NeurIPS 2024 Papers — Page 13
Conference on Neural Information Processing Systems · 4035 papers
Evaluation of Text-to-Video Generation Models: A Dynamics Perspective
Mingxiang Liao (University of Chinese Academy of Sciences), Xinyu Zhang (University of Adelaide)
GenerationData SynthesisLarge Language ModelOptical FlowVideoTextBenchmark
🎯 What it does: A dynamic dimension-based T2V evaluation protocol called DEVIL is proposed, and a benchmark containing multi-level dynamic text prompts is constructed.
Even Sparser Graph Transformers
Hamed Shirzad (University of British Columbia), Danica J. Sutherland (University of British Columbia)
Graph Neural NetworkTransformerGraph
🎯 What it does: A two-stage training method is proposed: first, a narrow network is trained to estimate attention scores, and then based on these scores, the graph is hierarchically sparsified and a wide network is trained, thereby improving the memory efficiency of the Graph Transformer.
Event-3DGS: Event-based 3D Reconstruction Using 3D Gaussian Splatting
Haiqian Han, Xiangyang Ji (Tsinghua University)
RestorationGenerationNeural Radiance FieldGaussian SplattingImagePoint Cloud
🎯 What it does: This paper proposes Event-3DGS, a method that directly processes event camera data for 3D reconstruction and novel view synthesis using 3D Gaussian splatting (3DGS).
Everyday Object Meets Vision-and-Language Navigation Agent via Backdoor
Keji He (Shandong University), Liang Wang (Pattern Recognition Institute of Automation, Chinese Academy of Sciences)
Adversarial AttackTransformerReinforcement LearningVision Language ModelImageTextMultimodality
🎯 What it does: A visual language navigation (VLN) backdoor attack method based on real object triggers is proposed, which can operate in normal environments but automatically stops execution when encountering specific object triggers.
Evidence of Learned Look-Ahead in a Chess-Playing Neural Network
Erik Jenner (University of California Berkeley), Stuart Russell (University of California Berkeley)
TransformerTabular
🎯 What it does: This study investigates whether the Leela Chess Zero neural network internally implements look-ahead search in actual chess positions, and verifies this through three pieces of causal and interpretability evidence.
Evidential Mixture Machines: Deciphering Multi-Label Correlations for Active Learning Sensitivity
Dayou Yu (Rochester Institute of Technology), Qi Yu (University of North Texas)
ClassificationData-Centric LearningTabular
🎯 What it does: This paper proposes and implements Evidential Mixture Machines (EMM), a multi-label active learning framework that combines a mixture Bernoulli model with deep evidence learning.
Evidential Stochastic Differential Equations for Time-Aware Sequential Recommendation
Krishna Prasad Neupane (Rochester Institute of Technology), Qi Yu (Rochester Institute of Technology)
Recommendation SystemSequentialStochastic Differential Equation
🎯 What it does: This paper proposes a stochastic differential equation model based on evidence learning (E-NSDE) to simultaneously capture the continuous evolution of user interests and the uncertainty of predictions in time-aware sequential recommendation.
EvolveDirector: Approaching Advanced Text-to-Image Generation with Large Vision-Language Models
Rui Zhao (National University of Singapore), Mike Zheng Shou
GenerationData SynthesisTransformerVision Language ModelDiffusion modelImageTextMultimodality
🎯 What it does: Utilizing large-scale visual language models (VLM) as 'directors', we collect data generated by advanced text-to-image models through public APIs to construct a dynamically iterable training set, training the baseline diffusion Transformer (DiT) model to approximate or even surpass the generative capabilities of various advanced text-to-image models.
Ex Uno Pluria: Insights on Ensembling in Low Precision Number Systems
Giung Nam (Korea Advanced Institute of Science and Technology), Juho Lee (Korea Advanced Institute of Science and Technology)
OptimizationComputational EfficiencyTransformerLarge Language ModelImage
🎯 What it does: An integrated method based on low-precision number systems, LPE-BSR, is proposed. By randomly rounding the weights of pre-trained models, it generates diverse model members and enhances inference performance and uncertainty estimation without additional training.
Exact Gradients for Stochastic Spiking Neural Networks Driven by Rough Signals
Christian Holberg (University of Copenhagen), Cristopher Salvi (Imperial College London)
Spiking Neural NetworkTime SeriesStochastic Differential Equation
🎯 What it does: This paper proposes a mathematical framework based on rough path theory for modeling and training stochastic spiking neural networks (SSNN) with noise, and provides the gradient recursion formula for event-driven stochastic differential equations (Event SDE);
Exact, Tractable Gauss-Newton Optimization in Deep Reversible Architectures Reveal Poor Generalization
Davide Buffelli (MediaTek Research), Alberto Bernacchia (MediaTek Research)
OptimizationTabular
🎯 What it does: This paper studies and implements a computable Gauss-Newton (GN) optimization in reversible deep networks (RevMLP), systematically evaluating the training speed and generalization performance of GN for the first time on large-scale practical networks.
Exactly Minimax-Optimal Locally Differentially Private Sampling
Hyun-Young Park (Korea Advanced Institute of Science and Technology), Si-Hyeon Lee (Korea Advanced Institute of Science and Technology)
OptimizationSafty and Privacy
🎯 What it does: This paper studies the problem of private sampling under local differential privacy, defines the privacy-utility trade-off (PUT), and proposes an optimal private sampling mechanism in the minimax sense.
Excluding the Irrelevant: Focusing Reinforcement Learning through Continuous Action Masking
Roland Stolz (Technical University of Munich), Matthias Althoff (Technical University of Munich)
Reinforcement Learning
🎯 What it does: A method for action masking in continuous action spaces is proposed and implemented, utilizing convex sets (polyhedra/zonotopes) to describe the effective action set related to states. Based on this, three masking strategies (generator masking, ray masking, distribution masking) are designed, and the gradient impact on policy gradient algorithms such as PPO is derived, followed by experimental validation in four benchmark environments.
Exclusively Penalized Q-learning for Offline Reinforcement Learning
Junghyuk Yeom (Ulsan National Institute of Science and Technology), Seungyul Han (Ulsan National Institute of Science and Technology)
Reinforcement LearningTabular
🎯 What it does: In offline reinforcement learning, a method called Exclusive Penalty Q-learning (EPQ) is proposed to address the distribution shift problem caused by overestimation errors, which only penalizes states lacking data support, and further introduces a prioritized dataset to enhance the effectiveness of the penalty.
Exocentric-to-Egocentric Video Generation
Jia-Wei Liu (Show Lab), Mike Zheng Shou (Show Lab)
GenerationData SynthesisDiffusion modelVideo
🎯 What it does: Using sparse appearance videos captured by four 360° appearance cameras to generate first-person perspective videos of the same scene and action.
Exogenous Matching: Learning Good Proposals for Tractable Counterfactual Estimation
Yikang Chen (Shanghai Key Laboratory of Trustworthy Computing, East China Normal University), Lili Tian (Shanghai Key Laboratory of Trustworthy Computing, East China Normal University)
Flow-based ModelTabular
🎯 What it does: A method based on importance sampling called Exogenous Matching is proposed for estimating counterfactual expressions that can be solved in general settings.
Expanding Sparse Tuning for Low Memory Usage
Shufan Shen (Institute of Computing Technology, Chinese Academy of Sciences), Shuhui Wang (Institute of Computing Technology, Chinese Academy of Sciences)
OptimizationComputational EfficiencyTransformerSupervised Fine-TuningImage
🎯 What it does: A low-memory sparse fine-tuning framework called SNELL is proposed, which combines the low-rank decomposition of LoRA with kernelization and employs a competitive sparse mechanism to achieve efficient parameter fine-tuning on large-scale pre-trained visual models.
Expectation Alignment: Handling Reward Misspecification in the Presence of Expectation Mismatch
Malek Mechergui (Colorado State University), Sarath Sreedharan (Colorado State University)
OptimizationReinforcement Learning from Human FeedbackReinforcement LearningTabular
🎯 What it does: The Expectation Alignment framework is proposed to represent and address the reward function mis-specification problem through human theoretical attitudes towards agents.
Expected Probabilistic Hierarchies
Marcel Kollovieh (Technical University of Munich), Stephan Günnemann
OptimizationGraph Neural NetworkGraphTabular
🎯 What it does: A hierarchical clustering method based on probability trees (EPH) is proposed, which directly optimizes the expected Dasgupta cost and Tree-Sampling Divergence.
Expectile Regularization for Fast and Accurate Training of Neural Optimal Transport
Nazar Buzun (Artificial Intelligence Research Institute), Dmitry V. Dylov (Skolkovo Institute of Science and Technology)
Image TranslationOptimizationComputational EfficiencyImageBenchmark
🎯 What it does: This paper proposes a neural optimal transport training method based on expected regularization, ENOT, which can efficiently and accurately solve optimal transport plans.
Expert-level protocol translation for self-driving labs
Yu-Zhe Shi (Peking University), Qining Wang (Peking University)
Autonomous DrivingRobotic IntelligenceDrug DiscoveryLarge Language ModelText
🎯 What it does: A protocol translation framework without human intervention has been designed, capable of converting natural language experimental protocols into machine-executable structured protocols layer by layer, ultimately generating a Protocol Dependence Graph (PDG) to automate the experimental process.
Explaining Datasets in Words: Statistical Models with Natural Language Parameters
Ruiqi Zhong (University of California Berkeley), Jacob Steinhardt (University of California Berkeley)
ClassificationOptimizationExplainability and InterpretabilityLarge Language ModelTextTime Series
🎯 What it does: A statistical model framework using natural language predicates as model parameters has been designed, and a general optimization algorithm has been proposed to learn these predicates through continuous relaxation and iterative refinement, applied to tasks such as clustering, time series, and multi-label classification.
Explanations that reveal all through the definition of encoding
Aahlad Manas Puli (New York University), Rajesh Ranganath (New York University)
Explainability and InterpretabilityText
🎯 What it does: This paper presents a formal definition of the 'encoding' phenomenon in explanations and develops a new evaluation method, STRIPE-X, which can effectively detect and distinguish between encoded and non-encoded explanations.
Explicit Eigenvalue Regularization Improves Sharpness-Aware Minimization
Haocheng Luo (Monash University), Trung Le (Monash University)
OptimizationConvolutional Neural NetworkTransformerImageStochastic Differential Equation
🎯 What it does: The paper studies the training dynamics of SAM, proposes a third-order continuous-time model (SDE), reveals the importance of the alignment between the perturbation vector and the maximum eigenvector of the Hessian for sharpness regularization, and based on this, proposes the Eigen-SAM algorithm, which explicitly regularizes the maximum eigenvalue.
Exploitation of a Latent Mechanism in Graph Contrastive Learning: Representation Scattering
Dongxiao He (Tianjin University), Weixiong Zhang (Hong Kong Polytechnic University)
Representation LearningGraph Neural NetworkContrastive LearningGraph
🎯 What it does: Proposed and implemented a graph contrastive learning framework SGRL based on representation scattering.
Exploiting Activation Sparsity with Dense to Dynamic-k Mixture-of-Experts Conversion
Filip Szatkowski (Warsaw University of Technology), Simone Scardapane (Sapienza University of Rome)
Computational EfficiencyTransformerMixture of ExpertsImageText
🎯 What it does: Transforming dense Transformer models into dynamic Mixture-of-Experts (MoE) to enhance inference efficiency through activation sparsity;
Exploiting Descriptive Completeness Prior for Cross Modal Hashing with Incomplete Labels
Haoyang Luo (Harbin Institute of Technology), Yadan Luo (University of Queensland)
RetrievalTransformerPrompt EngineeringContrastive LearningImageMultimodality
🎯 What it does: A cross-modal hashing method called PCRIL is proposed to address the similarity learning problem when labels are incomplete.
Exploiting LLM Quantization
Kazuki Egashira (ETH Zurich), Martin Vechev (ETH Zurich)
Safty and PrivacyAdversarial AttackAI Code AssistantTransformerLarge Language ModelText
🎯 What it does: This paper studies the potential security vulnerabilities of large language models (LLMs) during the zero-copy quantization process, proposing an attack framework that utilizes quantization differences to mask malicious behavior in full-precision models and activate malicious behavior after quantization. It employs projected gradient descent (PGD) for constrained training of the full-precision model, causing it to exhibit malicious outputs after quantization.
Exploiting Representation Curvature for Boundary Detection in Time Series
Yooju Shin (Korea Advanced Institute of Science and Technology), Jae-Gil Lee (Korea Advanced Institute of Science and Technology)
Anomaly DetectionRepresentation LearningContrastive LearningMultimodalityTime Series
🎯 What it does: This paper proposes a time series boundary detection method based on trajectory curvature representation called RECURVE. It uses a representation sequence generated by self-supervised representation learning as a curve, calculates the curvature at each time point, and identifies class boundaries based on small curvature values.
Exploiting the Replay Memory Before Exploring the Environment: Enhancing Reinforcement Learning Through Empirical MDP Iteration
Hongming Zhang (University of Alberta), Martin Müller (Chinese Academy of Sciences)
Reinforcement LearningSequential
🎯 What it does: A new reinforcement learning framework called Experience MDP Iteration (EMIT) is proposed, which improves the performance of reinforcement learning by constructing a series of experience MDPs using data from replay memory and learning the estimated Q function on these MDPs.
Exploration by Learning Diverse Skills through Successor State Representations
Paul-Antoine LE TOLGUENEC, Emmanuel Rachelson (ISAE-Supaero)
Robotic IntelligenceReinforcement Learning
🎯 What it does: By learning a set of diverse skills, the agent achieves comprehensive coverage of the state space in a reward-free environment.
Exploratory Retrieval-Augmented Planning For Continual Embodied Instruction Following
Minjong Yoo (Sungkyunkwan University), Honguk Woo (Sungkyunkwan University)
RetrievalRobotic IntelligenceTransformerLarge Language ModelTextRetrieval-Augmented Generation
🎯 What it does: The ExRAP framework is designed, combining LLM with temporal environmental memory to achieve query evaluation and execution planning for continuous instruction-following tasks, incorporating information gain exploration into the planning process.
Exploring Adversarial Robustness of Deep State Space Models
Biqing Qi (Tsinghua University), Bowen Zhou (Tsinghua University)
Adversarial AttackRecurrent Neural NetworkImage
🎯 What it does: Evaluated the robustness of various deep state space models (SSMs) under adversarial training (AT) and proposed a low-complexity adaptive scaling (AdS) module to enhance robustness.
Exploring and Exploiting the Asymmetric Valley of Deep Neural Networks
Xin-Chun Li (Nanjing University), De-Chuan Zhan (Nanjing University)
ClassificationFederated LearningImage
🎯 What it does: This paper systematically explores the asymmetric valleys in the loss landscape of deep neural networks (DNNs), investigates the impact of the consistency of the noise direction and the sign of the convergence point on valley symmetry, and applies this finding to model fusion and federated learning.
Exploring Behavior-Relevant and Disentangled Neural Dynamics with Generative Diffusion Models
Yule Wang (Georgia Institute of Technology), Anqi Wu (Georgia Institute of Technology)
GenerationRepresentation LearningRecurrent Neural NetworkDiffusion modelAuto EncoderVideo
🎯 What it does: An end-to-end framework based on behavioral information, Variational Autoencoder (VAE), and Video Diffusion Model (VDM) is proposed (BeNeDiff) to learn decoupled neural latent subspaces from head-fixed mouse whole-brain calcium imaging data and to explain the behavioral dynamics corresponding to each latent factor by generating behavioral videos.
Exploring Consistency in Graph Representations: from Graph Kernels to Graph Neural Networks
Xuyuan Liu (Dartmouth), Yujun Yan (Dartmouth)
ClassificationRepresentation LearningGraph Neural NetworkGraph
🎯 What it does: Proposes the idea of maintaining consistency in graph representation similarity between layers of Graph Neural Networks (GNNs) and introduces a consistency loss;
Exploring Context Window of Large Language Models via Decomposed Positional Vectors
zican Dong, Ji-Rong Wen (Renmin University of China)
TransformerLarge Language ModelText
🎯 What it does: This paper explores the positional information of large language models both within and outside the context window. It decomposes the hidden states to obtain positional information vectors, analyzes their formation process and impact on the attention mechanism, and based on this, proposes two training-free context window extension methods (positional information vector replacement and attention window extension).
Exploring DCN-like architecture for fast image generation with arbitrary resolution
Shuai Wang (Nanjing University), Limin Wang (Nanjing University)
GenerationData SynthesisConvolutional Neural NetworkFlow-based ModelImage
🎯 What it does: A pure convolutional generative model FlowDCN is proposed, which can quickly generate high-quality images at any resolution.
Exploring Fixed Point in Image Editing: Theoretical Support and Convergence Optimization
chen hang, Hongbin Wang (Ant Group)
RestorationOptimizationDiffusion modelImage
🎯 What it does: Discusses and proves the existence and uniqueness of fixed points in DDIM inversion, proposing an optimized convergence criterion to improve image editing and dehazing quality.
Exploring Jacobian Inexactness in Second-Order Methods for Variational Inequalities: Lower Bounds, Optimal Algorithms and Quasi-Newton Approximations
Artem Agafonov (Mohamed bin Zayed University of Artificial Intelligence), Dmitry Kamzolov (Mohamed bin Zayed University of Artificial Intelligence)
Optimization
🎯 What it does: This paper proposes a second-order method for variational inequalities that can achieve global convergence even in the presence of inaccuracies in the Jacobian matrix.
Exploring Low-Dimensional Subspace in Diffusion Models for Controllable Image Editing
Siyi Chen (University of Michigan), Qing Qu (University of Michigan)
GenerationData SynthesisDiffusion modelImage
🎯 What it does: This paper proposes LOCO Edit, an unsupervised, single-step low-dimensional subspace editing method that utilizes the posterior mean predictor of diffusion models to exploit local linearity and low-rank properties.
Exploring Molecular Pretraining Model at Scale
Xiaohong Ji (DP Technology), Weinan E (Peking University)
Drug DiscoveryTransformerGraph
🎯 What it does: A scalable molecular pre-training model, Uni-Mol2, is proposed and trained, utilizing a dual-track Transformer to jointly model atomic, graph, and geometric information, and is pre-trained on a large scale with 884M 3D molecular structures.
Exploring Structured Semantic Priors Underlying Diffusion Score for Test-time Adaptation
Mingjia Li (Beijing Institute of Technology), Wei Li (Inceptio Technology)
ClassificationSegmentationDomain AdaptationDiffusion modelScore-based ModelImage
🎯 What it does: A method called DUSA is proposed, which adapts during testing by utilizing structured semantic priors from diffusion model scores;
Exploring the Edges of Latent State Clusters for Goal-Conditioned Reinforcement Learning
Yuanlin Duan (Rutgers University), He Zhu (Rutgers University)
Robotic IntelligenceReinforcement LearningWorld ModelSequential
🎯 What it does: This paper proposes a target-conditioned reinforcement learning method based on clustering edge exploration, CE2, which utilizes latent space clustering to select reachable edge states as exploration targets, thereby efficiently exploring unknown environments and achieving diverse goals.
Exploring the Precise Dynamics of Single-Layer GAN Models: Leveraging Multi-Feature Discriminators for High-Dimensional Subspace Learning
Andrew Bond (Koç University), Zafer Dogan
GenerationData SynthesisGenerative Adversarial NetworkImageStochastic Differential EquationOrdinary Differential Equation
🎯 What it does: This study investigates the training dynamics of a single-layer GAN model in high-dimensional subspace learning and achieves faster and more accurate subspace learning through a multi-feature discriminator.
Exploring the Role of Large Language Models in Prompt Encoding for Diffusion Models
Bingqi Ma (SenseTime Research), Yu Liu (SenseTime Research)
GenerationTransformerLarge Language ModelPrompt EngineeringDiffusion modelImageTextMultimodality
🎯 What it does: This paper proposes a new framework called LLM-Infused Diffuser, which utilizes large language models (LLM) for text prompt encoding. It designs instruction injection, language token refiners, and collaborative refiners to overcome the issues of training objective mismatch and positional bias in LLMs. Based on this framework, the LLM-Infused Diffusion Transformer (LI-DiT) is constructed, significantly improving the quality of text-to-image generation and prompt understanding capabilities.
Exploring the trade-off between deep-learning and explainable models for brain-machine interfaces
Luis Hernan Cubillos, Cynthia Chestek
Explainability and InterpretabilityRobotic IntelligenceRecurrent Neural NetworkTime SeriesBiomedical Data
🎯 What it does: This study investigates the use of KalmanNet (a hybrid model combining Kalman filtering and RNN) for decoding fingertip movements in brain-computer interfaces, comparing it with traditional Kalman filtering, tcFNN, and LSTM in both offline and online experiments.
Exploring Token Pruning in Vision State Space Models
Zheng Zhan (Northeastern University), Yanzhi Wang (Northeastern University)
Object DetectionSegmentationComputational EfficiencyTransformerImage
🎯 What it does: An acceleration method based on token pruning is proposed for visual models based on state space models (SSM).
eXponential FAmily Dynamical Systems (XFADS): Large-scale nonlinear Gaussian state-space modeling
Matthew Dowling (Champalimaud Foundation), Il Memming Park (Champalimaud Foundation)
Computational EfficiencyAuto EncoderTime Series
🎯 What it does: A low-rank structured variational autoencoder framework, XFADS, has been developed for learning nonlinear high-dimensional Gaussian state space models and achieving efficient predictions.
Exponential Quantum Communication Advantage in Distributed Inference and Learning
Dar Gilboa (Google Quantum AI), Jarrod Ryan McClean
Federated LearningSafty and PrivacyComputational EfficiencyGraph Neural NetworkGraph
🎯 What it does: A framework for distributed inference and gradient training on quantum networks is proposed, demonstrating an exponential communication advantage under certain conditions.
Expressive Gaussian Human Avatars from Monocular RGB Video
Hezhen Hu (University of Texas at Austin), Zhangyang Wang (University of Cambridge)
RecognitionGenerationPose EstimationGaussian SplattingVideo
🎯 What it does: This paper proposes the EVA model based on 3D Gaussian and SMPL-X, which learns to drive expressive humanoid avatars from monocular RGB videos.
Extending Multi-modal Contrastive Representations
Ziang Zhang (Zhejiang University), Zhou Zhao (Zhejiang University)
RetrievalRepresentation LearningContrastive LearningImageTextMultimodalityPoint CloudAudio
🎯 What it does: The Ex-MCR method is proposed, utilizing pre-trained unimodal spaces to achieve multimodal contrastive learning with unpaired data through overlapping modalities, extending to unified representations beyond three modalities.
Extending Video Masked Autoencoders to 128 frames
Nitesh Bharadwaj Gundavarapu, Leonid Sigal (University of British Columbia)
ClassificationRepresentation LearningConvolutional Neural NetworkAuto EncoderVideo
🎯 What it does: This paper proposes an adaptive decoder masking strategy that enables video autoencoders to perform self-supervised pre-training on 128-frame long videos, significantly enhancing long-term temporal understanding capabilities.
Extensive-Form Game Solving via Blackwell Approachability on Treeplexes
Darshan Chakrabarti (Columbia University), Christian Kroer (Columbia University)
OptimizationReinforcement Learning from Human FeedbackReinforcement LearningAgentic AITabularBenchmark
🎯 What it does: A treeplex reward minimization framework based on Blackwell approximation is proposed, introducing algorithms such as PTB+ and Smooth PTB+, and solving the Nash equilibrium of two-player zero-sum extensive-form games through a self-play framework.
Externally Valid Policy Evaluation from Randomized Trials Using Additional Observational Data
Sofia Ek (Uppsala University), Dave Zachariah (Uppsala University)
TabularBenchmark
🎯 What it does: Using data from randomized controlled trials (RCT) and additional covariate information from the target population, we construct a non-parametric, finite sample valid lower bound curve for the losses of future individuals under any decision-making strategy.
Extracting Training Data from Molecular Pre-trained Models
Renhong Huang (Zhejiang University), Yang Yang (Lehigh University)
Adversarial AttackDrug DiscoveryGraph Neural NetworkReinforcement LearningGraph
🎯 What it does: This paper studies attack methods for extracting training data from molecular pre-training models and successfully implements effective data leakage attacks on non-generative molecular pre-training models for the first time.
Eye-gaze Guided Multi-modal Alignment for Medical Representation Learning
Chong Ma (ShanghaiTech University), Xiang Li (Massachusetts General Hospital)
ClassificationRetrievalRepresentation LearningTransformerContrastive LearningImageTextMultimodalityBiomedical Data
🎯 What it does: This paper proposes a framework for guiding the multimodal alignment of medical images and text using radiologists' gaze tracking data (EGMA), achieving more precise feature alignment during the pre-training phase through fine-grained alignment loss (EGF) and cross-modal mapping loss (EGM).
EZ-HOI: VLM Adaptation via Guided Prompt Learning for Zero-Shot HOI Detection
Qinqian Lei (National University of Singapore), Robby T. Tan (ASUS Intelligent Cloud Services)
Object DetectionTransformerLarge Language ModelPrompt EngineeringVision Language ModelMultimodality
🎯 What it does: An efficient zero-shot human-object interaction detection framework EZ-HOI is proposed, which utilizes prompt learning for adaptive handling of unseen interaction classes in visual-language models.
F-OAL: Forward-only Online Analytic Learning with Fast Training and Low Memory Footprint in Class Incremental Learning
Huiping Zhuang (South China University of Technology), Lap-Pui Chau (Hong Kong Polytechnic University)
ClassificationComputational EfficiencyRepresentation LearningTransformerImage
🎯 What it does: A forward online parsing learning (F-OAL) method is proposed, which combines a frozen pre-trained encoder with feature fusion and smooth projection, and updates the linear classifier through recursive least squares, achieving online class incremental learning without saving samples.
Face2QR: A Unified Framework for Aesthetic, Face-Preserving, and Scannable QR Code Generation
Xuehao Cui (Shanghai Jiao Tong University), Xiaohong Liu (Shanghai Jiao Tong University)
GenerationDiffusion modelImage
🎯 What it does: A three-stage Face2QR pipeline is designed, unifying the generation of facial identity, aesthetic background, and QR code scanning.
Facilitating Multimodal Classification via Dynamically Learning Modality Gap
Yang Yang (Nanjing University of Science and Technology), Yi Xu (Dalian University of Technology)
ClassificationRecognitionOptimizationConvolutional Neural NetworkTransformerContrastive LearningVideoMultimodalityAudio
🎯 What it does: This paper proposes a dynamic integration of unsupervised contrastive learning and supervised multimodal classification learning, utilizing the intervention of positive label fitting from contrastive learning to alleviate the issue of modality imbalance.
FACT or Fiction: Can Truthful Mechanisms Eliminate Federated Free Riding?
Marco Bornstein (University of Maryland), Furong Huang (University of Maryland)
Federated LearningImageBiomedical Data
🎯 What it does: A federated learning mechanism named FACT is proposed, which can eliminate free-riding and ensure system fairness even when participants may not be honest.
Factorized Diffusion Architectures for Unsupervised Image Generation and Segmentation
Xin Yuan (Google), Michael Maire (University of Chicago)
SegmentationGenerationDiffusion modelImage
🎯 What it does: Train a chunked diffusion model to simultaneously learn image generation and segmentation under unsupervised conditions.
FactorizePhys: Matrix Factorization for Multidimensional Attention in Remote Physiological Sensing
Jitesh Joshi (University College London), Youngjun Cho (University College London)
Convolutional Neural NetworkSupervised Fine-TuningVideo
🎯 What it does: A 3D-CNN model called FactorizePhys and a multi-dimensional attention module based on non-negative matrix factorization (NMF) called FSAM have been developed for end-to-end extraction of remote photoplethysmography (rPPG) signals from video frames.
FactorSim: Generative Simulation via Factorized Representation
Fan-Yun Sun (Stanford University), Nick Haber (Stanford University)
Robotic IntelligenceReinforcement Learning from Human FeedbackReinforcement LearningSequentialChain-of-Thought
🎯 What it does: The FACTORSIM framework is proposed, which can generate complete code simulations from natural language input to train agents, addressing the issue that existing methods only focus on partial challenges.
Fair Allocation in Dynamic Mechanism Design
Alireza Fallah (University of California), Annie S Ulichney
OptimizationTabular
🎯 What it does: This paper studies how to fairly allocate resources to two groups of buyers within T rounds under the framework of dynamic mechanism design, aiming to maximize the seller's discounted revenue while ensuring that each group receives at least a specified proportion of the allocation.
Fair and Welfare-Efficient Constrained Multi-Matchings under Uncertainty
Elita Lobo (University of Massachusetts Amherst), Yair Zick (University of Massachusetts Amherst)
OptimizationTabular
🎯 What it does: This study investigates the multi-matching problem that balances welfare efficiency and group fairness under resource allocation uncertainty.
Fair Bilevel Neural Network (FairBiNN): On Balancing fairness and accuracy via Stackelberg Equilibrium
Mehdi Yazdani-Jahromi (University of Central Florida), Ozlem Garibay
OptimizationTabularElectronic Health Records
🎯 What it does: Through a bi-level (Stackelberg) optimization framework, a neural network is constructed to achieve a Pareto optimal balance between accuracy and fairness (taking demographic fairness as an example);
Fair GLASSO: Estimating Fair Graphical Models with Unbiased Statistical Behavior
Madeline Navarro (Rice University), Santiago Segarra (Rice University)
Recommendation SystemOptimizationGraph Neural NetworkGraph
🎯 What it does: This paper proposes Fair GLASSO, a regularization method that balances fairness and sparsity when estimating Gaussian graphical models (GGM).
Fair Kernel K-Means: from Single Kernel to Multiple Kernel
Peng Zhou (Anhui University), Liang Du (Shanxi University)
OptimizationTabularBenchmark
🎯 What it does: A new fairness regularization term is introduced within the framework of Kernel K-Means and Multiple Kernel K-Means, leading to the development of Fair Kernel K-Means (FKKM) and its multi-kernel version (Fair Multiple Kernel K-Means, FMKKM), achieving a balance between fairness in clustering results and clustering quality.
Fair Online Bilateral Trade
François Bachoc (IMT Université Paul Sabatier Toulouse), Roberto Colomboni (Politecnico di Milano)
Optimization
🎯 What it does: This paper studies the fairness issue in pricing for online bilateral trading platforms and proposes a learning framework aimed at minimizing the utility of both buyers and sellers (fair profit);
Fair Secretaries with Unfair Predictions
Eric Balkanski (Columbia University), Andreas Maggiori (Columbia University)
🎯 What it does: This paper studies how to ensure a fair acceptance probability for the optimal candidate while utilizing predictive information to enhance performance in the learning-augmented setting of the secretary problem.
Fair Wasserstein Coresets
Zikai Xiong (Massachusetts Institute of Technology), Manuela Veloso (J.P.Morgan AI Research)
OptimizationData-Centric LearningTabular
🎯 What it does: This paper proposes the Fair Wasserstein Core Set (FWC), which compresses the original data by generating weighted synthetic representative samples while ensuring demographic fairness in downstream learning tasks.
Fairness and Efficiency in Online Class Matching
MohammadTaghi Hajiaghayi (University of Maryland), Max Springer (University of Maryland)
Recommendation SystemOptimizationGraph
🎯 What it does: A randomized online class fair matching algorithm is proposed, and it is proven to achieve non-wasteful matching with 1/2 approximation of class fairness (CEF), 1/2 approximation of class proportionality (CPROP), and 1/2 approximation of utilitarian social welfare (USW).
Fairness in Social Influence Maximization via Optimal Transport
Shubham Chowdhary (ETH Zurich), Florian Dorfler
OptimizationReinforcement LearningGraphPhysics Related
🎯 What it does: This paper proposes a new fairness metric—Mutual Fairness—for information dissemination in social networks, and based on this metric, designs a seed selection algorithm S3D that balances fairness and efficiency, significantly improving the fairness of information coverage among multiple groups.
Fairness without Harm: An Influence-Guided Active Sampling Approach
Jinlong Pang (University of California Santa Cruz), Yang Liu (University of California Santa Cruz)
ClassificationData-Centric LearningImageTabular
🎯 What it does: This paper proposes an active sampling framework (Fair Influential Sampling, FIS) that does not require training with labeled sensitive attributes. It selects new samples by evaluating their impact on accuracy and fairness on the validation set, thereby reducing fairness imbalance while maintaining model accuracy.
Fairness-Aware Estimation of Graphical Models
Zhuoping Zhou (University of Pennsylvania), Li Shen (University of Pennsylvania)
Recommendation SystemOptimizationGraph Neural NetworkGraphTabularPositron Emission TomographyAlzheimer's Disease
🎯 What it does: This paper proposes a fair graph model (Gaussian, Covariance, Ising) estimation framework, which introduces pairwise graph disparity error and combines it with a custom loss to construct a non-smooth multi-objective optimization aimed at eliminating bias caused by protected attributes.
Fairness-Aware Meta-Learning via Nash Bargaining
Yi Zeng (Meta AI), Ruoxi Jia (Virginia Tech)
OptimizationMeta LearningTabular
🎯 What it does: In fair machine learning, a two-stage Nash-Meta-Learning framework is proposed, first resolving supergradient conflicts through Nash Bargaining, and then optimizing for specific fairness objectives;
FairQueue: Rethinking Prompt Learning for Fair Text-to-Image Generation
Christopher T.H Teo, Ngai-man Cheung
GenerationData SynthesisPrompt EngineeringDiffusion modelImageText
🎯 What it does: This paper analyzes the quality issues of existing prompt-learning methods in fair text-to-image generation, discovering that the learned prompts exhibit biases. It subsequently proposes the FairQueue method (including Prompt Queuing and Attention Amplification) to enhance image quality while maintaining fairness.
FairWire: Fair Graph Generation
Oyku Deniz Kose, Yanning Shen (University of California Irvine)
GenerationData SynthesisGraph Neural NetworkDiffusion modelGraph
🎯 What it does: This paper studies the fairness bias present in graph structures, proposes a theoretical analysis, and designs a fair regularization L FairWire, which is further applied to the graph generation framework FairWire to produce fair synthetic graphs.
FashionR2R: Texture-preserving Rendered-to-Real Image Translation with Diffusion Models
Rui Hu (Zhejiang University), Huamin Wang (Style3D Research)
Image TranslationGenerationDiffusion modelImage
🎯 What it does: Convert computer-rendered fashion images into realistic images.
FasMe: Fast and Sample-efficient Meta Estimator for Precision Matrix Learning in Small Sample Settings
Xiao Tan (Southeast University), Beilun Wang (Southeast University)
Meta LearningTabularBiomedical DataMagnetic Resonance Imaging
🎯 What it does: A fast and sample-efficient meta-learning framework named FasMe is proposed for estimating accurate matrices in high-dimensional environments with insufficient sample sizes.
Fast and Memory-Efficient Video Diffusion Using Streamlined Inference
Zheng Zhan (Northeastern University), Yanzhi Wang (Northeastern University)
GenerationComputational EfficiencyDiffusion modelVideo
🎯 What it does: A training-independent inference framework named Streamlined Inference is proposed, which includes three core modules: Feature Slicer, Operator Grouping, and Step Rehash. It can significantly reduce peak memory usage and computational overhead without compromising video quality.
Fast Best-of-N Decoding via Speculative Rejection
Hanshi Sun (Carnegie Mellon University), Andrea Zanette (Carnegie Mellon University)
GenerationComputational EfficiencyTransformerLarge Language ModelReinforcement LearningText
🎯 What it does: A decoding algorithm named SPECULATIVE REJECTION is proposed, which aligns reasoning during generation by dynamically pruning sequences that are unlikely to produce high-quality responses in the early stages of the generation process using a reward model, achieving scores close to or better than traditional Best-of-N on a single GPU.
Fast Channel Simulation via Error-Correcting Codes
Sharang M. Sriramu (Cornell University), Aaron B. Wagner (Cornell University)
CompressionComputational Efficiency
🎯 What it does: This paper proposes a channel simulation method using Polar Codes called PolarSim, which allows Alice to compress the input she observes and send it to Bob under the premise of shared public randomness. Bob then generates an output that satisfies a given distribution based on the received information and the public randomness.
Fast Encoder-Based 3D from Casual Videos via Point Track Processing
Yoni Kasten (NVIDIA Research), Haggai Maron (Technion)
Object TrackingDepth EstimationComputational EfficiencyTransformerSimultaneous Localization and MappingVideo
🎯 What it does: By learning to map the 2D point trajectories in ordinary videos to 3D structures and camera motion, dynamic scene reconstruction is achieved.
Fast Graph Sharpness-Aware Minimization for Enhancing and Accelerating Few-Shot Node Classification
Yihong Luo (Hong Kong University of Science and Technology), Jing Tang (Hong Kong University of Science and Technology)
ClassificationOptimizationComputational EfficiencyGraph Neural NetworkGraph
🎯 What it does: Proposed a Fast Graph Sharpness-Aware Minimization (FGSAM/FGSAM+) to accelerate and enhance the generalization performance of few-shot node classification.
Fast Iterative Hard Thresholding Methods with Pruning Gradient Computations
Yasutoshi Ida (NTT Computer and Data Science Laboratories), Yasuhiro Fujiwara (NTT Communication Science Laboratories)
OptimizationComputational EfficiencyTabular
🎯 What it does: A fast iterative hard thresholding method is proposed, which accelerates the iterative hard thresholding (IHT) method by safely pruning unnecessary gradient computations to find k important elements in linear regression models.
Fast Last-Iterate Convergence of Learning in Games Requires Forgetful Algorithms
Yang Cai (Yale University), Weiqiang Zheng (Yale University)
Optimization
🎯 What it does: This paper studies the last iteration convergence of online learning algorithms in two-player zero-sum games, focusing on algorithms such as Optimistic FTRL (OFTRL) and its special form Optimistic Multiplicative Weights Update (OMWU), which lack a 'forgetting' mechanism. It proves that they cannot achieve a convergence rate independent of the problem size in the worst case.
Fast Proxy Experiment Design for Causal Effect Identification
Sepehr Elahi (École Polytechnique Fédérale de Lausanne), Patrick Thiran (École Polytechnique Fédérale de Lausanne)
OptimizationComputational EfficiencyGraph Neural NetworkGraph
🎯 What it does: This study proposes an algorithm for designing minimum cost experimental agents to achieve identifiable causal effects and improves upon the traditional 'minimum hitting set' method;
Fast Rates for Bandit PAC Multiclass Classification
Liad Erez (Tel Aviv University), Shay Moran (Technion)
ClassificationOptimizationStochastic Differential Equation
🎯 What it does: A new algorithm for multi-class PAC learning with bandwidth feedback is proposed, which can achieve ε-approximately optimal predictors in finite hypothesis classes or classes with finite Natarajan dimension;
Fast Rates in Stochastic Online Convex Optimization by Exploiting the Curvature of Feasible Sets
Taira Tsuchiya (University of Tokyo and RIKEN), Shinji Ito (University of Tokyo and RIKEN)
Optimization
🎯 What it does: This study explores Online Convex Optimization (OCO) and introduces a new condition and analysis method that achieves fast convergence rates by utilizing the curvature of the feasible set.
Fast samplers for Inverse Problems in Iterative Refinement models
Kushagra Pandey (University of California), Stephan Mandt (University of California)
RestorationSuper ResolutionDiffusion modelFlow-based ModelImage
🎯 What it does: A fast conditional sampling framework called Conditional Conjugate Integrators is designed, which does not require retraining and can be directly applied to pre-trained diffusion or flow models for linear inverse problems (such as super-resolution, deblurring, and inpainting).
Fast Sampling via Discrete Non-Markov Diffusion Models with Predetermined Transition Time
Zixiang Chen (University of California), Quanquan Gu (University of California)
GenerationData SynthesisComputational EfficiencyTransformerDiffusion modelText
🎯 What it does: A discrete non-Markov diffusion model (DNDM) is proposed, along with a training-independent accelerated sampling algorithm.
Fast T2T: Optimization Consistency Speeds Up Diffusion-Based Training-to-Testing Solving for Combinatorial Optimization
Yang Li (Shanghai Jiao Tong University), Junchi Yan (Shanghai Jiao Tong University)
OptimizationGraph Neural NetworkDiffusion modelGraph
🎯 What it does: A fast training-to-testing (Fast T2T) model based on optimized consistency is proposed, which can directly generate high-quality combinatorial optimization solutions from noise in a single step or a few steps.
Fast TRAC: A Parameter-Free Optimizer for Lifelong Reinforcement Learning
Aneesh Muppidi (Harvard University), Heng Yang (Harvard University)
OptimizationReinforcement LearningSequential
🎯 What it does: An optimizer called TRAC is proposed, which is hyperparameter-free and designed for lifelong reinforcement learning, capable of maintaining and improving performance under continuous task distribution shifts.
Fast Tree-Field Integrators: From Low Displacement Rank to Topological Transformers
Krzysztof Marcin Choromanski (Google DeepMind), Snigdha Chaturvedi (Google Research)
ClassificationComputational EfficiencyTransformerMeshGraph
🎯 What it does: A Fast Tree Field Integrator (FTFI) based on low displacement rank matrix theory is proposed, which can efficiently and accurately integrate tensor fields on weighted trees and approximate the metrics of general graphs through tree metrics.
Fast yet Safe: Early-Exiting with Risk Control
Metod Jazbec (University of Amsterdam), Eric Nalisnick (Johns Hopkins University)
SegmentationGenerationComputational EfficiencyImageText
🎯 What it does: This paper proposes a method for setting safe exit thresholds for Early Exit Neural Networks (EENNs) using a risk control framework, significantly improving inference speed while maintaining performance guarantees.
FAST: A Dual-tier Few-Shot Learning Paradigm for Whole Slide Image Classification
Kexue Fu (Qilu University of Technology), Manning Wang (Fudan University)
ClassificationTransformerPrompt EngineeringVision Language ModelImageBiomedical Data
🎯 What it does: A dual-layer few-shot learning paradigm called FAST is proposed to address the high labeling costs and data scarcity issues in whole slide image (WSI) classification.
FastDrag: Manipulate Anything in One Step
Xuanjia Zhao (Harbin Engineering University), Pengming Feng (Nanjing University)
GenerationData SynthesisDiffusion modelImage
🎯 What it does: This paper presents FastDrag, a one-step drag-and-drop image editing method based on latent diffusion models.