NeurIPS 2023 Papers — Page 12
Conference on Neural Information Processing Systems · 3218 papers
Fast Attention Requires Bounded Entries
Josh Alman (Columbia University), Zhao Song (Adobe Research)
Computational EfficiencyTransformerLarge Language Model
🎯 What it does: The paper studies the commonly used attention computation in large language models, proposing that when the input matrix elements are restricted to a small range, near-linear time approximate attention computation can be achieved; it also provides a proof based on the SETH hypothesis that no true sub-quadratic time algorithm exists when the elements are larger.
Fast Bellman Updates for Wasserstein Distributionally Robust MDPs
Zhuodong Yu (City University of Hong Kong), Chin Pang Ho (City University of Hong Kong)
OptimizationReinforcement LearningTabular
🎯 What it does: An efficient algorithm for computing distributed robust Bellman updates for distributed robust Markov decision processes (Wasserstein form) is proposed, along with a corresponding time complexity analysis.
Fast Conditional Mixing of MCMC Algorithms for Non-log-concave Distributions
Xiang Cheng (Massachusetts Institute of Technology), Yusong Zhu (Tsinghua University)
🎯 What it does: For globally mixed slow non-logarithmic convex distributions, this study investigates and proves that MCMC (such as Langevin Monte Carlo and Gibbs sampling) can mix quickly on local subsets, thus introducing and quantifying the concept of 'conditional mixing'.
Fast Exact Leverage Score Sampling from Khatri-Rao Products with Applications to Tensor Decomposition
Vivek Bharadwaj (University of California Berkeley), James Demmel (University of California Berkeley)
OptimizationComputational EfficiencyTabular
🎯 What it does: This paper proposes an efficient data structure for random row sampling based on exact leverage scores on Khatri–Rao product matrices, and applies it to CP decomposition of tensors.
Fast Model DeBias with Machine Unlearning
Ruizhe Chen (Zhejiang University), Zuozhu Liu (Zhejiang University)
OptimizationComputational EfficiencyImageTabular
🎯 What it does: A fast model debiasing framework FMD based on machine unlearning is proposed, which uses a small number of counterfactual samples to debias a trained model;
Fast Optimal Locally Private Mean Estimation via Random Projections
Hilal Asi (Apple Inc), Kunal Talwar (Apple Inc)
Safty and PrivacyComputational EfficiencyTabular
🎯 What it does: This paper studies the problem of local private mean estimation for high-dimensional vectors and proposes a new algorithmic framework called ProjUnit, which can achieve efficient mean estimation while preserving privacy.
Fast Optimal Transport through Sliced Generalized Wasserstein Geodesics
Guillaume Mahey (INSA Rouen Normandie), Nicolas Courty (Université Bretagne Sud)
Image TranslationSuper ResolutionOptimizationImagePoint Cloud
🎯 What it does: A new Wasserstein distance proxy called min-SWGG is proposed, which constructs the distance and corresponding transport plan by projecting the two distributions onto one dimension and matching the order after projection; a closed-form Wasserstein computation formula is also provided when the supporting distribution is on a line.
Fast Partitioned Learned Bloom Filter
Atsuki Sato (University of Tokyo), Yusuke Matsui (University of Tokyo)
OptimizationComputational EfficiencySupervised Fine-TuningTabular
🎯 What it does: This paper proposes two improved versions of the Partition Learning Bloom Filter (PLBF), namely fast PLBF and fast PLBF++, which reduce the number of dynamic programming (DP) table constructions and utilize matrix monotonicity, lowering the construction time of the original PLBF from O(Nk³) to O(Nk²) and O(Nk log N + Nk²), while maintaining or closely approaching the original memory efficiency and false positive rate.
Fast Projected Newton-like Method for Precision Matrix Estimation under Total Positivity
Jian-Feng CAI, Jiaxi Ying (Hong Kong University of Science and Technology)
OptimizationComputational EfficiencyTabularTime SeriesFinance Related
🎯 What it does: A fast projection Newton-type algorithm based on two-metric projection is proposed to solve the high-dimensional Gaussian precision matrix estimation problem under MTP2 constraints.
Fast Rank-1 Lattice Targeted Sampling for Black-box Optimization
Yueming Lyu (Agency for Science Technology and Research)
OptimizationTabular
🎯 What it does: A fast target sampling method using random Rank-1 grids (RLTS) is proposed, which is embedded into the sampling phase of black-box optimization, achieving efficient GP training and inference with O(n log n) complexity, significantly improving query efficiency.
Fast Scalable and Accurate Discovery of DAGs Using the Best Order Score Search and Grow Shrink Trees
Bryan Andrews (University of Minnesota), Erich Kummerfeld (University of Minnesota)
Biomedical DataMagnetic Resonance Imaging
🎯 What it does: A DAG learning algorithm BOSS based on permutation search is proposed, and a Grow-Shrink Tree (GST) cache structure is designed for efficient construction and scoring of DAGs, verifying its scalability and accuracy on large-scale highly connected variables (such as fMRI).
Fast Trainable Projection for Robust Fine-tuning
Junjiao Tian (Georgia Institute of Technology), Zsolt Kira (Georgia Institute of Technology)
ClassificationSegmentationOptimizationSupervised Fine-TuningImage
🎯 What it does: Proposes Fast Trainable Projection (FTP), an algorithm that achieves robust fine-tuning through learning layer-wise projection constraints.
FAST: a Fused and Accurate Shrinkage Tree for Heterogeneous Treatment Effects Estimation
Jia Gu (Peking University), JUN ZHOU
TabularTime SeriesBenchmark
🎯 What it does: A tree model called FAST is proposed to estimate heterogeneous treatment effects by integrating randomized trials and observational data, along with improved criteria for tree splitting.
Faster approximate subgraph counts with privacy
Dung Nguyen (University of Virginia), Anil Vullikanti (University of Virginia)
Safty and PrivacyComputational EfficiencyGraph Neural NetworkGraph
🎯 What it does: This paper proposes a fast differential privacy triangle counting and more general subgraph counting algorithm based on approximate smooth sensitivity;
Faster Differentially Private Convex Optimization via Second-Order Methods
Arun Ganesh (Google Research), Abhradeep Guha Thakurta
OptimizationSafty and PrivacyTabular
🎯 What it does: A differential privacy second-order optimization algorithm for convex optimization is proposed, which includes a private cubic regularized Newton method and a practical dual noise Newton update scheme for logistic regression.
Faster Discrete Convex Function Minimization with Predictions: The M-Convex Case
Taihei Oki (University of Tokyo), Shinsaku Sakaue (University of Tokyo)
Optimization
🎯 What it does: A framework for predicting the acceleration of M-convex function minimization using machine learning is proposed and implemented, with improved time complexity specifically for Laminar, Nested, and Box problems.
Faster Margin Maximization Rates for Generic Optimization Methods
Guanghui Wang (Georgia Institute of Technology), Jacob Abernethy (Google Research)
Optimization
🎯 What it does: This paper derives a faster convergence rate for margin maximization by transforming generic optimization methods (such as mirror descent and steepest descent) into the online learning dynamics of solving regularized bilinear games.
Faster Query Times for Fully Dynamic $k$-Center Clustering with Outliers
Leyla Biabani (Eindhoven University of Technology), Melanie Schmidt (Heinrich Heine University Düsseldorf)
OptimizationComputational Efficiency
🎯 What it does: A fully dynamic k-center clustering algorithm is designed in a dual-dimensional finite metric space, supporting point insertion/deletion and allowing for instant queries of any k and z for a (3+ε) approximate solution.
Faster Relative Entropy Coding with Greedy Rejection Coding
Gergely Flamich (University of Cambridge), José Miguel Hernández-Lobato
CompressionComputational EfficiencyAuto EncoderImage
🎯 What it does: Proposes Greedy Rejection Coding (GRC) and proves its correctness in any probability space, then presents two variants GRCS and GRCD, achieving significantly accelerated relative entropy coding on one-dimensional continuous distributions.
FD-Align: Feature Discrimination Alignment for Fine-tuning Pre-Trained Models in Few-Shot Learning
Kun Song (University of Science and Technology Beijing), Weiran Huang (Shanghai Jiao Tong University)
ClassificationDomain AdaptationTransformerPrompt EngineeringContrastive LearningImage
🎯 What it does: A fine-tuning method named FD-Align is proposed, which enhances the generalization of the CLIP model on downstream tasks with limited samples by maintaining pseudo-feature consistency.
Feature Adaptation for Sparse Linear Regression
Jonathan Kelner, Dhruv Rohatgi (Massachusetts Institute of Technology)
OptimizationTabular
🎯 What it does: A feature adaptation algorithm for sparse linear regression is proposed, which can still achieve near-optimal sample complexity in polynomial time even when the condition number of the covariance matrix is poor and only a few abnormal eigenvalues exist.
Feature Dropout: Revisiting the Role of Augmentations in Contrastive Learning
Alex Tamkin (Stanford University), Noah Goodman (Stanford University)
ClassificationRepresentation LearningContrastive LearningImageMultimodalityAudio
🎯 What it does: This paper constructs a synthetic image/audio dataset with multiple features and conducts experiments using contrastive learning (SimCLR, Viewmaker, etc.) on these datasets to explore the role of data augmentation in the foundational model's learning of multi-task features. It then provides a theoretical analysis of a linear contrastive learning model, proving that in certain cases, adding noise (i.e., disrupting certain features) can accelerate the learning of other features.
Feature Learning for Interpretable, Performant Decision Trees
Jack Henry Good, Artur Dubrawski (Carnegie Mellon University)
OptimizationExplainability and InterpretabilityImageTabular
🎯 What it does: A framework for alternating optimization of feature transformation and differentiable decision trees is proposed, achieving a single decision tree that is both compact and interpretable.
Feature learning via mean-field Langevin dynamics: classifying sparse parities and beyond
Taiji Suzuki (University of Tokyo), Atsushi Nitanda (Kyushu Institute of Technology)
ClassificationComputational EfficiencyTabularStochastic Differential Equation
🎯 What it does: This paper studies the sample complexity and generalization performance of Mean Field Langevin Dynamics (MFLD) in learning k-sparse even functions (second-order or general k-sparse even) and proposes a general theoretical framework along with theoretical proofs and experimental validation.
Feature Likelihood Divergence: Evaluating the Generalization of Generative Models Using Samples
Marco Jiralerspong (Université de Montréal), Gauthier Gidel (Université de Montréal)
GenerationData SynthesisContrastive LearningImage
🎯 What it does: A Feature Likelihood Divergence (FLD) evaluation metric is proposed to simultaneously measure the novelty, fidelity, and diversity of samples generated by models;
Feature Selection in the Contrastive Analysis Setting
Ethan Weinberger (University of Washington), Su-In Lee (University of Washington)
Auto EncoderContrastive LearningBiomedical Data
🎯 What it does: This study investigates the feature selection problem in the context of contrastive analysis and proposes a two-stage contrastive autoencoder combined with a stochastic gate for differentiable feature selection (CFS).
Feature-Learning Networks Are Consistent Across Widths At Realistic Scales
Nikhil Vyas (Harvard University), Cengiz Pehlevan (Harvard University)
TransformerImageText
🎯 What it does: Under the maximum update parameterization (µP), this study investigates and verifies the impact of width on the training dynamics, prediction results, internal representations, and dynamic phenomena (such as boundary stability) of feature learning networks at practical scales, demonstrating that wide networks exhibit consistency when the width is sufficiently large.
FeCAM: Exploiting the Heterogeneity of Class Distributions in Exemplar-Free Continual Learning
Dipam Goswami (Universitat Autònoma de Barcelona), Joost van de Weijer (Universitat Autònoma de Barcelona)
ClassificationRepresentation LearningConvolutional Neural NetworkTransformerImage
🎯 What it does: In sample-free category incremental learning, the FeCAM method is proposed, which significantly improves performance by freezing the feature extractor, constructing class prototypes and covariance matrices, and using Mahalanobis distance for classification.
Fed-CO$_{2}$: Cooperation of Online and Offline Models for Severe Data Heterogeneity in Federated Learning
Zhongyi Cai (ShanghaiTech University), Jingya Wang (ShanghaiTech University)
Federated LearningKnowledge DistillationImage
🎯 What it does: A unified federated learning framework, Fed-CO2, is proposed to simultaneously address two severe data heterogeneity issues: label distribution skew and feature skew. It achieves the integration of global knowledge and local exclusive knowledge through the collaboration of online and offline models.
Fed-FA: Theoretically Modeling Client Data Divergence for Federated Language Backdoor Defense
Zhiyuan Zhang (Peking University), Xu Sun (Peking University)
Federated LearningAdversarial AttackRecurrent Neural NetworkTransformerDiffusion modelText
🎯 What it does: This paper proposes the Fed-FA algorithm for detecting and eliminating clients with backdoors in federated learning, thereby achieving backdoor defense for NLP tasks.
Fed-GraB: Federated Long-tailed Learning with Self-Adjusting Gradient Balancer
Zikai Xiao (Zhejiang University), Zuozhu Liu (Zhejiang University)
ClassificationFederated LearningImage
🎯 What it does: Proposes the Fed-GraB framework, which combines a Global Long-Tail Prior Analyzer (DPA) and an Adaptive Gradient Balancer (SGB) to address the problem of Federated Long-Tail Learning (Fed-LT).
Federated Compositional Deep AUC Maximization
Xinwen Zhang (Temple University), Hongchang Gao (Temple University)
ClassificationOptimizationFederated LearningImage
🎯 What it does: A highly imbalanced data classification method suitable for distributed federated learning is proposed, which directly enhances the prediction performance of minority classes by maximizing AUC;
Federated Conditional Stochastic Optimization
Xidong Wu (University of Pittsburgh), Heng Huang (University of Maryland)
OptimizationFederated LearningTabular
🎯 What it does: This paper proposes a distributed algorithm for Federated Conditional Stochastic Optimization (FCSO), including FCSG, FCSG-M, and Acc-FCSG-M, and provides their theoretical convergence, sample complexity, and communication complexity.
Federated Learning via Meta-Variational Dropout
Insu Jeon (Seoul National University), Gunhee Kim (Seoul National University)
Federated LearningMeta LearningAuto EncoderImage
🎯 What it does: This paper proposes a Bayesian personalized federated learning framework called Meta-Variational Dropout (MetaVD), which uses a hypernetwork to predict the dropout rate for each client, achieving model personalization and compression in non-IID and data-scarce scenarios.
Federated Learning with Bilateral Curation for Partially Class-Disjoint Data
Ziqing Fan (Shanghai Jiao Tong University), Yanfeng Wang (Shanghai Jiao Tong University)
Federated LearningImage
🎯 What it does: This paper proposes a federated learning framework called FedGELA for the scenario of partially complete category data (PCDD), aiming to simultaneously improve the performance of global tasks and local personalization tasks.
Federated Learning with Client Subsampling, Data Heterogeneity, and Unbounded Smoothness: A New Algorithm and Lower Bounds
Michael Crawshaw (George Mason University), Mingrui Liu (George Mason University)
OptimizationFederated LearningRecurrent Neural NetworkTextSequential
🎯 What it does: EPISODE++ is proposed, a federated learning algorithm that achieves linear acceleration and lower communication costs under client sampling, data heterogeneity, and unbounded smoothness conditions;
Federated Learning with Manifold Regularization and Normalized Update Reaggregation
Xuming An (Beijing Institute of Technology), Yong Luo (Wuhan University)
Federated LearningConvolutional Neural NetworkImage
🎯 What it does: Proposed FedMRUR under the federated learning framework, utilizing two mechanisms: hyperplane graph fusion and normalized update aggregation, significantly reducing model inconsistency and compensating for the reduction in global update step size;
Federated Linear Bandits with Finite Adversarial Actions
Li Fan (University of Virginia), Cong Shen (University of Virginia)
Recommendation SystemFederated LearningTabular
🎯 What it does: The FedSupLinUCB algorithm is proposed for the federated linear contextual bandit problem, considering both asynchronous and synchronous communication modes.
Federated Multi-Objective Learning
Haibo Yang (Rochester Institute of Technology), Michinari Momma (Amazon.com Inc.)
OptimizationFederated LearningImage
🎯 What it does: A Federated Multi-Objective Learning (FMOL) framework is proposed, supporting the heterogeneity of objective functions and data across different clients, achieving Pareto stable convergence guarantees in federated learning for the first time.
Federated Spectral Clustering via Secure Similarity Reconstruction
Dong Qiao (Chinese University of Hong Kong), Jicong Fan (Chinese University of Hong Kong)
Federated LearningSafty and PrivacyTabular
🎯 What it does: A secure federated spectral clustering method is designed and implemented, utilizing kernel factorization to implicitly reconstruct the similarity matrix, and achieving differential privacy by adding noise to the data or factors.
FedFed: Feature Distillation against Data Heterogeneity in Federated Learning
Zhiqin Yang (Beihang University), Bo Han (Hong Kong Baptist University)
Federated LearningKnowledge DistillationAuto EncoderGenerative Adversarial NetworkImage
🎯 What it does: The FedFed framework is proposed to alleviate data heterogeneity in Federated Learning through performance-sensitive feature sharing with feature separation and differential privacy protection.
FedGame: A Game-Theoretic Defense against Backdoor Attacks in Federated Learning
Jinyuan Jia (Pennsylvania State University), Radha Poovendran (University of Washington)
Federated LearningImage
🎯 What it does: Developed FedGame, a federated learning backdoor attack defense framework based on minimax games;
FedGCN: Convergence-Communication Tradeoffs in Federated Training of Graph Convolutional Networks
Yuhang Yao (Carnegie Mellon University), Carlee Joe-Wong (Carnegie Mellon University)
Federated LearningSafty and PrivacyComputational EfficiencyGraph Neural NetworkGraph
🎯 What it does: This paper proposes FedGCN, which uses a federated learning framework for semi-supervised node classification on a single large graph, significantly reducing communication overhead during the training process.
FedL2P: Federated Learning to Personalize
Royson Lee (University of Cambridge), Nicholas Donald Lane
Federated LearningMeta LearningImageAudio
🎯 What it does: In the framework of federated learning, FedL2P is proposed to automatically generate batch normalization statistics weights and hierarchical learning rates for each client by learning two types of meta-networks (BNNet and LRNet), achieving personalized fine-tuning strategies.
FedNAR: Federated Optimization with Normalized Annealing Regularization
Junbo Li (Mohamed bin Zayed University of Artificial Intelligence), Hongyi Wang (Carnegie Mellon University)
OptimizationFederated LearningImageText
🎯 What it does: The study investigates the impact of weight decay on convergence and generalization in federated learning, and proposes the FedNAR algorithm to adaptively control weight decay.
Few-Shot Class-Incremental Learning via Training-Free Prototype Calibration
Qi-Wei Wang (Nanjing University), Han-Jia Ye (Nanjing University)
ClassificationRecognitionImage
🎯 What it does: This paper proposes a prototype calibration method called TEEN, which does not require additional training, to address the issue of new classes being misclassified into old classes in few-shot class incremental learning.
Few-shot Generation via Recalling Brain-Inspired Episodic-Semantic Memory
Zhibin Duan (Xidian University), Mingyuan Zhou
GenerationData SynthesisDiffusion modelAuto EncoderImage
🎯 What it does: This paper proposes a Variable Structured Memory Module (VSM) that simultaneously stores episodic and semantic memories, enhancing the adaptability of few-shot generation models.
FGPrompt: Fine-grained Goal Prompting for Image-goal Navigation
Xinyu Sun (South China University of Technology), Mingkui Tan (South China University of Technology)
Robotic IntelligenceConvolutional Neural NetworkReinforcement LearningPrompt EngineeringImage
🎯 What it does: This paper proposes the Fine-grained Goal Prompting (FGPrompt) framework to enhance the performance of image target navigation (ImageNav).
FiGURe: Simple and Efficient Unsupervised Node Representations with Filter Augmentations
Chanakya Ekbote (Microsoft Research India), Ramakrishna B Bairi
Computational EfficiencyRepresentation LearningGraph Neural NetworkContrastive LearningGraph
🎯 What it does: This paper proposes an unsupervised graph node representation learning method called FiGURe, which utilizes a filter bank as a multi-spectral view and combines random Fourier feature projection to achieve low-dimensional and efficient embedding.
Find What You Want: Learning Demand-conditioned Object Attribute Space for Demand-driven Navigation
Hongcheng Wang (Peking University), Hao Dong (Peking University)
Robotic IntelligenceTransformerLarge Language ModelContrastive LearningTextMultimodality
🎯 What it does: This paper proposes a Demand-Driven Navigation (DDN) task, where agents search for objects that satisfy natural language needs (e.g., 'I am thirsty') in unknown environments, rather than being given specific objects. It also introduces the extraction of attribute features under demand conditions using large language models, aligning them with visual features through CLIP to form a visual attribute space that enhances navigation performance.
Finding Counterfactually Optimal Action Sequences in Continuous State Spaces
Stratis Tsirtsis (Max Planck Institute for Software Systems), Manuel Gomez Rodriguez
OptimizationReinforcement Learning from Human FeedbackReinforcement LearningTime SeriesBiomedical DataElectronic Health Records
🎯 What it does: This study investigates the problem of finding counterfactual optimal action sequences in continuous state spaces, constructing a framework based on continuous MDP and reversible structural causal models (SCM).
Finding Local Minima Efficiently in Decentralized Optimization
Wenhan Xian (University of Maryland), Heng Huang (University of Maryland)
Recommendation SystemOptimizationTabular
🎯 What it does: A decentralized stochastic gradient algorithm named PEDESTAL is proposed, which can efficiently escape saddle points and find local optimal solutions for non-convex optimization problems.
Finding Order in Chaos: A Novel Data Augmentation Method for Time Series in Contrastive Learning
Berken Utku Demirel (ETH Zurich), Christian Holz (ETH Zurich)
ClassificationRecognitionData-Centric LearningAuto EncoderContrastive LearningTime Series
🎯 What it does: A custom mixup data augmentation method specifically designed for quasi-periodic time series data in self-supervised contrastive learning is proposed, avoiding the destruction of signal information caused by traditional mixup.
Finding Safe Zones of Markov Decision Processes Policies
Lee Cohen (TTI-Chicago), Michal Moshkovitz (Bosch Center for AI)
OptimizationReinforcement LearningTabular
🎯 What it does: The SAFEZONE problem is proposed: under a given Markov Decision Process (MDP) policy, find a subset of states such that most trajectories remain within this subset, and a feasible approximation algorithm is provided.
Fine-Grained Cross-View Geo-Localization Using a Correlation-Aware Homography Estimator
Xiaolong Wang (Zhejiang University), Yu Zhang (Zhejiang University)
Pose EstimationDomain AdaptationComputational EfficiencyConvolutional Neural NetworkContrastive LearningImage
🎯 What it does: This paper proposes HC-Net, which uses a differentiable spherical transformation to project ground panoramic images into bird's-eye views, and directly aligns satellite images through a single iterative depth homomorphic estimator with relevant perception, achieving accurate GPS positioning and orientation.
Fine-grained Expressivity of Graph Neural Networks
Jan Böker (RWTH Aachen University), Christopher Morris (RWTH Aachen University)
Graph Neural NetworkGraph
🎯 What it does: This paper presents a fine-grained expressiveness theory of Message Passing Neural Networks (MPNN) on graph subsets and provides a continuous measure of similarity between graph subsets.
Fine-Grained Human Feedback Gives Better Rewards for Language Model Training
Zeqiu Wu (University of Washington), Hannaneh Hajishirzi (University of Washington)
Reinforcement Learning from Human FeedbackTransformerLarge Language ModelReinforcement LearningText
🎯 What it does: This paper proposes the FINE-GRAINED RLHF framework, enabling language models to utilize fine-grained human feedback (such as sentence-level error annotations) to train reward models and perform reinforcement learning;
Fine-grained Late-interaction Multi-modal Retrieval for Retrieval Augmented Visual Question Answering
Weizhe Lin (University of Cambridge), Bill Byrne (University of Cambridge)
RetrievalTransformerLarge Language ModelVision Language ModelContrastive LearningImageTextMultimodality
🎯 What it does: Proposes Fine-grained Late-interaction Multi-modal Retrieval (FLMR) to enhance the knowledge retrieval quality of Retrieval Augmented Visual Question Answering (RA-VQA), thereby improving VQA performance.
Fine-Grained Theoretical Analysis of Federated Zeroth-Order Optimization
Jun Chen (Huazhong Agricultural University), Hao Deng (Huazhong Agricultural University)
OptimizationFederated LearningImage
🎯 What it does: This paper conducts a fine-grained theoretical analysis of the Federated Zero-order Optimization (FedZO) algorithm, providing upper bounds for its generalization error and optimization error, covering both synchronous and asynchronous modes.
Fine-Grained Visual Prompting
Lingfeng Yang (Nanjing University of Science and Technology), Jian Yang (Beijing Academy of Artificial Intelligence)
ClassificationObject DetectionSegmentationTransformerPrompt EngineeringVision Language ModelImage
🎯 What it does: This paper proposes Fine-Grained Visual Prompts (FGVP), which obtain semantic masks through the Segment Anything model and apply Blur Reverse Mask visual prompts on images to achieve zero-shot instance-level tasks (such as pointing expression understanding and part detection).
Fine-Tuning Language Models with Just Forward Passes
Sadhika Malladi (Princeton University), Sanjeev Arora (Princeton University)
OptimizationTransformerLarge Language ModelSupervised Fine-TuningPrompt EngineeringText
🎯 What it does: A memory-efficient zero-order optimizer MeZO is proposed, which can fine-tune large-scale language models using only forward passes.
FineMoGen: Fine-Grained Spatio-Temporal Motion Generation and Editing
Mingyuan Zhang (Nanyang Technological University), Ziwei Liu (Nanyang Technological University)
GenerationTransformerLarge Language ModelMixture of ExpertsDiffusion modelVideoText
🎯 What it does: The FineMoGen framework is proposed, supporting fine-grained spatio-temporal motion generation and editing, and a HuMMan-MoGen dataset covering 2,968 videos and 102,336 fine-grained descriptions is constructed.
Finite Population Regression Adjustment and Non-asymptotic Guarantees for Treatment Effect Estimation
Mehrdad Ghadiri (Massachusetts Institute of Technology), Anup Rao
Tabular
🎯 What it does: This paper studies the use of regression adjustment techniques to reduce the variance of treatment effect estimates through auxiliary covariates in a finite population experimental setting, and provides non-asymptotic precision bounds.
Finite-Time Analysis of Single-Timescale Actor-Critic
Xuyang Chen (National University of Singapore), Lin Zhao (National University of Singapore)
Reinforcement Learning
🎯 What it does: This paper provides a theoretical analysis of the finite-time convergence of the single-time-scale Actor-Critic (AC) algorithm under continuous state space, linear function approximation, and Markovian sampling, proving that its sample complexity to reach an ε-approximate stationary point is approximately ˜O(ε⁻²) (O(ε⁻²) under i.i.d. sampling).
Finite-Time Analysis of Whittle Index based Q-Learning for Restless Multi-Armed Bandits with Neural Network Function Approximation
GUOJUN XIONG, Jian Li (Stony Brook University)
Reinforcement Learning
🎯 What it does: This paper proposes Neural-Q-Whittle, a Whittle index Q-learning algorithm based on neural network function approximation, aimed at solving the restless multi-armed bandit (RMAB) problem under a model-free (Markov observation) setting, and provides a finite-time convergence analysis.
Finite-Time Logarithmic Bayes Regret Upper Bounds
Alexia Atsidakou (University of Texas), sujay sanghavi
🎯 What it does: This paper presents for the first time the finite-time logarithmic Bayesian upper bound of the UCB algorithm in multi-armed and linear Bandits, breaking through the previous ~O(√n) upper bound;
FIRAL: An Active Learning Algorithm for Multinomial Logistic Regression
Youguang Chen (Oden Institute for Computational Engineering and Sciences University of Texas at Austin), George Biros (Oden Institute for Computational Engineering and Sciences University of Texas at Austin)
ClassificationOptimizationImage
🎯 What it does: This study investigates pooling active learning, proposing theories and algorithms for multi-class logistic regression. It proves that the Fisher Information Ratio (FIR) can bound the excess risk and designs the FIRAL active learning algorithm based on this.
First Order Methods with Markovian Noise: from Acceleration to Variational Inequalities
Aleksandr Beznosikov (Innopolis University), Eric Moulines (Ecole Polytechnique)
Optimization
🎯 What it does: This paper proposes a unified stochastic batch size framework for gradient methods with Markov noise, achieving optimal finite-time performance in non-convex, strongly convex optimization, and variational inequality (VI) problems.
First Order Stochastic Optimization with Oblivious Noise
Ilias Diakonikolas (University of Wisconsin-Madison), Christos Tzamos (University of Athens)
Optimization
🎯 What it does: Proposes and solves a stochastic optimization problem in the presence of oblivious noise, providing an algorithm to find approximate steady points within the list solvable framework.
First- and Second-Order Bounds for Adversarial Linear Contextual Bandits
Julia Olkhovskaya (Delft University of Technology), Chen-Yu Wei (Massachusetts Institute of Technology)
Adversarial AttackReinforcement Learning
🎯 What it does: This paper studies the setting of adversarial linear contextual multi-armed bandits and proposes improvements to the expected regret bounds of the algorithm under known distributions in contextual situations.
Fitting trees to $\ell_1$-hyperbolic distances
Joon-Hyeok Yim, Anna Gilbert
Graph
🎯 What it does: This study investigates how to fit a given distance using trees (or equidistant trees) in finite metric spaces and presents the tree fitting algorithm HCCROOTEDTREEFIT for the ℓ1 norm; it also proposes a theoretical framework that associates hyperbolicity vectors with tree fitting errors.
Fixing the NTK: From Neural Network Linearizations to Exact Convex Programs
Rajat Vadiraj Dwaraknath (Stanford University), Mert Pilanci (Stanford University)
OptimizationTabular
🎯 What it does: This paper studies the relationship between the convex rewriting of finite-width gated ReLU networks and the infinite-width Neural Tangent Kernel (NTK), proving that the NTK is equivalent to a weighted combination of masked kernels. It improves the NTK weights using multi-kernel learning and Iteratively Reweighted Least Squares (IRLS) methods, resulting in better training performance.
Flat Seeking Bayesian Neural Networks
Van-Anh Nguyen (Monash University), Trung Le (Monash University)
ClassificationOptimizationConvolutional Neural NetworkImage
🎯 What it does: This paper proposes Sharpness-Aware Posterior (SA-Posterior), which incorporates the consideration of model flatness in the posterior inference of Bayesian neural networks, resulting in models sampled from the posterior that are broader and more robust.
FlatMatch: Bridging Labeled Data and Unlabeled Data with Cross-Sharpness for Semi-Supervised Learning
Zhuo Huang (University of Sydney), Tongliang Liu (University of Sydney)
ClassificationConvolutional Neural NetworkSupervised Fine-TuningImage
🎯 What it does: This paper proposes FlatMatch, a semi-supervised learning method that bridges labeled and unlabeled data through cross-sharpness regularization, enhancing the model's generalization performance.
Flexible Attention-Based Multi-Policy Fusion for Efficient Deep Reinforcement Learning
Zih-Yun Chiu (University of California), Michael C. Yip (University of California)
Reinforcement LearningSequential
🎯 What it does: The KGRL framework and KIAN model are proposed, enabling RL to utilize and integrate any external policies to achieve knowledge accessibility, sample efficiency, transferability, composability, and incremental learning.
Flocks of Stochastic Parrots: Differentially Private Prompt Learning for Large Language Models
Haonan Duan (University of Toronto), Franziska Boenisch (University of Toronto)
Safty and PrivacyTransformerLarge Language ModelPrompt EngineeringText
🎯 What it does: A new method for implementing differential privacy (prompt) learning in large language models (LLM) is proposed, including both soft prompt-based PromptDPSGD and PromptPATE suitable for black-box APIs;
Flow Factorized Representation Learning
Yue Song (University of Trento), Max Welling (University of Amsterdam)
GenerationRepresentation LearningFlow-based ModelImage
🎯 What it does: This paper proposes Flow Factorized Representation Learning, which utilizes the gradient potential field learned in the latent space (dynamically optimal transport evolving over time) to generate decomposable transformation paths, achieving factorized modeling of input transformations.
Flow Matching for Scalable Simulation-Based Inference
Jonas Bernhard Wildberger, Bernhard Schölkopf (Max Planck Institute for Intelligent Systems)
Flow-based ModelTime SeriesSequentialPhysics Related
🎯 What it does: A Bayesian posterior estimation method using continuous normalizing flows and flow matching (FMPE) is proposed, which can achieve scalable and directly assessable posterior distributions in simulation inference.
Flow-Attention-based Spatio-Temporal Aggregation Network for 3D Mask Detection
Yuxin Cao (Tsinghua University), Minhui Xue (CSIRO)
RecognitionObject DetectionConvolutional Neural NetworkOptical FlowImageVideo
🎯 What it does: Proposes the FASTEN framework, which utilizes facial optical flow networks, flow attention, and spatiotemporal aggregation to achieve 3D mask defense;
Flow-Based Feature Fusion for Vehicle-Infrastructure Cooperative 3D Object Detection
Haibao Yu (University of Hong Kong), Zaiqing Nie (Tsinghua University)
Object DetectionAutonomous DrivingFlow-based ModelPoint Cloud
🎯 What it does: A vehicle-infrastructure collaborative 3D detection framework FFNet based on feature flow is proposed, which utilizes feature flow to predict future features to compensate for temporal asynchrony and achieve low communication overhead.
Flow: Per-instance Personalized Federated Learning
Kunjal Panchal (University of Massachusetts), Hui Guan (University of Massachusetts)
Federated LearningImageText
🎯 What it does: This paper proposes a method called Flow that simultaneously achieves personalized models for each client and each instance in federated learning, utilizing dynamic routing to determine whether to use local parameters or global parameters during inference.
FlowCam: Training Generalizable 3D Radiance Fields without Camera Poses via Pixel-Aligned Scene Flow
Cameron Omid Smith (Massachusetts Institute of Technology), Vincent Sitzmann (Massachusetts Institute of Technology)
GenerationPose EstimationNeural Radiance FieldSimultaneous Localization and MappingOptical FlowVideo
🎯 What it does: Through a single forward inference, without prior camera pose, jointly reconstruct a general 3D neural field and estimate the camera trajectory.
FlowPG: Action-constrained Policy Gradient with Normalizing Flows
Janaka Chathuranga Brahmanage (Singapore Management University), Akshat Kumar (Singapore Management University)
Reinforcement LearningFlow-based ModelTabular
🎯 What it does: Developed and implemented a constraint-based reinforcement learning method called FlowPG, which directly generates actions that satisfy constraints during the reinforcement learning process, avoiding traditional projection solutions.
FLSL: Feature-level Self-supervised Learning
Qing Su (Georgia State University), Shihao Ji (Duke University)
Object DetectionSegmentationTransformerContrastive LearningImageVideo
🎯 What it does: A feature layer self-supervised learning method based on ViT, called FLSL, is designed, utilizing a dual-layer clustering framework that combines mean shift clustering and k-means, allowing image features to achieve semantic clustering at both local and global levels, significantly improving the performance of dense prediction tasks.
FLuID: Mitigating Stragglers in Federated Learning using Invariant Dropout
Irene Wang (University of British Columbia), Divya Mahajan (Microsoft)
Federated LearningConvolutional Neural NetworkRecurrent Neural NetworkImageText
🎯 What it does: The FLuID framework is proposed, which dynamically identifies and discards 'invariant' neurons that contribute little to the global model through Invariant Dropout, thereby generating sub-models for slow devices (stragglers) to alleviate their computational and communication load;
FOCAL: Contrastive Learning for Multimodal Time-Series Sensing Signals in Factorized Orthogonal Latent Space
Shengzhong Liu (Shanghai Jiao Tong University), Tarek Abdelzaher (University of Illinois at Urbana-Champaign)
Anomaly DetectionRepresentation LearningTransformerContrastive LearningMultimodalityTime Series
🎯 What it does: This paper proposes a self-supervised contrastive learning framework named FOCAL, specifically designed to extract shared and private features from multimodal time series perception signals and achieve semantic representation through contrastive learning.
Focus on Query: Adversarial Mining Transformer for Few-Shot Segmentation
Yuan Wang (University of Science and Technology of China), Tianzhu Zhang (University of Science and Technology of China)
SegmentationTransformerGenerative Adversarial NetworkImage
🎯 What it does: A query-centered few-shot semantic segmentation model AMFormer is proposed, which achieves precise segmentation of query images using adversarial mining Transformer with only rough support information.
Focus Your Attention when Few-Shot Classification
Haoqing Wang (Peking University), Zhi-Hong Deng (Peking University)
ClassificationMeta LearningTransformerPrompt EngineeringContrastive LearningImage
🎯 What it does: For the few-shot image classification task, this paper designs a Position Prompt to guide the pre-trained visual Transformer to focus on the key entities in the image that are most relevant to the current category during the fine-tuning process, thereby improving classification performance.
Focused Transformer: Contrastive Training for Context Scaling
Szymon Tworkowski (IDEAS NCBR), Piotr Miłoś (IDEAS NCBR)
TransformerContrastive LearningText
🎯 What it does: A Focused Transformer (FOT) is proposed, which extends the effective context length of the model by training the attention layer through contrastive learning.
Follow-ups Also Matter: Improving Contextual Bandits via Post-serving Contexts
Chaoqi Wang (University of Chicago), Haifeng Xu (University of Chicago)
Recommendation SystemReinforcement LearningTabular
🎯 What it does: This paper proposes a novel contextual linear multi-armed bandit model that considers the 'post-service context' and designs a corresponding online learning algorithm called poLinUCB.
For SALE: State-Action Representation Learning for Deep Reinforcement Learning
Scott Fujimoto (Mila McGill University), David Meger (Mila McGill University)
Representation LearningReinforcement LearningSequential
🎯 What it does: This paper proposes the SALE (State-Action Representation Learning) method, which enhances representation learning for low-dimensional continuous control tasks by learning state-action embeddings, and combines it with TD3 to form a new TD7 algorithm.
ForecastPFN: Synthetically-Trained Zero-Shot Forecasting
Samuel Dooley (Abacus.AI), Colin White (California Institute of Technology)
TransformerTime Series
🎯 What it does: This paper proposes ForecastPFN, a zero-shot time series forecasting model pre-trained with synthetic data, capable of making predictions without further training.
ForkMerge: Mitigating Negative Transfer in Auxiliary-Task Learning
Junguang Jiang (Tsinghua University), Mingsheng Long (Tsinghua University)
Domain AdaptationOptimizationImage
🎯 What it does: The ForkMerge method is proposed to dynamically find the optimal auxiliary task weights through periodic forking and merging models, thereby alleviating the negative transfer problem in auxiliary task learning.
Formalizing locality for normative synaptic plasticity models
Colin Bredenberg (Mila Quebec AI Institute), Guillaume Lajoie (Mila Quebec AI Institute)
🎯 What it does: A unified S_p-locality framework is proposed, using probabilistic graphical models and set theory to define synaptically accessible variables, categorizing various learning algorithms into different locality classes;
Formulating Discrete Probability Flow Through Optimal Transport
Pengze Zhang (Sun Yat-sen University), Xiaohua Xie (Sun Yat-sen University)
GenerationData SynthesisOptimizationDiffusion modelScore-based ModelImage
🎯 What it does: Proposed a discrete probability flow theory and implemented corresponding sampling methods.
Foundation Model is Efficient Multimodal Multitask Model Selector
Fanqing Meng (Shanghai AI Laboratory), Ping Luo (University of Hong Kong)
ClassificationOptimizationComputational EfficiencyTransformerLarge Language ModelVision Language ModelImageTextMultimodality
🎯 What it does: This paper studies a model selection method called EMMS, which can quickly predict the performance of pre-trained models in multi-modal multi-task scenarios.
FouriDown: Factoring Down-Sampling into Shuffling and Superposing
Qi Zhu (University of Science and Technology of China), Feng Zhao (University of Science and Technology of China)
RestorationConvolutional Neural NetworkImage
🎯 What it does: This paper proposes a general framework for downsampling in the Fourier domain—FouriDown, which can dynamically learn frequency weighting and overlapping mechanisms based on image context, addressing the bias and aliasing issues caused by traditional static weighting.
FourierGNN: Rethinking Multivariate Time Series Forecasting from a Pure Graph Perspective
Kun Yi (Beijing Institute of Technology), Zhendong Niu (HeFei University of Technology)
Graph Neural NetworkTime Series
🎯 What it does: Redefines the multivariate time series forecasting problem as a pure graph neural network task on a hyper-variable graph and designs FourierGNN for prediction.
FourierHandFlow: Neural 4D Hand Representation Using Fourier Query Flow
Jihyun Lee (KAIST), Tae-Kyun Kim (Imperial College London)
Pose EstimationRepresentation LearningGraph Neural NetworkOptical FlowVideo
🎯 What it does: A four-dimensional continuous representation of the hand based on Fourier query flow is proposed, utilizing RGB video to learn the spatiotemporal continuous reconstruction of hand shapes.
Fractal Landscapes in Policy Optimization
Tao Wang (University of California San Diego), Sicun Gao (University of California San Diego)
OptimizationReinforcement LearningSequential
🎯 What it does: The study proves that in continuous control problems, the objective function of policy gradients may exhibit fractal (non-smooth) characteristics, and proposes a method for estimating the Hölder index from samples.