ICML 2023 Papers — Page 6
International Conference on Machine Learning · 1828 papers
Efficiently predicting high resolution mass spectra with graph neural networks
Michael Murphy (Massachusetts Institute of Technology), Thomas Butler (Enveda Biosciences)
Graph Neural NetworkGraph
🎯 What it does: This paper studies a novel graph neural network GRAFF-MS for predicting high-resolution mass spectrometry.
Eliminating Adversarial Noise via Information Discard and Robust Representation Restoration
Dawei Zhou (Xidian University), Tongliang Liu (University of Sydney)
RestorationRepresentation LearningAdversarial AttackTransformerAuto EncoderGenerative Adversarial NetworkImage
🎯 What it does: This paper proposes a defense model based on information discard and robust representation recovery, which destroys the spatial structure of noise by randomly generating complementary masks on adversarial samples, and utilizes adversarial games to guide the removal of non-robust features and recover robust representations.
ELSA: Efficient Label Shift Adaptation through the Lens of Semiparametric Models
Qinglong Tian (University of Waterloo), Jiwei Zhao (University of Wisconsin Madison)
Domain AdaptationImage
🎯 What it does: This paper studies the unsupervised domain adaptation problem of label shift, proposing a moment matching framework based on a semi-parametric model, and on this basis, provides an importance weight estimator for Efficient Label Shift Adaptation (ELSA).
EM-Network: Oracle Guided Self-distillation for Sequence Learning
Ji Won Yoon (Seoul National University), Nam Soo Kim (Seoul National University)
Knowledge DistillationTransformerSequentialAudio
🎯 What it does: This paper proposes EM-Network, a method that utilizes target sequences to generate oracle guidance and enhances the performance of seq2seq models in single-stage self-distillation.
Emergence of Adaptive Circadian Rhythms in Deep Reinforcement Learning
aqeel labash, Raul Vicente (Institute of Computer Science University of Tartu)
Recurrent Neural NetworkReinforcement LearningTime SeriesSequential
🎯 What it does: In a feeding task with periodic lighting, the study investigates how deep reinforcement learning agents spontaneously generate and internalize circadian rhythms, thereby achieving prediction and adaptation to environmental cycles.
Emergence of Sparse Representations from Noise
Trenton Bricken (Harvard University), Gabriel Kreiman (Harvard Medical School)
ClassificationRepresentation LearningConvolutional Neural NetworkTransformerAuto EncoderImageText
🎯 What it does: The study introduces random noise during the training of neural networks and finds that noise training can automatically produce sparse activations and enable hidden layer neurons to learn receptor fields similar to those in biological visual systems.
Emergent Agentic Transformer from Chain of Hindsight Experience
Hao Liu (University of California), Pieter Abbeel (University of California)
TransformerReinforcement LearningAgentic AIMultimodality
🎯 What it does: This study trained a Transformer model called Agentic Transformer (AT), which utilizes multiple sub-optimal trajectories concatenated in ascending order of rewards (Chain of Hindsight Experience) and re-labels target returns. The model predicts actions autoregressively, allowing it to learn combined sub-optimal experiences during the training phase and improve performance through multiple rounds of trial and error during the testing phase.
Emergent Asymmetry of Precision and Recall for Measuring Fidelity and Diversity of Generative Models in High Dimensions
Mahyar Khayatkhoei (University of Southern California), Wael AbdAlmageed (University of Southern California)
GenerationData SynthesisConvolutional Neural NetworkGenerative Adversarial NetworkImage
🎯 What it does: This study investigates the limitations of the evaluation metrics Precision and Recall using k-nearest neighbor approximation in high-dimensional spaces. It discovers and proves the phenomenon of 'emergent asymmetry' and proposes the symmetric metrics symPrecision and symRecall, followed by experimental validation on synthetic and real image data.
Enabling First-Order Gradient-Based Learning for Equilibrium Computation in Markets
Nils Kohring (Technical University of Munich), Martin Bichler (Technical University of Munich)
OptimizationReinforcement LearningTabularFinance Related
🎯 What it does: To achieve market equilibrium calculations, a differentiable softened auction game is introduced, and a first-order gradient method is used to solve the strategy;
End-to-end Differentiable Clustering with Associative Memories
Bishwajit Saha (Rensselaer Polytechnic Institute), Parikshit Ram (IBM Research)
Tabular
🎯 What it does: A differentiable clustering algorithm ClAM is implemented using Associative Memory (AM) dynamics and energy functions, retaining hard cluster assignments.
End-to-End Full-Atom Antibody Design
Xiangzhe Kong (Tsinghua University), Yang Liu (Tsinghua University)
Drug DiscoveryProtein Structure PredictionGraph Neural NetworkBiomedical Data
🎯 What it does: An end-to-end all-atom antibody design framework called dyMEAN is proposed, which can simultaneously generate the complete sequence and full three-dimensional structure of antibodies, as well as perform antigen-antibody docking, given only the three-dimensional structure of the epitope and an incomplete antibody sequence.
End-to-End Learning for Stochastic Optimization: A Bayesian Perspective
Yves Rychener (Ecole Polytechnique Fédérale de Lausanne), Tobias Sutter (University of Konstanz)
OptimizationTime Series
🎯 What it does: This paper proposes an end-to-end framework for learning decision mappings directly from data and provides its Bayesian interpretation. Based on this interpretation, new training algorithms for empirical risk minimization (ERM) and distributionally robust optimization (DRO) are designed.
End-to-End Multi-Object Detection with a Regularized Mixture Model
Jaeyoung Yoo (NAVER WEBTOON AI), Nojun Kwak (Seoul National University)
Object DetectionMixture of ExpertsImage
🎯 What it does: An end-to-end multi-object detection framework D-RMM based on a mixed model is proposed, which directly learns the target distribution through negative log-likelihood.
End-to-end Training of Deep Boltzmann Machines by Unbiased Contrastive Divergence with Local Mode Initialization
Shohei Taniguchi (University of Tokyo), Yutaka Matsuo (University of Tokyo)
GenerationContrastive LearningImage
🎯 What it does: An unbiased contrastive divergence (UCD-LMI) algorithm based on Metropolis-Hastings maximum coupling and local mode initialization is proposed, enabling deep Boltzmann machines (DBM) to be trained end-to-end without greedy pre-training.
Enforcing Hard Constraints with Soft Barriers: Safe Reinforcement Learning in Unknown Stochastic Environments
Yixuan Wang (Northwestern University), Qi Zhu
OptimizationSafty and PrivacyReinforcement LearningSequentialStochastic Differential Equation
🎯 What it does: In an unknown continuous stochastic environment, a safety reinforcement learning framework based on soft barrier functions is proposed, utilizing generative models to learn environmental dynamics, and through bi-level optimization, simultaneously learning soft barriers, generative models, and policies to achieve direct encoding of hard safety probability constraints.
Enhancing Activity Prediction Models in Drug Discovery with the Ability to Understand Human Language
Philipp Seidl (Johannes Kepler University), Günter Klambauer (Johannes Kepler University)
Drug DiscoveryTransformerContrastive LearningTextBiomedical Data
🎯 What it does: A model called CLAMP is proposed, which can achieve zero/few-shot activity prediction by understanding experimental text during inference.
Entity Divider with Language Grounding in Multi-Agent Reinforcement Learning
Ziluo Ding (Peking University), Zongqing Lu (Peking University)
Robotic IntelligenceReinforcement LearningAgentic AITextMultimodality
🎯 What it does: This paper proposes a multi-agent reinforcement learning framework named EnDi, which utilizes natural language manuals and target instructions to represent environmental entities, allowing each agent to independently define sub-goals (self and others), thereby achieving entity-level sub-goal allocation and collaborative decision-making.
Entropy-driven Unsupervised Keypoint Representation Learning in Videos
Ali Younes (Technische Universitat Darmstadt), Georgia Chalvatzaki
Object DetectionObject TrackingRepresentation LearningConvolutional Neural NetworkVideo
🎯 What it does: A purely unsupervised keypoint learning method based on image spatial entropy, MINT, is proposed for extracting sparse and spatiotemporally consistent keypoint representations from videos.
Equivariance with Learned Canonicalization Functions
Sékou-Oumar Kaba (McGill University), Siamak Ravanbakhsh (Mila Quebec Artificial Intelligence Institute)
ClassificationSegmentationOptimizationConvolutional Neural NetworkImagePoint Cloud
🎯 What it does: This paper proposes achieving equivariance in networks through a learned canonicalization function, eliminating the need to impose symmetry constraints on the network architecture.
Equivariant Architectures for Learning in Deep Weight Spaces
Aviv Navon (Bar Ilan University), Haggai Maron (Nvidia)
Domain AdaptationContrastive LearningImage
🎯 What it does: This paper studies and proposes a deep network weight space equivariant network architecture called DWSNet, which can directly handle sequences of network weights and is suitable for tasks such as weight editing and domain transfer.
Equivariant Polynomials for Graph Neural Networks
Omri Puny (Weizmann Institute of Science), Yaron Lipman (Weizmann Institute of Science)
Graph Neural NetworkGraph
🎯 What it does: This paper proposes a new expression hierarchy of equivariant polynomials based on graphs, provides a complete basis for equivariant polynomials, designs a greedy algorithm to determine the computability of polynomials, and enhances the expressive power of existing GNNs through precomputed polynomial features, ultimately achieving state-of-the-art results in graph isomorphism and molecular property prediction tasks.
ESC: Exploration with Soft Commonsense Constraints for Zero-shot Object Navigation
Kaiwen Zhou (University of California), Xin Eric Wang (University of California)
Object DetectionRobotic IntelligenceTransformerLarge Language ModelVision Language ModelImage
🎯 What it does: This paper proposes the ESC framework, which uses a pre-trained GLIP vision-language model for scene semantic recognition, and infers the common-sense relationships between target objects and rooms, as well as adjacent objects, using large language models (such as DeBERTa). The inference results are then transformed into soft logic constraints in Probabilistic Soft Logic (PSL) to guide Frontier-Based Exploration for zero-experience target object navigation.
Escaping saddle points in zeroth-order optimization: the power of two-point estimators
Zhaolin Ren (Harvard University), Na Li (Harvard University)
Optimization
🎯 What it does: This paper proves that using only two function evaluations (two-point estimation) along with isotropic perturbation in a zero-order gradient descent method can effectively escape saddle points in non-convex functions and converge to approximate second-order saddle points.
Estimating Causal Effects using a Multi-task Deep Ensemble
Ziyang Jiang (Duke University), David Carlson (Duke University)
OptimizationData-Centric LearningImageMultimodality
🎯 What it does: A multi-task deep ensemble framework named CMDE is proposed to learn shared and group-specific information from control and treatment groups, thereby estimating the Conditional Average Treatment Effect (CATE).
Estimating Heterogeneous Treatment Effects: Mutual Information Bounds and Learning Algorithms
Xingzhuo Guo (Tsinghua University), Mingsheng Long (Tsinghua University)
OptimizationRepresentation LearningAdversarial AttackGenerative Adversarial NetworkContrastive LearningTabularBiomedical Data
🎯 What it does: This paper proposes the use of mutual information to select bias and derives a new error bound, subsequently designing the MitNet algorithm to estimate heterogeneous treatment effects.
Estimating Joint Treatment Effects by Combining Multiple Experiments
Yonghan Jung (Purdue University), Elias Bareinboim (Columbia University)
Supervised Fine-TuningReinforcement LearningTabular
🎯 What it does: This paper studies how to use multiple edge experiments (i.e., experiments that randomize only a portion of the treatment variables) to estimate the joint causal effects of multidimensional treatments, proposing identifiable conditions and corresponding estimators.
Estimating Possible Causal Effects with Latent Variables via Adjustment
Tian-Zuo Wang (Nanjing University), Zhi-Hua Zhou (Nanjing University)
Graph
🎯 What it does: In the presence of latent variables, this study investigates how to estimate the set of causal effects that can be adjusted for covariates from a given partially ancestral graph (PAG) and proposes a method to obtain this set without enumerating directed acyclic graphs (DAGs) or mixed ancestral graphs (MAGs).
Estimating the Contamination Factor's Distribution in Unsupervised Anomaly Detection
Lorenzo Perini (KU Leuven), Arto Klami (University of Helsinki)
Anomaly DetectionTabular
🎯 What it does: A method for estimating the posterior distribution of the contamination factor (anomaly ratio) in unsupervised anomaly detection is proposed.
Estimation Beyond Data Reweighting: Kernel Method of Moments
Heiner Kremer (Max Planck Institute for Intelligent Systems), Jia-Jie Zhu (Weierstrass Institute for Applied Analysis and Stochastics)
Optimization
🎯 What it does: This paper proposes the Kernel Method of Moments (KMM), a method based on Maximum Mean Discrepancy (MMD), which breaks through the limitations of traditional methods that only reweight empirical distributions by approximating continuous distributions to satisfy (conditional) moment constraints.
Evaluating Self-Supervised Learning via Risk Decomposition
Yann Dubois (Stanford University), Percy Liang (Stanford University)
Representation LearningContrastive LearningImage
🎯 What it does: Decomposing the risk of SSL, four error components are proposed (approximation, usability, probe generalization, encoder generalization), along with a consistent and efficient estimator;
Evaluating Unsupervised Denoising Requires Unsupervised Metrics
Adria Marcos Morales, Carlos Fernandez-Granda (New York University)
RestorationConvolutional Neural NetworkSupervised Fine-TuningImageVideo
🎯 What it does: Two unsupervised denoising evaluation metrics (uMSE, uPSNR) that can be computed based solely on noisy images are proposed and validated, with experiments conducted on various real and synthetic datasets.
Eventual Discounting Temporal Logic Counterfactual Experience Replay
Cameron Voloshin (California Institute of Technology), Yisong Yue (California Institute of Technology)
Autonomous DrivingReinforcement LearningSequential
🎯 What it does: A value function based on 'final discount' and an LTL counterfactual experience replay method are designed to learn policies that satisfy LTL specifications without the need for sparse rewards.
Everyone's Preference Changes Differently: A Weighted Multi-Interest Model For Retrieval
Hui Shi (University of California San Diego), Jishen Zhao (University of California San Diego)
RetrievalRecommendation SystemRecurrent Neural NetworkTransformerSupervised Fine-TuningTime SeriesSequential
🎯 What it does: A multi-interest preference (MIP) model based on clustering enhancement is proposed, which generates multi-dimensional user embeddings using temporal self-attention and learns interest weights to improve candidate retrieval performance.
Evidential Interactive Learning for Medical Image Captioning
Ervine Zheng (Rochester Institute of Technology), Qi Yu (Rochester Institute of Technology)
GenerationTransformerReinforcement LearningImageTextBiomedical Data
🎯 What it does: An interactive learning framework based on evidence uncertainty estimation is proposed, which continuously updates the model with user feedback to improve the accuracy and reliability of medical image description generation.
Evolving Semantic Prototype Improves Generative Zero-Shot Learning
Shiming Chen (Carnegie Mellon University), Kun Zhang (Mohamed bin Zayed University of Artificial Intelligence)
GenerationData SynthesisConvolutional Neural NetworkGenerative Adversarial NetworkImage
🎯 What it does: A Dynamic Semantic Prototype Evolution (DSP) method is proposed, which gradually aligns predefined semantic prototypes with real prototypes by jointly training a visual-to-semantic mapping network and a prototype evolution network, thereby generating samples that better match real visual features and improving the performance of generative zero-shot learning.
Ewald-based Long-Range Message Passing for Molecular Graphs
Arthur Kosmala (Max Planck Institute for the Science of Light), Stephan Günnemann
Drug DiscoveryGraph Neural NetworkGraphPhysics Related
🎯 What it does: A non-local information transfer framework based on Ewald summation (Ewald message passing) is proposed, which can be seamlessly integrated with existing MPNN models to learn long-range interactions in molecular graphs.
Exact Inference in High-order Structured Prediction
Chuyang Ke (Purdue University), Jean Honorio (Purdue University)
OptimizationGraph
🎯 What it does: This paper proposes an exact inference method for high-order structure prediction tasks, first obtaining the signs of labels through convex constraints of high-order Laplacian tensors, and then determining the final labels using node observations; it also introduces the properties of hypergraph structures based on hyperedge expansion and proves the high-order Cheeger inequality.
Existence and Estimation of Critical Batch Size for Training Generative Adversarial Networks with Two Time-Scale Update Rule
Naoki Sato (Meiji University), Hideaki Iiduka (Meiji University)
GenerationOptimizationGenerative Adversarial NetworkImage
🎯 What it does: This study investigates the theoretical relationship between batch size, number of training steps, and stochastic first-order (SFO) complexity under the two-time scale update rule (TTUR) with a constant learning rate in Generative Adversarial Networks (GANs), and verifies the existence of an optimal batch size.
Expectation-Complete Graph Representations with Homomorphisms
Pascal Welke (University of Bonn), Thomas Gärtner
Graph Neural NetworkGraph
🎯 What it does: An Expectation-Complete graph embedding method is proposed, which obtains graph representations that can be computed in expected polynomial time by randomly sampling the homomorphic counts in the Lovász vector.
Expected Gradients of Maxout Networks and Consequences to Parameter Initialization
Hanna Tseran (Max Planck Institute for Mathematics in the Sciences), Guido Montufar (University of California Los Angeles)
ClassificationOptimizationImage
🎯 What it does: This study investigates the distribution of input and parameter gradients in Maxout networks, deriving upper and lower bounds for the gradient moments and providing a robust initialization strategy.
Expertise Trees Resolve Knowledge Limitations in Collective Decision-Making
Axel Abels (Universite Libre de Bruxelles), Ann Nowe (Vrije Universiteit Brussel)
ClassificationOptimizationReinforcement LearningTabular
🎯 What it does: This paper studies how to utilize local expertise among experts to improve collective decision-making when expert opinions vary with problem instances.
Exphormer: Sparse Transformers for Graphs
Hamed Shirzad (University of British Columbia), Ali Kemal Sinop (Google)
Graph Neural NetworkTransformerGraph
🎯 What it does: This paper proposes a graph Transformer architecture based on sparse attention—EXPHORMER, which combines virtual global nodes, random expander graphs, and local neighborhood attention to achieve parallel processing of global information capture and local detail.
Explainability as statistical inference
Hugo Henri Joseph Senetaire (Technical University of Denmark), Pierre-Alexandre Mattei (Universite Cote d'Azur)
Explainability and InterpretabilityImageBenchmark
🎯 What it does: This paper proposes a framework called LEX that views interpretability as a statistical inference problem, modeling the interpreter as a latent variable and learning both the interpreter and predictor through maximum likelihood.
Explainable Data-Driven Optimization: From Context to Decision and Back Again
Alexandre Forel (Polytechnique Montreal), Thibaut Vidal
OptimizationExplainability and InterpretabilityTabular
🎯 What it does: This study investigates providing interpretable causal explanations for context-driven optimization processes using machine learning predictors, proposing to explain why data-driven decisions outperform traditional approaches by solving for the nearest adversarial context through integer programming.
Explaining Reinforcement Learning with Shapley Values
Daniel Beechey (University of Bath), Özgür Şimşek (University of Bath)
Explainability and InterpretabilityReinforcement Learning
🎯 What it does: A Shapley value-based reinforcement learning explanation framework called SVERL is proposed, and the SVERL-P method is developed to assess the contribution of features to agent performance.
Explaining the effects of non-convergent MCMC in the training of Energy-Based Models
Elisabeth Agoritsas (University of Geneva), Beatriz Seoane (Universidad Complutense de Madrid)
GenerationConvolutional Neural NetworkDiffusion modelImage
🎯 What it does: This study analyzes the impact of using non-convergent (short time, random initial) Markov chain sampling when training energy-based models (EBM), providing a theoretical framework and validating its effects on generation quality and model behavior.
Explore and Exploit the Diverse Knowledge in Model Zoo for Domain Generalization
Yimeng Chen (Chinese Academy of Sciences), Zhi-Ming Ma
Domain AdaptationImageBenchmark
🎯 What it does: This study investigates how to enhance domain generalization performance by leveraging the diversity of different models in a pre-trained model library, proposing analysis and fusion methods for diversity shift and correlation shift.
Exploring Chemical Space with Score-based Out-of-distribution Generation
Seul Lee (KAIST), Sung Ju Hwang (KAIST)
GenerationDrug DiscoveryGraph Neural NetworkDiffusion modelScore-based ModelGraphBiomedical DataStochastic Differential Equation
🎯 What it does: A molecular generation framework called MOOD is proposed, which utilizes a score-based diffusion model to explore out-of-distribution (OOD) beyond the training distribution, and generates molecules that meet multiple target attributes through attribute prediction gradient guidance.
Exploring Model Dynamics for Accumulative Poisoning Discovery
Jianing Zhu (Hong Kong Baptist University), Bo Han (Hong Kong Baptist University)
ClassificationAnomaly DetectionConvolutional Neural NetworkGenerative Adversarial NetworkImage
🎯 What it does: A model dynamic-based Memorization Discrepancy metric is proposed to identify cumulative poisoning samples in real-time data streams, and a Discrepancy-aware Sample Correction (DSC) defense algorithm is designed based on this metric.
Exploring the Benefits of Training Expert Language Models over Instruction Tuning
Joel Jang (Korea Advanced Institute of Science and Technology), Minjoon Seo (Korea Advanced Institute of Science and Technology)
TransformerLarge Language ModelSupervised Fine-TuningMixture of ExpertsTextRetrieval-Augmented Generation
🎯 What it does: This study investigates the generalization ability of Expert Language Models (Expert LM) on unseen tasks after single-task training, comparing it with Multi-Task Instruction Fine-Tuning Models (MT LM).
Exploring the Limits of Model-Targeted Indiscriminate Data Poisoning Attacks
Yiwei Lu (University of Waterloo), Yaoliang Yu (University of Waterloo)
Adversarial AttackConvolutional Neural NetworkImage
🎯 What it does: The research model targets the accessibility of indiscriminate data poisoning attacks and proposes a computable threshold τ, demonstrating its threshold effect across different models and datasets.
Exponential Smoothing for Off-Policy Learning
Imad AOUALI, Anna Korba (Criteo AI Lab)
ClassificationReinforcement Learning from Human FeedbackReinforcement LearningImage
🎯 What it does: An exponential smoothing regularization for Inverse Propensity Score (IPS) is proposed, and based on this, a two-sided PAC-Bayes generalization bound is provided. This bound is directly optimized as the learning objective, addressing the issues of non-differentiability and difficulty in assessing sample complexity associated with traditional hard clipping of IPS.
Extending Conformal Prediction to Hidden Markov Models with Exact Validity via de Finetti's Theorem for Markov Chains
Buddhika Nettasinghe (University of Iowa), Mahantesh M Halappanavar
Time SeriesSequentialFinance Related
🎯 What it does: A general conformal prediction framework is proposed to generate confidence sets for unknown state sequences under hidden Markov models (HMM).
Extending Kernel PCA through Dualization: Sparsity, Robustness and Fast Algorithms
Francesco Tonin (KU Leuven), Johan Suykens
OptimizationTabular
🎯 What it does: By treating KPCA as a differential convex function, a dual form is constructed and the gradient descent/DC algorithm is utilized to address the expensive issues of traditional SVD, achieving robust and sparse feature learning within the same framework.
Extrapolated Random Tree for Regression
Yuchao Cai (Renmin University of China), Hanfang Yang (Renmin University of China)
🎯 What it does: A new tree-based algorithm called Extrapolated Random Tree Regression (ERTR) is proposed, which can adapt to the arbitrary smoothness of regression functions while maintaining the interpretability of trees.
Extrapolative Controlled Sequence Generation via Iterative Refinement
Vishakh Padmakumar (New York University), Ankur P Parikh
GenerationData SynthesisOptimizationTransformerTextSequentialBiomedical Data
🎯 What it does: This paper proposes an Iterative Controlled Extrapolation (ICE) method that achieves attribute value extrapolation generation through gradual local editing on sequences.
Facial Expression Recognition with Adaptive Frame Rate based on Multiple Testing Correction
Andrey Savchenko
RecognitionComputational EfficiencyVideo
🎯 What it does: An adaptive frame rate video emotion recognition framework based on multiple testing correction is proposed, which dynamically decides whether to terminate frame processing using the Bayesian-Hodges program.
FaDIn: Fast Discretized Inference for Hawkes Processes with General Parametric Kernels
Guillaume Staerman (Université Paris-Saclay), Thomas Moreau (Université Paris-Saclay)
Time SeriesSequential
🎯 What it does: This paper proposes a fast discretization inference framework called FaDIn for estimating multivariate Hawkes processes with finite support general parametric kernels.
FAENet: Frame Averaging Equivariant GNN for Materials Modeling
Alexandre AGM Duval, David Rolnick (McGill University)
Graph Neural NetworkGraph
🎯 What it does: In the material modeling task, a general framework is proposed to achieve E(3) equivariance through Stochastic Frame Averaging (SFA), and within this framework, a lightweight, fast, and expressive graph neural network FAENet is designed; this network does not require structural constraints on symmetry and directly utilizes atomic relative position information; SFA enhances the model's symmetry at the data level.
Fair and Accurate Decision Making through Group-Aware Learning
Ramtin Hosseini (University of California San Diego), Pengtao Xie (University of California San Diego)
Domain AdaptationOptimizationNeural Architecture SearchConvolutional Neural NetworkMixture of ExpertsImage
🎯 What it does: This paper proposes a decision-making framework based on Learning by Grouping (LBG), which enhances both fairness and accuracy for multi-class tasks by dividing task samples into several subgroups and training dedicated models for each subgroup.
Fair and Optimal Classification via Post-Processing
Ruicheng Xian (University of Illinois Urbana-Champaign), Han Zhao (University of Illinois Urbana-Champaign)
ClassificationOptimizationTabular
🎯 What it does: In multi-group multi-class classification tasks with noise, the inherent trade-off between demographic parity fairness and error rate is studied, and an optimal post-processing algorithm based on Wasserstein barycenter is proposed.
Fair and Robust Estimation of Heterogeneous Treatment Effects for Policy Learning
Kwangho Kim (Korea University), Jose R Zubizarreta
OptimizationTabular
🎯 What it does: Proposes a non-parametric estimation of heterogeneous treatment effects under fairness constraints and applies it to optimize decision-making strategies.
Fair Densities via Boosting the Sufficient Statistics of Exponential Families
Alexander Soen (Australian National University), Richard Nock (Google Research)
OptimizationTabular
🎯 What it does: A boosting-based preprocessing algorithm FBDE is proposed, which can gradually approach the original data distribution while maintaining fairness.
Fair Neighbor Embedding
Jaakko Peltonen (Tampere University), Jyrki Nummenmaa (Tampere University)
RetrievalOptimizationExplainability and InterpretabilityTabular
🎯 What it does: Two information retrieval-based neighbor embedding methods, Fair-NeRV and Fair-t-NeRV, are proposed for achieving unbiased nonlinear dimensionality reduction and visualization.
Fair yet Asymptotically Equal Collaborative Learning
Xiaoqiang Lin (National University of Singapore), Bryan Kian Hsiang Low (National University of Singapore)
Federated LearningImageTime Series
🎯 What it does: This paper proposes a collaborative learning framework for streaming data, ensuring that nodes receive rewards that match their true contributions by fully exploring and then utilizing during the contribution evaluation phase.
FAIRER: Fairness as Decision Rationale Alignment
Tianlin Li (Nanyang Technological University), Yang Liu (Zhejiang Sci-Tech University)
ClassificationTabular
🎯 What it does: A method called Decision Rationale Alignment is proposed to enhance the fairness of deep networks.
Fairness in Matching under Uncertainty
Siddartha Devic (University of Southern California), Aleksandra Korolova (Princeton University)
Recommendation SystemOptimizationTabular
🎯 What it does: A framework for two-sided matching fairness considering 'uncertainty' is proposed, along with a distributed matching algorithm that is fair and utility-maximizing based on linear programming.
Fairness in Streaming Submodular Maximization over a Matroid Constraint
Marwa El Halabi (Samsung Research America), Jakub Tarnawski (Microsoft Research)
Recommendation SystemOptimizationGraphTabular
🎯 What it does: This paper presents the maximization problem of monotone submodular functions with fairness constraints under matrix constraints in a streaming environment, along with corresponding algorithms and theoretical bounds.
FARE: Provably Fair Representation Learning with Practical Certificates
Nikola Jovanović (ETH Zurich), Martin Vechev (ETH Zurich)
Representation LearningAuto EncoderTabular
🎯 What it does: The FARE method is proposed, which uses a constrained encoder to generate fair representations and provides a confidence upper bound for any subsequent classifier (fairness proof).
Fascinating Supervisory Signals and Where to Find Them: Deep Anomaly Detection with Scale Learning
Hongzuo Xu (National University of Defense Technology), Ning Liu (National University of Defense Technology)
Anomaly DetectionTabular
🎯 What it does: This paper proposes a deep anomaly detection method based on scale learning, called SLAD.
Fast $(1+\varepsilon)$-Approximation Algorithms for Binary Matrix Factorization
Ameya Velingker (Google Research), Samson Zhou (University of California Berkeley and Rice University)
OptimizationTabular
🎯 What it does: An efficient (1 + ε)-approximation algorithm is proposed for the Binary Matrix Factorization (BMF) problem, aiming to approximate the input matrix A as the product of low-rank factors U and V.
Fast Algorithms for Distributed k-Clustering with Outliers
Junyu Huang (Central South University), Jianxin Wang (Central South University)
Anomaly DetectionOptimizationComputational EfficiencyTabular
🎯 What it does: Two sampling algorithms are proposed in distributed k-clustering (including outliers), significantly reducing local running time and eliminating dependence on instance aspect ratio.
Fast as CHITA: Neural Network Pruning with Combinatorial Optimization
Riade Benbaki (Massachusetts Institute of Technology), Rahul Mazumder (Massachusetts Institute of Technology)
OptimizationComputational EfficiencyConvolutional Neural NetworkImage
🎯 What it does: A neural network pruning framework called CHITA is proposed, which transforms the pruning problem into ℓ0 constrained sparse regression and performs iterative hard threshold updates on a low-rank Hessian-free representation, achieving efficient one-time or multi-stage pruning.
Fast Combinatorial Algorithms for Min Max Correlation Clustering
Sami Davies (Northwestern University), Heather Newman (Carnegie Mellon University)
OptimizationGraph
🎯 What it does: A fast combinatorial algorithm for Min Max correlation clustering is proposed, which can achieve a constant factor approximation on complete graphs and run in linear or near-linear time.
Fast Excess Risk Rates via Offset Rademacher Complexity
Chenguang Duan (Wuhan University), Jerry Zhijian Yang (Wuhan University)
Optimization
🎯 What it does: A unified framework is proposed through offset Rademacher complexity, providing a fast convergence rate for the expected excess risk of general Lipschitz continuous loss functions under the absence of Bernstein conditions, covering non-parametric learning for linear, LAD, logistic regression, sparse models, and deep ReLU networks.
Fast Federated Machine Unlearning with Nonlinear Functional Theory
Tianshi Che (Auburn University), Jun Huan (AWS AI Labs)
Federated LearningImage
🎯 What it does: A novel fast federated machine unlearning algorithm (FFMU) based on PCMU random gradient smoothing and the Nemytskii operator is proposed, achieving synchronous training and unlearning.
Fast Inference from Transformers via Speculative Decoding
Yaniv Leviathan (Google Research), Yossi Matias (Google Research)
GenerationComputational EfficiencyTransformerLarge Language ModelText
🎯 What it does: A speculative decoding-based inference acceleration method is proposed and implemented, which improves generation speed by first using a small approximate model to generate multiple token candidates, and then having the target large model verify them in parallel, without changing the model structure or output distribution.
Fast Online Node Labeling for Very Large Graphs
Baojian Zhou (Fudan University), Reza Babanezhad Harikandeh
OptimizationComputational EfficiencyGraph Neural NetworkGraph
🎯 What it does: This paper proposes a fast node label prediction method based on an online relaxation framework (FASTONL), which approximates the inverse of the graph kernel matrix using the local approximation algorithm FIFOPUSH, significantly reducing the time and space complexity of a single prediction while maintaining an effective regret of O(√n^(1+γ)).
Fast Online Value-Maximizing Prediction Sets with Conformal Cost Control
Zhen Lin (University of Illinois), Jimeng Sun (University of Illinois)
Recommendation SystemOptimizationTabularBiomedical DataElectronic Health Records
🎯 What it does: This paper proposes an online multi-label prediction framework called FavMac, which maximizes business value while satisfying user-defined cost constraints.
Fast Private Kernel Density Estimation via Locality Sensitive Quantization
Tal Wagner (Amazon), Nina Mishra (Amazon)
Safty and PrivacyComputational EfficiencyTabular
🎯 What it does: A framework based on Local Sensitivity Quantization (LSQ) is proposed for efficiently achieving differential privacy kernel density estimation (DP-KDE) for high-dimensional data.
Fast Rates for Maximum Entropy Exploration
Daniil Tiapkin (Higher School of Economics), Pierre MENARD
OptimizationReinforcement LearningTabular
🎯 What it does: This paper studies two types of maximum entropy exploration problems in a reward-free environment: Maximum Visit Entropy Exploration (MVEE) and Maximum Trajectory Entropy Exploration (MTEE). For MVEE, we designed EntGame and its regularized version RegEntGame, achieving improved sample complexity; for MTEE, we proposed two algorithms, UCBVI‑Ent and RL‑Explore‑Ent, and demonstrated that they can be used to solve the optimal policy for entropy-regularized MDPs.
Fast Rates in Time-Varying Strongly Monotone Games
Yu-Hu Yan (Nanjing University), Zhi-Hua Zhou (Nanjing University)
OptimizationTime Series
🎯 What it does: A new decentralized online algorithm is proposed for time-varying strongly monotone games, significantly improving existing results and achieving a fast convergence rate that matches the best guarantees in the time-invariant case.
Fast Sampling of Diffusion Models via Operator Learning
Hongkai Zheng (California Institute of Technology), Anima Anandkumar (NVIDIA)
GenerationData SynthesisComputational EfficiencyDiffusion modelImageOrdinary Differential Equation
🎯 What it does: A DSNO network is constructed using neural operators and temporal convolution blocks, allowing for single-step prediction of the probability flow trajectory of the diffusion model, enabling sampling to be completed with a single forward pass of the model.
Fast, Differentiable and Sparse Top-k: a Convex Analysis Perspective
Michael Eli Sander, Mathieu Blondel (Google Research)
OptimizationTransformerMixture of ExpertsImage
🎯 What it does: Proposed differentiable and sparse top-k and top-k-magnitude operators, treating them as linear programming on the permutahedron, achieving seamless differentiability through p-norm regularization;
Faster Gradient-Free Algorithms for Nonsmooth Nonconvex Stochastic Optimization
Lesi Chen (Fudan University), Luo Luo (Fudan University)
OptimizationAdversarial AttackImageTabular
🎯 What it does: A new gradient-free algorithm GFM+ is proposed for solving non-smooth non-convex stochastic optimization problems, and the theoretical complexity of the algorithm achieving (δ,ϵ)-Goldstein stationary points is provided.
Faster Rates of Convergence to Stationary Points in Differentially Private Optimization
Raman Arora (Johns Hopkins University), Enayat Ullah (Johns Hopkins University)
OptimizationSafty and Privacy
🎯 What it does: This paper studies how to quickly approximate the infinite feasible points of Lipschitz and smooth functions under differential privacy constraints, providing new convergence rates for non-convex, convex, and generalized linear models.
Feature Directions Matter: Long-Tailed Learning via Rotated Balanced Representation
Gao Peifeng (University of Chinese Academy of Sciences), Qingming Huang (University of Chinese Academy of Sciences)
ClassificationRepresentation LearningImage
🎯 What it does: This paper proposes a Representation-Balanced Learning Framework (RBL) that learns balanced feature representations in long-tail classification tasks by introducing a learnable orthogonal matrix to rotate a fixed Equiangular Tight Frame (ETF), thereby enhancing the model's generalization performance.
Feature Expansion for Graph Neural Networks
Jiaqi Sun (Tsinghua University), Yujiu Yang (Tsinghua University)
Graph Neural NetworkGraph
🎯 What it does: This paper analyzes graph neural networks from the perspective of feature space, proposing a feature subspace flattening method and providing theoretical and experimental support.
Feature learning in deep classifiers through Intermediate Neural Collapse
Akshay Rangamani (Massachusetts Institute of Technology), tomaso a poggio
ClassificationRepresentation LearningConvolutional Neural NetworkImage
🎯 What it does: This study investigates the feature learning of intermediate layers in deep classifiers and explores whether the phenomenon of 'Neural Collapse' occurs in the intermediate layers during the later stages of training in neural networks.
Feature Programming for Multivariate Time Series Prediction
Alex Daniel Reneau, Han Liu (Northwestern University)
Convolutional Neural NetworkRecurrent Neural NetworkTime SeriesPhysics Related
🎯 What it does: The paper proposes a programmable feature engineering framework for generating high-order features for multivariate time series forecasting.
Featured Graph Coarsening with Similarity Guarantees
Manoj Kumar (Indian Institute of Technology Delhi), Sandeep Kumar (Indian Institute of Technology Delhi)
OptimizationGraph Neural NetworkGraph
🎯 What it does: A graph coarsening method that jointly utilizes the original graph matrix and node features is proposed, learning to obtain a connected coarsened graph and the corresponding feature matrix, while ensuring that the original graph and the coarsened graph are ε-similar in terms of smoothness and spectral structure.
Fed-CBS: A Heterogeneity-Aware Client Sampling Mechanism for Federated Learning via Class-Imbalance Reduction
Jianyi Zhang (Duke University), Hai Li
Federated LearningSafty and PrivacyComputational EfficiencyImage
🎯 What it does: A class-balanced client sampling mechanism called Fed-CBS is proposed to enhance the performance of global models in federated learning under non-IID data.
FedAvg Converges to Zero Training Loss Linearly for Overparameterized Multi-Layer Neural Networks
Bingqing Song (University of Minnesota), Mingyi Hong (University of Minnesota)
OptimizationFederated LearningImage
🎯 What it does: In the framework of federated learning, this study investigates the use of over-parameterized multi-layer neural networks with a wide first layer and hierarchical width, proving that even under highly heterogeneous data distributions, Vanilla FedAvg (Local SGD) can converge to a solution with training error close to zero at a linear rate.
FedBR: Improving Federated Learning on Heterogeneous Data via Local Learning Bias Reduction
Yongxin Guo (Chinese University of Hong Kong), Tao Lin (Westlake University)
Federated LearningConvolutional Neural NetworkTransformerContrastive LearningImage
🎯 What it does: This paper proposes FedBR, a unified federated learning framework that reduces local learning bias by using label-agnostic pseudo-data at each client and combining two regularization steps, thereby improving convergence speed and final accuracy on heterogeneous data.
FedCR: Personalized Federated Learning Based on Across-Client Common Representation with Conditional Mutual Information Regularization
Hao Zhang (Shanghai Jiao Tong University), Hongkai Xiong (Shanghai Jiao Tong University)
Federated LearningMixture of ExpertsImage
🎯 What it does: The FedCR framework is proposed in multi-client personalized federated learning, utilizing cross-client common representation learning and CMI regularization to enhance the generalization performance of local models.
FedDisco: Federated Learning with Discrepancy-Aware Collaboration
Rui Ye (Shanghai Jiao Tong University), Yanfeng Wang (Shanghai AI Laboratory)
Federated LearningImageText
🎯 What it does: In response to the heterogeneity of class distribution in federated learning, a deviation-based aggregation weight method called FedDisco is proposed.
Federated Adversarial Learning: A Framework with Convergence Analysis
Xiaoxiao Li (University of British Columbia), Jiaming Yang (University of Michigan)
OptimizationFederated LearningAdversarial Attack
🎯 What it does: A general federated adversarial learning framework (FAL) is proposed, combining adversarial training with FedAvg, and a global convergence analysis is provided for over-parameterized ReLU networks.
Federated Conformal Predictors for Distributed Uncertainty Quantification
Charles Lu (Massachusetts Institute of Technology), Ramesh Raskar (Massachusetts Institute of Technology)
Federated LearningImageBiomedical Data
🎯 What it does: This study proposes the Federated Conformal Prediction (FCP) framework in a federated learning environment for distributed uncertainty quantification in multi-client heterogeneous data distributions.
Federated Heavy Hitter Recovery under Linear Sketching
Adria Gascon (Google Research), Ananda Theertha Suresh (Google Research)
Federated LearningSafty and PrivacyReinforcement LearningText
🎯 What it does: This study investigates the communication-accuracy trade-off for adversary recovery and approximate histogram in multi-round linear sketches under federated analysis, and presents an optimal protocol.
Federated Linear Contextual Bandits with User-level Differential Privacy
Ruiquan Huang (Pennsylvania State University), Jing Yang (Pennsylvania State University)
OptimizationFederated LearningSafty and Privacy
🎯 What it does: This paper proposes and analyzes a Federated Linear Contextual Bandits framework under user-level differential privacy (user-level DP) constraints, and designs an algorithm named ROBIN based on greedy exploration and private averaging, achieving approximately optimal learning error under central differential privacy (CDP) conditions.