAT SAT instances has been designed and implemented.
Hardness of Learning Neural Networks under the Manifold Hypothesis
Bobak Kiani, Melanie Weber (Harvard University)
ClassificationRecognitionImage
🎯 What it does: This paper studies the feasibility of neural network learning under the manifold hypothesis and provides geometric conditions for learnable and unlearnable manifolds.
Harmonizing Stochasticity and Determinism: Scene-responsive Diverse Human Motion Prediction
Zhenyu Lou (Zhejiang University), Hong Zhou (Zhejiang University)
GenerationPose EstimationTransformerDiffusion modelPoint Cloud
🎯 What it does: This paper proposes the DiMoP3D framework, which utilizes 3D point clouds and historical motion to achieve diverse and physically consistent human motion prediction.
Harmonizing Visual Text Comprehension and Generation
Zhen Zhao (East China Normal University), Yuan Xie (East China Normal University)
GenerationTransformerLarge Language ModelVision Language ModelDiffusion modelImageTextMultimodality
🎯 What it does: Developed the TextHarmony unified multimodal generation model, which balances visual text understanding and generation, and introduces Slide-LoRA to achieve modality consistency.
Harnessing Multiple Correlated Networks for Exact Community Recovery
Miklos Z. Racz, Jifan Zhang (Northwestern University)
Graph Neural NetworkGraph
🎯 What it does: This paper studies the precise information-theoretic thresholds for multi-graph collaborative community recovery and graph matching in the context of a multi-related stochastic block model (SBM), proposing a complete threshold criterion for achieving exact community recovery on an arbitrary constant number of related graphs.
Harnessing small projectors and multiple views for efficient vision pretraining
Arna Ghosh (Mila Quebec AI Institute), Blake Aaron Richards
Computational EfficiencyRepresentation LearningConvolutional Neural NetworkContrastive LearningImage
🎯 What it does: This study investigates the efficiency and sample efficiency of self-supervised visual pre-training, proposing a theoretical foundation and practical recommendations for low-dimensional projection heads and multi-view augmentation.
HAWK: Learning to Understand Open-World Video Anomalies
Jiaqi Tang (Hong Kong University of Science and Technology), Ying-Cong Chen (Hong Kong University of Science and Technology)
Anomaly DetectionTransformerVision Language ModelOptical FlowVideoTextMultimodality
🎯 What it does: A video-language model named HAWK is proposed, specifically designed to understand and describe abnormal behaviors in videos, and supports interactive question answering.
HC-GAE: The Hierarchical Cluster-based Graph Auto-Encoder for Graph Representation Learning
Lu Bai (Beijing Normal University), Edwin Hancock (University of York)
ClassificationRepresentation LearningGraph Neural NetworkAuto EncoderGraph
🎯 What it does: This paper proposes a Hierarchical Clustering Graph Autoencoder (HC-GAE), which decomposes the graph into subgraphs through hard node assignment and compresses them using local convolution, then reconstructs them through soft node assignment to obtain representations that can be used for both node classification and graph classification.
HDR-GS: Efficient High Dynamic Range Novel View Synthesis at 1000x Speed via Gaussian Splatting
Yuanhao Cai (Johns Hopkins University), Alan Yuille (Johns Hopkins University)
GenerationData SynthesisComputational EfficiencyGaussian SplattingSimultaneous Localization and MappingImagePoint Cloud
🎯 What it does: A new perspective synthesis framework HDR-GS based on high dynamic range Gaussian splatting has been developed.
HEALNet: Multimodal Fusion for Heterogeneous Biomedical Data
Konstantin Hemker (University of Cambridge), Mateja Jamnik (University of Cambridge)
ClassificationExplainability and InterpretabilityTransformerMultimodalityBiomedical Data
🎯 What it does: This paper proposes and implements HEALNet, a hybrid early fusion attention network that integrates multimodal (image, tabular, graph) data for survival analysis of multiple cancers.
Heavy-Tailed Class Imbalance and Why Adam Outperforms Gradient Descent on Language Models
Frederik Kunstner (University of British Columbia), Alberto Bietti (Flatiron Institute)
OptimizationTransformerLarge Language ModelImageText
🎯 What it does: This paper demonstrates through experimental and theoretical analysis that the heavy-tailed class imbalance encountered during language model training leads to slow progress of stochastic gradient descent (SGD) on low-frequency classes, while Adam and sign descent optimization methods are insensitive to this issue, thereby explaining the phenomenon of Adam significantly outperforming SGD on large language models.
HENASY: Learning to Assemble Scene-Entities for Interpretable Egocentric Video-Language Model
Khoa Vo, Ngan Hoang Le
RecognitionRetrievalExplainability and InterpretabilityTransformerVision Language ModelContrastive LearningVideoText
🎯 What it does: The HENASY framework is designed to provide interpretable entity-level representations of egocentric videos using a hierarchical entity assembly approach, and it integrates global and entity information through an entity-aware decoder.
HEPrune: Fast Private Training of Deep Neural Networks With Encrypted Data Pruning
Yancheng Zhang (University of Central Florida), Qian Lou (University of Central Florida)
Federated LearningSafty and PrivacyComputational EfficiencyConvolutional Neural NetworkSupervised Fine-TuningImage
🎯 What it does: This paper presents HEPrune, a framework for implementing data pruning in a Fully Homomorphic Encryption (FHE) environment, making encrypted data training more efficient.
Heterogeneity-Guided Client Sampling: Towards Fast and Efficient Non-IID Federated Learning
Huancheng Chen (University of Texas at Austin), Haris Vikalo (University of Texas at Austin)
Federated LearningImageText
🎯 What it does: This paper proposes HiCS-FL, a hierarchical clustering sampling method based on output layer bias gradient estimation of data heterogeneity, designed for efficiently selecting clients in communication-constrained non-IID Federated Learning scenarios.
HGDL: Heterogeneous Graph Label Distribution Learning
Yufei Jin (Florida Atlantic University), Xingquan Zhu (Florida Atlantic University)
ClassificationRepresentation LearningGraph Neural NetworkTransformerGraph
🎯 What it does: This paper proposes a heterogeneous graph label distribution learning (HGDL) framework that can predict the probability distribution of target nodes across multiple categories, achieving a finer-grained characterization of node functions compared to single-label or multi-label classification.
HHD-GP: Incorporating Helmholtz-Hodge Decomposition into Gaussian Processes for Learning Dynamical Systems
Hao Xu (University of Hong Kong), Jia Pan (University of Hong Kong)
OptimizationExplainability and InterpretabilityComputational EfficiencyGaussian SplattingTime SeriesPhysics RelatedOrdinary Differential Equation
🎯 What it does: This paper proposes a Helmholtz-Hodge decomposition model based on Gaussian processes (HHD-GP), which can simultaneously learn the curl-free and div-free components of dynamic systems, and achieve interpretability and high predictive performance by imposing symmetry constraints on these components.
HiCo: Hierarchical Controllable Diffusion Model for Layout-to-image Generation
Bocheng, Yuhui Yin (360 AI Research)
GenerationData SynthesisDiffusion modelImage
🎯 What it does: This paper presents HiCo—a diffusion model based on a multi-branch, hierarchical control approach for generating high-quality images from layouts and text.
HiCoM: Hierarchical Coherent Motion for Dynamic Streamable Scenes with 3D Gaussian Splatting
Qiankun Gao (Peking University), Jian Zhang (Peking University)
CompressionComputational EfficiencyGaussian SplattingVideo
🎯 What it does: The HiCoM framework is proposed to achieve online reconstruction of multi-view streaming dynamic scenes, significantly improving training speed, rendering performance, and storage compression.
Hierarchical and Density-based Causal Clustering
Kwangho Kim (Korea University), Edward Kennedy
Tabular
🎯 What it does: This paper proposes and implements a causal clustering method based on hierarchical clustering and density clustering, and provides its theoretical convergence rate.
Hierarchical Federated Learning with Multi-Timescale Gradient Correction
Wenzhi Fang (Purdue University), Christopher Brinton
Federated LearningConvolutional Neural NetworkImage
🎯 What it does: A multi-time scale gradient correction (MTGC) algorithm is designed for model drift correction in multi-level federated learning, addressing client and group-level model drift issues caused by multi-layer non-IID.
Hierarchical Hybrid Sliced Wasserstein: A Scalable Metric for Heterogeneous Joint Distributions
Khai Nguyen (University of Texas at Austin), Nhat Ho (University of Texas at Austin)
OptimizationRepresentation LearningAuto EncoderMesh
🎯 What it does: This paper proposes a new distance metric called Hierarchical Hybrid Sliced Wasserstein (H2SW), specifically designed for comparing heterogeneous joint distributions supported on different spaces, and experimentally validates its superiority in 3D mesh deformation, 3D mesh autoencoder training, and cross-dataset comparisons.
Hierarchical Object-Aware Dual-Level Contrastive Learning for Domain Generalized Stereo Matching
Yikun Miao (Beijing Institute of Technology), Thambipillai Srikanthan (Nanyang Technological University)
Depth EstimationDomain AdaptationContrastive LearningImage
🎯 What it does: This paper proposes a Hierarchical Object-aware Dual Contrastive Learning (HODC) framework to enhance the generalization performance of end-to-end stereo matching networks when transferring from synthetic domains to real domains.
Hierarchical Programmatic Option Framework
Yu-An Lin (National Taiwan University), Shao-Hua Sun (National Taiwan University)
Reinforcement LearningSequential
🎯 What it does: A hierarchical procedural option framework (HIPO) is proposed, where interpretable procedures serve as low-level options, and high-level policies are learned to address long-term, repetitive reinforcement learning tasks.
Hierarchical Selective Classification
Shani Goren (Technion), Ran El-Yaniv (Technion)
ClassificationKnowledge DistillationImage
🎯 What it does: This paper proposes Hierarchical Selective Classification (HSC), which allows the model to revert to higher-level nodes in the hierarchy when uncertain, thereby reducing risk without completely abandoning the prediction.
Hierarchical Uncertainty Exploration via Feedforward Posterior Trees
Elias Nehme (Technion Israel Institute of Technology), Tomer Michaeli (Technion Israel Institute of Technology)
Image TranslationRestorationSegmentationComputational EfficiencyConvolutional Neural NetworkImageMultimodality
🎯 What it does: An algorithm for posterior distribution hierarchical tree quantization (Posterior Trees) is proposed, which can output results after a single forward inference for visualization and uncertainty analysis of multimodal image inverse problems.
Hierarchical Visual Feature Aggregation for OCR-Free Document Understanding
Jaeyoo Park (Seoul National University), Bohyung Han (Seoul National University)
RecognitionSegmentationTransformerLarge Language ModelSupervised Fine-TuningImageMultimodality
🎯 What it does: A framework for OCR-free document understanding is proposed, utilizing multi-scale visual features and integrating a Hierarchical Visual Feature Aggregation (HVFA) module with a relative text position prediction task for instruction fine-tuning.
Hierarchy-Agnostic Unsupervised Segmentation: Parsing Semantic Image Structure
Simone Rossetti (Sapienza University of Rome), fiora pirri
Object DetectionSegmentationContrastive LearningImage
🎯 What it does: This paper proposes a completely unsupervised hierarchical structure insensitive image semantic segmentation method, which recursively partitions images into semantic regions and constructs a tree structure to achieve pixel-level semantic parsing.
High Rank Path Development: an approach to learning the filtration of stochastic processes
Jiajie Tao (University College London), Chong Liu (ShanghaiTech University)
GenerationData SynthesisGenerative Adversarial NetworkTime SeriesSequentialFinance Related
🎯 What it does: This paper proposes a high-order path development characteristic function (HRPCF) and defines a corresponding distance (HRPCFD) to measure extended weak convergence. It constructs an HRPCF-GAN for conditional time series generation and subsequently validates its effectiveness through hypothesis testing and generative model experiments.
High-dimensional (Group) Adversarial Training in Linear Regression
Yiling Xie (Georgia Institute of Technology), Xiaoming Huo (Georgia Institute of Technology)
OptimizationAdversarial AttackTabular
🎯 What it does: This paper conducts a non-asymptotic consistency analysis of ℓ∞ adversarial training for high-dimensional linear regression models, proving that its prediction error can achieve the minimal attainable rate for sparse parameter classes, and that group adversarial training can further reduce the error when a group sparse structure is present.
High-probability complexity bounds for stochastic non-convex minimax optimization
Yassine Laguel (Universite Cote d'Azur), Mert Gurbuzbalaban
OptimizationTabularStochastic Differential Equation
🎯 What it does: The paper presents a high-probability complexity upper bound for non-convex-convex minimax problems satisfying the PL condition using the sm-AGDA algorithm.
High-Resolution Image Harmonization with Adaptive-Interval Color Transformation
Quanling Meng (Harbin Institute of Technology), Liqiang Nie (Harbin Institute of Technology)
Image HarmonizationTransformerImage
🎯 What it does: This paper studies the harmonization problem of high-resolution images and proposes the Adaptive-Interval Color Transformation (AICT) method, which achieves pixel-level color transformation and adaptively adjusts the sampling interval through low-resolution predicted position-dependent 3D LUT.
Higher-Order Causal Message Passing for Experimentation with Complex Interference
Mohsen Bayati (Stanford Graduate School of Business), Ruoxuan Xiong (Emory University)
GraphTime Series
🎯 What it does: The Higher-Order Causal Message-Passing (HO-CMP) estimator is proposed to estimate the total treatment effect (TTE) by observing multi-time point data under unknown network interference.
Higher-Rank Irreducible Cartesian Tensors for Equivariant Message Passing
Viktor Zaverkin (NEC Laboratories Europe), Mathias Niepert (University of Stuttgart)
GraphBenchmarkPhysics Related
🎯 What it does: A covariant information transfer neural network based on higher-order irreducible Cartesian tensors is proposed and implemented for learning atomic-level potential and forces.
HippoRAG: Neurobiologically Inspired Long-Term Memory for Large Language Models
Bernal Jimenez Gutierrez, Yu Su (Ohio State University)
RetrievalTransformerLarge Language ModelTextRetrieval-Augmented Generation
🎯 What it does: A retrieval framework called HippoRAG is proposed based on the hippocampal memory index theory, which constructs a knowledge graph using LLM and achieves one-time multi-hop retrieval through Personalized PageRank, addressing the limitations of traditional RAG in knowledge integration.
Historical Test-time Prompt Tuning for Vision Foundation Models
Jingyi Zhang (Nanyang Technological University), Shijian Lu (University of Chinese Academy of Sciences)
Object DetectionSegmentationDomain AdaptationPrompt EngineeringImage
🎯 What it does: A historical test-time prompt tuning (HisTPT) framework is proposed for online learning and optimizing visual foundation model prompts in unlabeled test streams.
HLM-Cite: Hybrid Language Model Workflow for Text-based Scientific Citation Prediction
Qianyue Hao (Tsinghua University), Yong Li (Tsinghua University)
TransformerLarge Language ModelSupervised Fine-TuningTextRetrieval-Augmented GenerationChain-of-Thought
🎯 What it does: Proposes a scientific literature citation prediction work based on a hybrid language model, defines the core citation concept, and implements a two-stage retrieval + LLM reasoning pipeline.
HOI-Swap: Swapping Objects in Videos with Hand-Object Interaction Awareness
Zihui Xue (University of Texas at Austin), Kristen Grauman (University of Texas at Austin)
Object DetectionObject TrackingGenerationData SynthesisDiffusion modelOptical FlowVideo
🎯 What it does: We propose HOI-Swap, a two-stage diffusion model-based video object replacement framework that can accurately replace objects in hand-object interaction scenes while maintaining natural hand-object interactions.
Hollowed Net for On-Device Personalization of Text-to-Image Diffusion Models
Wonguk Cho (Qualcomm AI Research), Fatih Porikli (Qualcomm AI Research)
GenerationData SynthesisSupervised Fine-TuningDiffusion modelImageText
🎯 What it does: Implementing personalized learning of text-to-image diffusion models on resource-constrained devices, using Hollowed Net for low-memory fine-tuning.
Homology Consistency Constrained Efficient Tuning for Vision-Language Models
Huatian Zhang (University of Science and Technology of China), Zhendong Mao (University of Science and Technology of China)
ClassificationDomain AdaptationComputational EfficiencyTransformerVision Language ModelImageTextMultimodality
🎯 What it does: This paper proposes a homotopy consistency constraint based on persistent homology for efficiently transferring large visual-language models in low-data environments, maintaining the structural consistency of the latent manifolds of images and text.
HonestLLM: Toward an Honest and Helpful Large Language Model
Chujie Gao (Mohamed Bin Zayed University of Artificial Intelligence), Xiangliang Zhang (University of Notre Dame)
TransformerLarge Language ModelSupervised Fine-TuningPrompt EngineeringTextBenchmark
🎯 What it does: A framework is proposed to enhance the usefulness of large language models (LLMs) while maintaining honesty. It first defines the principle of honesty and constructs a specialized evaluation dataset called HONESET, and then designs two methods—training-free (curiosity-driven prompts) and two-stage fine-tuning—to improve the model's honesty and helpfulness.
Honor Among Bandits: No-Regret Learning for Online Fair Division
Ariel D. Procaccia (Harvard University), Shirley Zhang (Harvard University)
OptimizationReinforcement Learning
🎯 What it does: This paper studies the problem of online fair allocation of indivisible goods. When the players' valuations of the goods are unknown, a learning algorithm based on multi-armed bandits is designed to maximize social welfare while ensuring expected fairness.
HOPE: Shape Matching Via Aligning Different K-hop Neighbourhoods
Barakeel Fanseu Kamhoua (Hong Kong University of Science and Technology), Huamin Qu (Hong Kong University of Science and Technology)
Graph Neural NetworkMesh
🎯 What it does: The HOPE framework is proposed, which uses local mapping distortion metrics to filter out unmatched vertices and iteratively refines shape matching through multi-scale k-hop neighborhood consistency.
HORSE: Hierarchical Representation for Large-Scale Neural Subset Selection
Binghui Xie (Chinese University of Hong Kong), James Cheng (Chinese University of Hong Kong)
Recommendation SystemDrug DiscoveryTransformerTabular
🎯 What it does: A new attention-based hierarchical subset selection network called HORSE is proposed for learning subset functions and performing subset selection on large-scale sets.
How Control Information Influences Multilingual Text Image Generation and Editing?
Boqiang Zhang (University of Science and Technology of China), Hongtao Xie (University of Science and Technology of China)
GenerationData SynthesisDiffusion modelImageText
🎯 What it does: In the task of visual text generation and editing, a multi-stage framework called TextGen based on ControlNet is proposed, which enhances the feature extraction of control information using Fourier-enhanced convolution and optimizes the output of control features through a frequency balancing mechanism, ultimately achieving a unified approach for multi-language text generation and editing.
How Diffusion Models Learn to Factorize and Compose
Qiyao Liang (Massachusetts Institute of Technology), Ila R Fiete (Massachusetts Institute of Technology)
GenerationData SynthesisRepresentation LearningDiffusion modelImagePhysics Related
🎯 What it does: By training a conditional DDPM on a 32×32 2D Gaussian image dataset, a systematic study was conducted on whether diffusion models can learn factored representations and their performance in combinatorial reasoning and interpolation.
How Do Large Language Models Acquire Factual Knowledge During Pretraining?
Hoyeon Chang (Korea Advanced Institute of Science and Technology), Minjoon Seo (Korea Advanced Institute of Science and Technology)
TransformerLarge Language ModelText
🎯 What it does: This study investigates how large language models acquire factual knowledge during the pre-training phase, constructs a FICTIONAL KNOWLEDGE dataset, and measures the processes of memory acquisition and forgetting through the insertion of training instances.
How do Large Language Models Handle Multilingualism?
Yiran Zhao (National University of Singapore), Lidong Bing (DAMO Academy, Alibaba Group)
GenerationData SynthesisTransformerLarge Language ModelSupervised Fine-TuningText
🎯 What it does: This paper studies the processing mechanisms of large language models in multilingual tasks, proposes a three-stage multilingual workflow (MWork), and designs an unsupervised method for detecting parallel language-specific neurons (PLND) to validate and utilize these neurons for multilingual enhancement with limited data.
How does Architecture Influence the Base Capabilities of Pre-trained Language Models? A Case Study Based on FFN-Wider and MoE Transformers
Xin Lu (Harbin Institute of Technology), Dongliang Xu (Du Xiaoman Science Technology Co., Ltd.)
TransformerLarge Language ModelMixture of ExpertsText
🎯 What it does: This paper studies the impact of FFN-Wider Transformer and MoE Transformer on the baseline capabilities of pre-trained language models, and proposes a Combination Enhanced Architecture (CEA) to restore and enhance baseline capabilities.
How Does Black-Box Impact the Learning Guarantee of Stochastic Compositional Optimization?
Jun Chen (Huazhong Agricultural University), Bin Gu (Jilin University)
OptimizationFederated Learning
🎯 What it does: This paper conducts a theoretical analysis of the generalization and optimization guarantees of Stochastic Combination Optimization (SCO) algorithms in a black-box (zeroth-order) setting, constructs a more compact stability framework, and applies the analysis results to Vertical Federated Learning (VFL) algorithms, providing the first generalization and convergence bounds based on algorithm stability.
How does Gradient Descent Learn Features --- A Local Analysis for Regularized Two-Layer Neural Networks
Mo Zhou (University of Washington), Rong Ge (Duke University)
OptimizationKnowledge Distillation
🎯 What it does: In the teacher-student framework, the feature learning mechanism of gradient descent in two-layer networks is studied, and it is proven that under sufficiently wide conditions and with weight decay, gradient descent can converge to the target network in polynomial time and align the student neurons with the teacher direction.
How does Inverse RL Scale to Large State Spaces? A Provably Efficient Approach
Filippo Lazzati (Politecnico di Milano), Alberto Maria Metelli (Politecnico di Milano)
Reinforcement Learning from Human FeedbackReinforcement Learning
🎯 What it does: This paper studies the scalability of inverse reinforcement learning in large state spaces, proving that traditional feasible reward sets cannot be efficiently learned in Linear MDPs, and proposes a reward compatibility framework along with the design of the sample-efficient CATY-IRL algorithm.
How Does Message Passing Improve Collaborative Filtering?
Mingxuan Ju (Snap Inc.), Tong Zhao (Snap Inc.)
Recommendation SystemGraph Neural NetworkContrastive LearningGraph
🎯 What it does: This paper systematically studies the mechanism by which message passing enhances the performance of collaborative filtering (CF) and proposes a testing enhancement framework TAG-CF that aggregates messages only once during inference.
How does PDE order affect the convergence of PINNs?
Chang hoon Song, Myungjoo Kang (Seoul National University)
Physics RelatedOrdinary Differential Equation
🎯 What it does: This paper analyzes the convergence of the gradient flow (GF) of Physics-Informed Neural Networks (PINN) from a theoretical perspective, revealing the impact of the order of partial differential equations (PDE), dimensions, and the power of ReLU activation on convergence.
How Does Variance Shape the Regret in Contextual Bandits?
Zeyu Jia (Massachusetts Institute of Technology), Chen-Yu Wei (University of Virginia)
Reinforcement Learning
🎯 What it does: This paper studies how reward variance affects regret in achievable contextual multi-armed bandits, providing variance-dependent upper and lower bounds for weak and strong opponents, and proposes an efficient algorithm based on the eluder dimension.
How Far Can Transformers Reason? The Globality Barrier and Inductive Scratchpad
Emmanuel Abbe (Apple), Omid Saremi (Apple)
TransformerSequential
🎯 What it does: This paper explores the 'global reasoning barrier' of Transformers during zero-shot learning from both theoretical and experimental perspectives. It introduces a globality metric to measure the globality of data distribution and proves that tasks with high globality are difficult for Transformers to learn effectively. Subsequently, it introduces the scratchpad technique, particularly the inductive scratchpad, demonstrating its significant effect on improving OOV/length generalization.
How JEPA Avoids Noisy Features: The Implicit Bias of Deep Linear Self Distillation Networks
Etai Littwin (Apple), Joshua M. Susskind
Knowledge DistillationRepresentation LearningAuto EncoderTime SeriesOrdinary Differential Equation
🎯 What it does: This paper theoretically and experimentally compares two self-supervised learning paradigms—JEPA (Joint Embedding Prediction) and MAE (Masked Autoencoder)—through the analyzable dynamics of deep linear networks, revealing their implicit biases in feature learning order.
How many classifiers do we need?
Hyunsuk Kim (University of California), Michael W. Mahoney (University of California)
ClassificationConvolutional Neural NetworkImage
🎯 What it does: This paper studies the error rate of multi-class voting ensembles, proposes and analyzes a new polarization indicator, provides its upper bound, and discusses its relationship with voting error and model disagreement.
How Molecules Impact Cells: Unlocking Contrastive PhenoMolecular Retrieval
Philip Fradkin (University of Toronto), Dominique Beaini (Valence Labs)
RetrievalDrug DiscoveryGraph Neural NetworkContrastive LearningImageMultimodality
🎯 What it does: The MolPhenix model was constructed to achieve molecular-cell morphology retrieval based on contrastive learning, enabling the retrieval of corresponding active molecules from cell morphology images under zero-shot conditions.
How Sparse Can We Prune A Deep Network: A Fundamental Limit Perspective
Qiaozhe Zhang (Huazhong University of Science and Technology), Yingzhuang Liu (Huazhong University of Science and Technology)
OptimizationConvolutional Neural NetworkImage
🎯 What it does: The theoretical limits of deep network pruning were studied, providing lower and upper bounds for the pruning ratio, which were experimentally verified to be nearly consistent;
How to Boost Any Loss Function
Richard Nock (Google Research), Yishay Mansour (Tel Aviv University)
Optimization
🎯 What it does: A new boosting algorithm called SECBOOST is proposed, which can optimize almost any loss function without requiring the loss function to be differentiable, convex, or continuous.
How to Continually Adapt Text-to-Image Diffusion Models for Flexible Customization?
Jiahua Dong (Mohamed bin Zayed University of Artificial Intelligence), Fahad Khan
GenerationData SynthesisDiffusion modelImageText
🎯 What it does: A text-to-image diffusion model (CIDM) capable of continuous learning of multiple custom concepts is proposed, addressing the challenges of catastrophic forgetting and concept neglect in the context of Concept Incremental Customization (CIFC).
How to Solve Contextual Goal-Oriented Problems with Offline Datasets?
Ying Fan (University of Wisconsin-Madison), Ching-An Cheng (Microsoft Research)
OptimizationReinforcement LearningSequential
🎯 What it does: This paper proposes a Contextual goal-Oriented Data Augmentation (CODA) method to address the contextual goal-oriented problem in offline data using known trajectories and context-goal pairs.
How to Use Diffusion Priors under Sparse Views?
Qisen Wang (Beihang University), Jia Li (Beihang University)
GenerationData SynthesisOptimizationDiffusion modelScore-based ModelGaussian SplattingImage
🎯 What it does: This paper proposes a correction of the score distillation sampling (SDS) of diffusion models using a sparse perspective linear prior (inline prior) to achieve new view synthesis under a sparse perspective, and combines this method with 3D Gaussian Splatting (3DGS);
How Transformers Utilize Multi-Head Attention in In-Context Learning? A Case Study on Sparse Linear Regression
Xingwu Chen (Hong Kong University), Difan Zou (Hong Kong University)
OptimizationTransformerTabular
🎯 What it does: This paper studies the mechanism of multi-head attention in the first layer of the Transformer for preprocessing, and single-head attention in subsequent layers for optimization, conducting experimental and theoretical analysis in the context of sparse linear regression learning.
Human Expertise in Algorithmic Prediction
Rohan Alur (Massachusetts Institute of Technology), Devavrat Shah (Massachusetts Institute of Technology)
ClassificationRecognitionConvolutional Neural NetworkImageBiomedical Data
🎯 What it does: A framework based on algorithm indistinguishable subsets is proposed, utilizing expert feedback to improve predictions, and a method is provided to test whether expert information can be captured by the algorithm.
Human-3Diffusion: Realistic Avatar Creation via Explicit 3D Consistent Diffusion Models
Yuxuan Xue, Gerard Pons-Moll
GenerationData SynthesisDiffusion modelGaussian SplattingImage
🎯 What it does: Using a single RGB image, we construct a consistent and 3D unified realistic human avatar in real-time by combining a 2D multi-view diffusion prior with a 3D Gaussian splats generator.
Human-Object Interaction Detection Collaborated with Large Relation-driven Diffusion Models
Liulei Li (University of Technology Sydney), Yi Yang (Zhejiang University)
Object DetectionDiffusion modelContrastive LearningImage
🎯 What it does: This paper proposes DIFFUSIONHOI, a human-object interaction detection framework based on diffusion models.
Humanoid Locomotion as Next Token Prediction
Ilija Radosavovic (University of California Berkeley), Jitendra Malik (University of California Berkeley)
Robotic IntelligenceTransformerReinforcement LearningVideoMultimodality
🎯 What it does: Transforming whole-body robot walking control into a next-token prediction task, training a causal Transformer to predict sensor-action sequences, and achieving zero-shot deployment in real environments.
HumanSplat: Generalizable Single-Image Human Gaussian Splatting with Structure Priors
Panwang Pan (ByteDance), Yebin Liu (Xiamen University)
RestorationGenerationTransformerDiffusion modelGaussian SplattingImage
🎯 What it does: A 3D high-fidelity human body reconstruction method called HumanSplat is proposed, which can directly generate a 3D Gaussian Splatting representation from a single input image and achieve high-quality new perspective rendering.
HumanVLA: Towards Vision-Language Directed Object Rearrangement by Physical Humanoid
Xinyu Xu (Shanghai Jiao Tong University), Cewu Lu (Shanghai Jiao Tong University)
Knowledge DistillationRobotic IntelligenceConvolutional Neural NetworkTransformerReinforcement LearningVision-Language-Action ModelImageTextMultimodality
🎯 What it does: A visual and language-based physical humanoid robot object rearrangement system, HumanVLA, has been developed, utilizing a teacher-student framework to distill the teacher policy obtained from reinforcement learning into a student model that can execute tasks solely based on first-person vision and natural language instructions.
HuRef: HUman-REadable Fingerprint for Large Language Models
Boyi Zeng (Shanghai Jiao Tong University), Zhouhan Lin (Shanghai Jiao Tong University)
RecognitionGenerationData SynthesisSafty and PrivacyTransformerLarge Language ModelGenerative Adversarial NetworkContrastive LearningText
🎯 What it does: A human-readable fingerprint (HuRef) is proposed for large language models (LLM) by extracting invariants from model parameters and mapping them to natural images, publicly disclosing the fingerprint without leaking parameters, and ensuring the authenticity of the fingerprint through zero-knowledge proofs.
Hybrid Generative AI for De Novo Design of Co-Crystals with Enhanced Tabletability
Nina Gubina (ITMO University), Vladimir Vinogradov (ITMO University)
OptimizationDrug DiscoveryGraph Neural NetworkTransformerAuto EncoderGenerative Adversarial NetworkTabular
🎯 What it does: GEMCODE is proposed and implemented, a collaborative generation pipeline that integrates generative models and evolutionary optimization for the design of co-crystal ligands with target formulation properties (such as suppressibility) based on drug molecules;
Hybrid Mamba for Few-Shot Segmentation
Qianxiong Xu (Nanyang Technological University), Rui Zhao (SenseTime Research)
SegmentationConvolutional Neural NetworkImage
🎯 What it does: Using the linear complexity Mamba model to achieve cross-sequence dependency fusion in few-shot segmentation.
Hybrid Reinforcement Learning Breaks Sample Size Barriers In Linear MDPs
Kevin Tan (University of Pennsylvania), Yuting Wei (University of Pennsylvania)
Reinforcement Learning
🎯 What it does: This paper proposes two hybrid reinforcement learning algorithms, namely Online-to-Offline (RAPPEL) and Offline-to-Online (HYRULE), and provides theoretical sample complexity and optimal bounds on return/error in linear MDP environments.
Hybrid Top-Down Global Causal Discovery with Local Search for Linear and Nonlinear Additive Noise Models
Sujai Hiremath (Cornell University), Kyra Gan (Cornell University)
Tabular
🎯 What it does: This paper proposes a hybrid causal discovery framework that combines functional causal models with constraint search. It provides hierarchical topological sorting algorithms (LHTS, NHTS) for linear and nonlinear additive noise models, as well as a non-parametric edge pruning algorithm (ED) based on local conditional sets, achieving precise reconstruction of global causal graphs.
HYDRA-FL: Hybrid Knowledge Distillation for Robust and Accurate Federated Learning
Momin Ahmad Khan (University of Massachusetts Amherst), Fatima M. Anwar
Federated LearningKnowledge DistillationAdversarial AttackImage
🎯 What it does: In federated learning, this paper investigates and addresses the issue of knowledge distillation methods amplifying the impact of model poisoning attacks, and proposes the HYDRA-FL hybrid distillation technique, which mitigates attack amplification by utilizing shallow distillation and reducing the KD weight of the final layer, while maintaining or improving performance in non-attack scenarios.
Hydra: Bidirectional State Space Models Through Generalized Matrix Mixers
Sukjun Hwang (Carnegie Mellon University), Albert Gu (Carnegie Mellon University)
TransformerImageText
🎯 What it does: A framework that unifies sequence mixers into matrix mixers is proposed, and based on this, a new bidirectional state space model called Hydra is designed as a scalable, data-dependent, and efficient sequence mixer.
HYDRA: Model Factorization Framework for Black-Box LLM Personalization
Yuchen Zhuang (Georgia Institute of Technology), Bo Dai (Georgia Institute of Technology)
GenerationRetrievalRecommendation SystemTransformerLarge Language ModelTextBenchmarkRetrieval-Augmented Generation
🎯 What it does: The HYDRA framework is proposed, which combines retrieval-re-ranking with a black-box LLM adapter to achieve personalized generation of user behavior history without accessing model parameters.
HydraLoRA: An Asymmetric LoRA Architecture for Efficient Fine-Tuning
Chunlin Tian (University of Macau), Cheng-zhong Xu
TransformerLarge Language ModelSupervised Fine-TuningMixture of ExpertsText
🎯 What it does: This paper proposes HydraLoRA, an asynchronous LoRA structure that automatically identifies and learns the intrinsic substructures of data by sharing matrix A and multiple matrix B, achieving efficient fine-tuning without domain knowledge.
HydraViT: Stacking Heads for a Scalable ViT
Janek Haberer (Kiel University), Olaf Landsiedel (Kiel University)
ClassificationRecognitionComputational EfficiencyTransformerMixture of ExpertsImage
🎯 What it does: A scalable Vision Transformer (HydraViT) is proposed, which generates multiple sub-networks by randomly cropping attention heads and embedding dimensions during training.
Hyper-opinion Evidential Deep Learning for Out-of-Distribution Detection
Jingen Qu (Tongji University), Qiguang Huang (Tongji University)
ClassificationAnomaly DetectionConvolutional Neural NetworkImage
🎯 What it does: This paper proposes Hyper-opinion Evidential Deep Learning (HEDL), which extends traditional Evidential Deep Learning (EDL) based on evidence theory. It utilizes hyper-opinion to simultaneously extract sharp and fuzzy evidence and transforms it into polynomial opinions through projection, thereby improving the accuracy of out-of-distribution (OOD) detection and classification.
Hyper-SD: Trajectory Segmented Consistency Model for Efficient Image Synthesis
Yuxi Ren (ByteDance), Xuefeng Xiao (ByteDance)
GenerationData SynthesisKnowledge DistillationDiffusion modelScore-based ModelImageOrdinary Differential Equation
🎯 What it does: Proposes the Hyper-SD framework, which combines trajectory segment consistency distillation, human feedback learning, and score distillation to achieve high-quality image generation under 1-8 step inference, and provides a unified LoRA plugin.
Hyperbolic Embeddings of Supervised Models
Richard Nock (Google Research), Manfred K Warmuth
ClassificationTabular
🎯 What it does: This paper proposes a complete scheme for embedding supervised models into hyperbolic geometry, primarily focusing on decision trees and their ensemble models.
HyperLogic: Enhancing Diversity and Accuracy in Rule Learning with HyperNets
Yang Yang (Chinese University of Hong Kong), Shuang Li (Chinese University of Hong Kong)
ClassificationExplainability and InterpretabilityMixture of ExpertsTabularFinance Related
🎯 What it does: Proposes the HyperLogic framework, which uses hypernetworks to generate weights for the main rule learning network, thereby enhancing the diversity and accuracy of rule learning while maintaining interpretability.
HyperPrism: An Adaptive Non-linear Aggregation Framework for Distributed Machine Learning over Non-IID Data and Time-varying Communication Links
Haizhou Du (Shanghai University of Electric Power), Linghe Kong (Shanghai Jiao Tong University)
OptimizationFederated LearningReinforcement LearningImage
🎯 What it does: The HyperPrism framework is designed to use adaptive nonlinear aggregation in distributed machine learning to address the model divergence issues caused by non-IID data and time-varying communication links.
Hypothesis Testing the Circuit Hypothesis in LLMs
Claudia Shi (Columbia University), David Blei
TransformerLarge Language ModelText
🎯 What it does: This paper proposes a formal hypothesis testing framework to evaluate whether the 'circuits' in large language models meet idealized properties, and conducts experimental validation on six known circuits.
HYSYNTH: Context-Free LLM Approximation for Guiding Program Synthesis
Shraddha Barke (University of California San Diego), Nadia Polikarpova (University of California San Diego)
OptimizationAI Code AssistantTransformerLarge Language ModelPrompt EngineeringText
🎯 What it does: This paper proposes a hybrid program synthesis method called HYSYNTH, which utilizes program samples generated by LLM to learn a PCFG and then guides a context-free pushdown search to solve program synthesis tasks across multiple domains.
I Don't Know: Explicit Modeling of Uncertainty with an [IDK] Token
Roi Cohen (Hasso Plattner Institute), Gerard de Melo (Hasso Plattner Institute)
GenerationTransformerLarge Language ModelText
🎯 What it does: A special [IDK] (I Don't Know) token is added to the vocabulary of large language models, and a new self-supervised objective (IDK-tuning) is adopted during the continuous pre-training phase, allowing the model to transfer probability mass to this token when uncertain, thereby clearly expressing uncertainty during generation and reducing hallucinations.
I2EBench: A Comprehensive Benchmark for Instruction-based Image Editing
Yiwei Ma (Xiamen University), Rongrong Ji (Xiamen University)
Image TranslationGenerationData SynthesisLarge Language ModelPrompt EngineeringImageTextBenchmark
🎯 What it does: I2EBench is proposed, which includes a comprehensive evaluation benchmark with over 2000 images and 4000 instructions for assessing instruction-based image editing models.
ID-to-3D: Expressive ID-guided 3D Heads via Score Distillation Sampling
Francesca Babiloni (Imperial College London), Stefanos Zafeiriou (Imperial College London)
GenerationData SynthesisTransformerDiffusion modelScore-based ModelImage
🎯 What it does: Using a small number of 'in-the-wild' facial images and text prompts, we generate high-quality 3D facial heads (geometry + texture) that are identity-consistent and capable of expressing various emotions through Score Distillation Sampling combined with lightweight fine-tuning of a 2D diffusion model.
Identifiability Analysis of Linear ODE Systems with Hidden Confounders
Yuanyuan Wang (University of Melbourne), Mingming Gong (University of Melbourne)
Time SeriesOrdinary Differential Equation
🎯 What it does: This paper conducts an identifiability analysis of linear ODE systems with hidden confounding factors, proposing identifiability conditions under different hidden variable structures.
Identifiability Guarantees for Causal Disentanglement from Purely Observational Data
Ryan Welch (Massachusetts Institute of Technology), Caroline Uhler (Massachusetts Institute of Technology)
OptimizationScore-based ModelGraph
🎯 What it does: This paper proposes a theory and algorithm for causal identifiability using the zero variance property of the score function under a model with only observational data, linear mixing, and nonlinear Gaussian noise.
Identifiable Object-Centric Representation Learning via Probabilistic Slot Attention
Avinash Kori (Imperial College London), Fabio De Sousa Ribeiro (Imperial College London)
Object DetectionRepresentation LearningImage
🎯 What it does: Proposes the Probability Slot Attention (PSA) algorithm, introducing an aggregated Gaussian mixture prior in the object slot space to achieve recognizability of object center representation in an unsupervised manner.
Identifiable Shared Component Analysis of Unpaired Multimodal Mixtures
Subash Timilsina (Oregon State University), Xiao Fu (Oregon State University)
RetrievalDomain AdaptationGenerative Adversarial NetworkMultimodalityBiomedical Data
🎯 What it does: The study investigates how to learn shared components through distribution matching in different modality unaligned linear mixed models.
Identification and Estimation of the Bi-Directional MR with Some Invalid Instruments
Feng Xie (Beijing Technology and Business University), Zhi Geng (Beijing Technology and Business University)
TabularBiomedical Data
🎯 What it does: In pure observational data, the identifiability of the bi-directional Mendelian randomization (MR) model was studied, and an algorithm based on pseudo-residuals, PReBiM, was proposed to automatically discover valid instrumental variables (IVs) from candidate gene variants, determine causal direction, and estimate bidirectional causal effects.
Identification of Analytic Nonlinear Dynamical Systems with Non-asymptotic Guarantees
Negin Musavi (University of Illinois Urbana-Champaign), Yingying Li (University of Illinois Urbana-Champaign)
Time SeriesPhysics RelatedStochastic Differential EquationOrdinary Differential Equation
🎯 What it does: This paper studies system identification of linear parameterized nonlinear systems with real analytic characteristic functions under non-active exploration conditions, providing the non-asymptotic convergence rates of LSE and SME, and verifying the convergence rates through pendulum and quadrotor simulation experiments consistent with the theory;
Identify Then Recommend: Towards Unsupervised Group Recommendation
Yue Liu (Ant Group National University of Singapore), Wenliang Zhong (Ant Group)
Recommendation SystemContrastive LearningTabular
🎯 What it does: A completely unsupervised group recommendation framework ITI (Identify Then Recommend) is proposed, which first automatically identifies user groups in the user embedding space and then performs group recommendation through self-supervised pretext tasks.
Identifying and Solving Conditional Image Leakage in Image-to-Video Diffusion Model
Min Zhao (Tsinghua University), Jun Zhu (Tsinghua University)
GenerationData SynthesisDiffusion modelImageVideoStochastic Differential Equation
🎯 What it does: This study addresses and solves the Conditional Image Leakage (CIL) problem in Image-to-Video Diffusion Models (I2V-DM), proposing an early start time during the inference phase and optimized initialization noise (Analytic-Init), as well as a time-varying noise distribution (TimeNoise) during the training phase to reduce reliance on conditional images and enhance video motion performance.
Identifying Causal Effects Under Functional Dependencies
Yizuo Chen (University of California), Adnan Darwiche (University of California)
Graph Neural NetworkGraph
🎯 What it does: Introducing functional dependence (where variables are determined by parent variables but the specific function is unknown) in causal graphs to enhance the identifiability of causal effects, this paper elaborates on functional elimination and functional projection operations, and combines them with existing ID algorithms to construct decidable rules for functional identifiability.
Identifying Equivalent Training Dynamics
William T Redman, Igor Mezic (AIMdyn Inc.)
Convolutional Neural NetworkTransformerImage
🎯 What it does: This paper utilizes Koopman operator theory for data-driven isomorphism identification of deep neural network training dynamics.