ICLR 2024 Papers — Page 8
International Conference on Learning Representations · 2260 papers
fairret: a Framework for Differentiable Fairness Regularization Terms
Maarten Buyl (Ghent University), Tijl De Bie (Ghent University)
Tabular
🎯 What it does: A differentiable fair regularization framework called FAIRRET is proposed, which can quantify and minimize bias in machine learning models.
FairSeg: A Large-Scale Medical Image Segmentation Dataset for Fairness Learning Using Segment Anything Model with Fair Error-Bound Scaling
Yu Tian (Harvard University), Mengyu Wang (Harvard University)
SegmentationTransformerSupervised Fine-TuningImageBiomedical DataMagnetic Resonance Imaging
🎯 What it does: The first fairness dataset for medical image segmentation, Harvard-FairSeg, is proposed, along with a fairness loss reweighting method based on group error bounds (FEBS) and new equivalence metrics (ES-Dice/ES-IoU) for evaluating and enhancing the fairness of segmentation models.
FairTune: Optimizing Parameter Efficient Fine Tuning for Fairness in Medical Image Analysis
Raman Dutt (University of Edinburgh), Timothy Hospedales (Samsung AI Center)
OptimizationSupervised Fine-TuningImageBiomedical DataMagnetic Resonance Imaging
🎯 What it does: Proposes the FairTune framework, which utilizes parameter-efficient fine-tuning (PEFT) to automatically optimize fairness in medical imaging tasks, addressing the fairness generalization gap.
Faithful and Efficient Explanations for Neural Networks via Neural Tangent Kernel Surrogate Models
Andrew William Engel (Pacific Northwest National Laboratory), Tony Chiang (Pacific Northwest National Laboratory)
Safty and PrivacyExplainability and InterpretabilityComputational EfficiencyConvolutional Neural NetworkTransformerImage
🎯 What it does: This paper proposes and evaluates a series of approximate empirical neural tangent kernel (eNTK) models (such as trNTK, proj‑trNTK, proj‑pNTK) and uses them as kernel functions to construct linear kernel generalized linear models (kGLM) to approximate and explain the decision-making process of deep neural networks.
Faithful Explanations of Black-box NLP Models Using LLM-generated Counterfactuals
Yair Ori Gat (Technion), Roi Reichart (Technion)
Explainability and InterpretabilityTransformerLarge Language ModelPrompt EngineeringContrastive LearningText
🎯 What it does: Two model-agnostic causal explanation methods are proposed, using large language models to generate adversarial texts or learning to match embedding spaces to approximate counterfactuals, thereby assessing the causal impact of NLP models on high-level concepts.
Faithful Rule Extraction for Differentiable Rule Learning Models
Xiaxia Wang (University of Oxford), Ian Horrocks (University of Oxford)
OptimizationExplainability and InterpretabilityGraphBenchmark
🎯 What it does: This paper proposes a method for extracting a sound and complete rule set from differentiable rule learning models (especially DRUM), addressing the lack of formal guarantees in existing methods.
Faithful Vision-Language Interpretation via Concept Bottleneck Models
Songning Lai (King Abdullah University of Science and Technology), Di Wang (King Abdullah University of Science and Technology)
OptimizationExplainability and InterpretabilityTransformerLarge Language ModelVision Language ModelImage
🎯 What it does: This paper proposes a 'Faithful Vision-Language Concept (FVLC)' for the Concept Bottleneck Model (CBM), addressing the instability issues in concept generation and input perturbation without labels.
Fake It Till Make It: Federated Learning with Consensus-Oriented Generation
Rui Ye (Shanghai Jiao Tong University), Siheng Chen (Shanghai Jiao Tong University)
Federated LearningKnowledge DistillationConvolutional Neural NetworkImage
🎯 What it does: This paper proposes FedCOG—a data-correction-based federated learning framework that utilizes a global model to generate complementary data and enhances local model consistency through soft label distillation, thereby alleviating model drift caused by data heterogeneity.
Fantastic Gains and Where to Find Them: On the Existence and Prospect of General Knowledge Transfer between Any Pretrained Model
Karsten Roth (Tübingen AI Center and University of Tübingen), Zeynep Akata (Tübingen AI Center and University of Tübingen)
Knowledge DistillationImage
🎯 What it does: This study explores the feasibility of transferring complementary knowledge between arbitrary pre-trained models and proposes a data partition distillation method based on continual learning to achieve knowledge transfer without performance degradation.
Fantastic Generalization Measures are Nowhere to be Found
Michael Gastpar (École Polytechnique Fédérale de Lausanne), Thomas Weinberger (École Polytechnique Fédérale de Lausanne)
🎯 What it does: The paper rigorously proves through theoretical analysis that in over-parameterized environments, any generalization bounds that rely solely on the training set and the model (as well as those that depend on the algorithm but still maintain distribution independence) cannot achieve uniform tightness, revealing an inherent trade-off between learning performance and estimability.
Fast and unified path gradient estimators for normalizing flows
Lorenz Vaitl (Technical University of Berlin), Pan Kessel (Genentech)
OptimizationFlow-based ModelTabularPhysics Related
🎯 What it does: This paper proposes a unified and fast path gradient estimator to accelerate the training of normalized flows under forward and backward KL objectives.
Fast Ensembling with Diffusion Schrödinger Bridge
Hyunsu Kim (KAIST), Juho Lee (KAIST)
ClassificationComputational EfficiencyKnowledge DistillationDiffusion modelScore-based ModelImage
🎯 What it does: This paper proposes the Diffusion Bridge Network (DBN), which constructs a conditional diffusion bridge between the output of a single model and the output of the entire ensemble model, allowing for the approximation of deep ensemble predictions using a single forward pass and a lightweight scoring network.
Fast Equilibrium of SGD in Generic Situations
Zhiyuan Li (Toyota Technological Institute at Chicago), Zhiren Wang (Pennsylvania State University)
OptimizationConvolutional Neural NetworkImageStochastic Differential Equation
🎯 What it does: This paper proves that under general conditions, stochastic gradient descent (SGD) with weight decay converges rapidly to a local equilibrium in normalized networks, supporting the 'fast equilibrium conjecture.'
Fast Hyperboloid Decision Tree Algorithms
Philippe Chlenski (Columbia University), Itsik Pe'er (Columbia University)
ClassificationOptimizationComputational EfficiencyTabular
🎯 What it does: Two algorithms suitable for hyperbolic space, HYPERDT and HYPERRF, are proposed, implementing node partitioning using geometric inner products on hyperplanes.
Fast Imitation via Behavior Foundation Models
Matteo Pirotta (Meta), Yann Ollivier (Meta)
Computational EfficiencyKnowledge DistillationRobotic IntelligenceReinforcement Learning from Human FeedbackReinforcement LearningSequential
🎯 What it does: This paper proposes using the Forward-Backward (FB) framework in the Behavior Foundation Model (BFM) to quickly generate imitation strategies from a small amount of expert demonstration data, without the need for additional reinforcement learning or fine-tuning during testing.
Fast Updating Truncated SVD for Representation Learning with Sparse Matrices
Haoran Deng (Zhejiang University), Shiliang Pu (Hikvision Research Institute)
Recommendation SystemRepresentation LearningGraphTabular
🎯 What it does: An algorithm for rapidly updating sparse matrix truncated singular value decomposition (truncated SVD) has been developed, supporting incremental updates by row/column/weight while maintaining high accuracy.
Fast Value Tracking for Deep Reinforcement Learning
Frank Shih (Purdue University), Faming Liang (Purdue University)
Reinforcement LearningTabularStochastic Differential Equation
🎯 What it does: A deep reinforcement learning sampling algorithm based on Kalman filtering and Langevin dynamics—LKTD—is proposed for efficiently sampling from the posterior distribution of value function parameters.
Fast-DetectGPT: Efficient Zero-Shot Detection of Machine-Generated Text via Conditional Probability Curvature
Guangsheng Bao (Westlake University), Yue Zhang (Westlake University)
ClassificationComputational EfficiencyTransformerLarge Language ModelText
🎯 What it does: A zero-shot detector named Fast-DetectGPT is proposed for efficiently distinguishing between machine-generated text and human-written text.
Fast-ELECTRA for Efficient Pre-training
Chengyu Dong (University of California), Xiaodong Liu (Microsoft Research)
OptimizationComputational EfficiencyTransformerLarge Language ModelTextBenchmark
🎯 What it does: This paper proposes Fast-ELECTRA, an efficient pre-training method that utilizes existing language models as auxiliary models and constructs learning curves through temperature scaling.
Fast, Expressive $\mathrm{SE}(n)$ Equivariant Networks through Weight-Sharing in Position-Orientation Space
Erik J Bekkers, David W. Romero (Vrije Universiteit Amsterdam)
Computational EfficiencyDrug DiscoveryConvolutional Neural NetworkDiffusion modelPoint Cloud
🎯 What it does: This paper proposes a weight-sharing SE(n) equivariant convolutional network on homotopy spaces, constructing a scalable 3D point cloud processing architecture called P Θ NITA.
Faster Approximation of Probabilistic and Distributional Values via Least Squares
Weida Li, Yaoliang Yu (University of Waterloo)
OptimizationComputational EfficiencySupervised Fine-TuningTabular
🎯 What it does: A general estimator GELS based on least squares and its variants are proposed for quickly approximating all probability values (including Shapley, Banzhaf, etc.) and distributed values, along with theoretical convergence analysis and an unsupervised training framework TrELS.
Faster Sampling from Log-Concave Densities over Polytopes via Efficient Linear Solvers
Oren Mangoubi (Worcester Polytechnic Institute), Nisheeth K. Vishnoi (Yale University)
OptimizationComputational Efficiency
🎯 What it does: This paper studies the problem of sampling from log-concave distributions under polyhedral constraints and proposes an efficient Markov chain algorithm to achieve this goal.
FasterViT: Fast Vision Transformers with Hierarchical Attention
Ali Hatamizadeh (NVIDIA), Pavlo Molchanov (NVIDIA)
ClassificationObject DetectionSegmentationConvolutional Neural NetworkTransformerImage
🎯 What it does: Designed and implemented FasterViT, a hybrid network that integrates CNN and Vision Transformer, aimed at achieving high throughput and high accuracy for high-resolution images; and proposed a Hierarchical Attention (HAT) module to efficiently capture global and local dependencies.
FeatUp: A Model-Agnostic Framework for Features at Any Resolution
Stephanie Fu, William T. Freeman (Massachusetts Institute of Technology Google)
SegmentationDepth EstimationTransformerNeural Radiance FieldImage
🎯 What it does: A model-agnostic feature upsampling framework called FeatUp is proposed, which utilizes multi-view consistency loss to recover high-resolution information from low-resolution deep features, enhancing the performance of downstream dense prediction tasks without altering semantics.
Feature Collapse
Thomas Laurent (Loyola Marymount University), Xavier Bresson (National University of Singapore)
ClassificationExplainability and InterpretabilityRepresentation LearningRecurrent Neural NetworkText
🎯 What it does: This study investigates how early layers of the network assign the same features to words with the same roles (i.e., feature collapse) in the synthetic sentence classification task, and demonstrates that this collapse can achieve interpretable and generalizable representations in the limit of large samples.
Feature emergence via margin maximization: case studies in algebraic tasks
Depen Morwani (Harvard University), Sham M. Kakade
🎯 What it does: This paper studies the features learned by sufficiently wide single hidden layer neural networks under minimal regularization in algebraic tasks such as modular addition, sparse parity, and finite group operations, and proves that they are actually Fourier features or irreducible representations.
Feature-aligned N-BEATS with Sinkhorn divergence
Joonhun Lee (Seoul National University), Kyunghyun Park (Nanyang Technological University)
Time SeriesFinance Related
🎯 What it does: Proposes the Feature-aligned N-BEATS model, which achieves cross-domain time series forecasting through stacked hierarchical feature alignment;
FedCDA: Federated Learning with Cross-rounds Divergence-aware Aggregation
Haozhao Wang (Nanyang Technological University), Tianwei Zhang (Nanyang Technological University)
OptimizationFederated LearningConvolutional Neural NetworkImage
🎯 What it does: A new federated learning aggregation method called FedCDA is proposed, which can select the local models with the least divergence for aggregation across different communication rounds, thereby improving the performance of the global model on non-IID data.
FedCompass: Efficient Cross-Silo Federated Learning on Heterogeneous Client Devices Using a Computing Power-Aware Scheduler
Zilinghan Li (University of Illinois at Urbana-Champaign), Ravi Madduri (Argonne National Laboratory)
Federated LearningComputational EfficiencyImage
🎯 What it does: We propose FedCompass, a semi-asynchronous federated learning framework that utilizes a computation power-aware scheduler to enable different clients to nearly synchronize their local training within a group, reducing model staleness and improving efficiency.
FedDA: Faster Adaptive Gradient Methods for Federated Constrained Optimization
Junyi Li (University of Maryland), Heng Huang (University of Maryland)
OptimizationFederated LearningTabular
🎯 What it does: This paper proposes the FedDA (Fast Adaptive Dual-Averaging) framework to address constrained optimization problems in federated learning, and provides its achievable sample complexity and communication complexity.
Federated Causal Discovery from Heterogeneous Data
Loka Li (Mohamed bin Zayed University of Artificial Intelligence), Kun Zhang (Carnegie Mellon University)
Federated LearningSafty and PrivacyTabularBiomedical DataMagnetic Resonance ImagingFinance Related
🎯 What it does: A constraint-based federated causal discovery method, FedCDH, is proposed, which can learn causal structures through summary statistics in decentralized and heterogeneous data environments.
Federated Orthogonal Training: Mitigating Global Catastrophic Forgetting in Continual Federated Learning
Yavuz Faruk Bakman (University of Southern California), Salman Avestimehr (University of Southern California)
Federated LearningSafty and PrivacyConvolutional Neural NetworkImage
🎯 What it does: A Federated Orthogonal Training (FOT) framework is proposed to mitigate global catastrophic forgetting in continual federated learning.
Federated Q-Learning: Linear Regret Speedup with Low Communication Cost
Zhong Zheng (Pennsylvania State University), Jing Yang (Pennsylvania State University)
Federated LearningReinforcement Learning
🎯 What it does: This paper proposes two model-free Q-learning algorithms for federated reinforcement learning (FedQ-Hoeffding and FedQ-Bernstein), achieving linear regret acceleration in tabular, finite-horizon MDP scenarios of multi-agent collaborative learning while maintaining logarithmic control over communication costs.
Federated Recommendation with Additive Personalization
Zhiwei Li (Australian AI Institute), Tianyi Zhou (University of Maryland)
Recommendation SystemFederated LearningSafty and PrivacyTabular
🎯 What it does: The FedRAP method is proposed, which achieves bidirectional personalized recommendation in federated learning through global sparse item embeddings and local personalized embeddings, while only uploading the sparse global matrix to reduce communication and privacy leakage risks.
Federated Text-driven Prompt Generation for Vision-Language Models
Chen Qiu (Bosch Center for AI), Wan-Yi Lin (Bosch Center for AI)
ClassificationFederated LearningTransformerPrompt EngineeringVision Language ModelImage
🎯 What it does: In the context of federated learning, a text-driven prompt generation method called FedTPG is proposed, which utilizes the pre-trained CLIP model to map task-related textual information into context-aware soft prompt vectors, enabling joint training across clients and tasks.
Federated Wasserstein Distance
Alain Rakotomamonjy (Criteo AI Lab), Liva Ralaivola (Criteo AI Lab)
Federated LearningComputational EfficiencyImage
🎯 What it does: The FedWaD algorithm is proposed to compute the Wasserstein distance between two distributions in a federated environment without sharing samples.
FedHyper: A Universal and Robust Learning Rate Scheduler for Federated Learning with Hypergradient Descent
Ziyao Wang (University of Maryland), Ang Li (University of Maryland)
OptimizationFederated LearningImage
🎯 What it does: This paper proposes FEDHYPER, a general robust learning rate scheduler based on supergradient descent, for adaptive adjustment of global and local learning rates in federated learning.
FedImpro: Measuring and Improving Client Update in Federated Learning
Zhenheng Tang (Hong Kong Baptist University), Xiaowen Chu (Hong Kong University of Science and Technology)
Federated LearningImage
🎯 What it does: This study investigates the client drift problem in federated learning and proposes to enhance the generalization performance of the global model by decoupling the network and sharing similar feature distributions.
FedInverse: Evaluating Privacy Leakage in Federated Learning
Di Wu (University of Southern Queensland), Atul Sajjanhar (Swinburne University of Technology)
Federated LearningSafty and PrivacyAdversarial AttackConvolutional Neural NetworkGenerative Adversarial NetworkImage
🎯 What it does: Proposes the FedInverse framework to assess the risk of model inversion attacks (MI) on privacy leakage in federated learning systems.
FedLoGe: Joint Local and Generic Federated Learning under Long-tailed Data
Zikai Xiao (Zhejiang University), Zuozhu Liu (Zhejiang University)
Federated LearningImage
🎯 What it does: This paper proposes the FedLoGe framework, which simultaneously improves the performance of the global model and individual client models in federated long-tail learning.
FedP3: Federated Personalized and Privacy-friendly Network Pruning under Model Heterogeneity
Kai Yi (King Abdullah University of Science and Technology), Lingjuan Lyu (Sony Artificial Intelligence)
Federated LearningSafty and PrivacyConvolutional Neural NetworkImage
🎯 What it does: The FedP3 algorithm is proposed in the context of federated learning, addressing model heterogeneity through a two-level pruning strategy of global and local pruning to achieve personalized networks for each client, while only sending selected layer parameters to the server, enhancing privacy protection and communication efficiency.
FedTrans: Client-Transparent Utility Estimation for Robust Federated Learning
Mingkun Yang (Delft University of Technology), Jie Yang (Delft University of Technology)
Federated LearningConvolutional Neural NetworkImage
🎯 What it does: Proposes the FedTrans framework, which utilizes server-side auxiliary data to estimate the utility of client updates through a Bayesian model and variational inference, achieving transparent client selection under noise and heterogeneous environments, thereby enhancing the robustness of federated learning.
FedWon: Triumphing Multi-domain Federated Learning Without Normalization
Weiming Zhuang (Sony AI), Lingjuan Lyu (Sony AI)
Domain AdaptationFederated LearningConvolutional Neural NetworkImage
🎯 What it does: A model called FedWon is proposed, which removes all normalization layers in multi-domain federated learning and re-parameterizes using Weighted Standardized Convolution (WSConv).
Ferret: Refer and Ground Anything Anywhere at Any Granularity
Haoxuan You (Columbia University), Yinfei Yang (Apple)
ClassificationObject DetectionSegmentationRetrievalTransformerLarge Language ModelVision Language ModelImageTextMultimodality
🎯 What it does: Ferret, a multimodal large language model, has been designed to understand references for arbitrary shapes and locate them semantically in images.
Few-Shot Detection of Machine-Generated Text using Style Representations
Rafael Alberto Rivera Soto (Lawrence Livermore National Laboratory), Nicholas Andrews (Johns Hopkins University)
ClassificationObject DetectionMeta LearningTransformerLarge Language ModelContrastive LearningText
🎯 What it does: Proposes a few-shot machine-generated text detection method based on writing style representation.
Few-shot Hybrid Domain Adaptation of Image Generator
Hengjia Li (State Key Lab of CAD and CG, Zhejiang University), Xiaofei He (State Key Lab of CAD and CG, Zhejiang University)
GenerationData SynthesisDomain AdaptationGenerative Adversarial NetworkImage
🎯 What it does: Proposes the task of few-shot hybrid domain adaptation (HDA), training a pre-trained generator to synthesize images that fuse multi-domain attributes.
FFB: A Fair Fairness Benchmark for In-Processing Group Fairness Methods
Xiaotian Han (Texas A&M University), Xia Hu (Rice University)
ClassificationOptimizationTabularBenchmark
🎯 What it does: This paper proposes and implements a standardized benchmark framework called Fair Fairness Benchmark (FFB) for evaluating and comparing in-processing group fairness methods for binary classification tasks.
Fiber Monte Carlo
Nick Richardson (Princeton University), Ryan P Adams
OptimizationImagePoint Cloud
🎯 What it does: A Fiber Monte Carlo method based on line segment sampling is proposed, which can achieve differentiable estimates in integrals with parameter discontinuities, suitable for tasks such as rendering, topology optimization, and convex hulls.
Fine-Tuned Language Models Generate Stable Inorganic Materials as Text
Nate Gruver (New York University), Zachary Ward Ulissi
GenerationTransformerLarge Language ModelSupervised Fine-TuningPrompt EngineeringTextPhysics Related
🎯 What it does: Fine-tuning the text representation of crystal structures using pre-trained large language models to generate stable inorganic materials that meet physical constraints.
Fine-tuning Aligned Language Models Compromises Safety, Even When Users Do Not Intend To!
Xiangyu Qi (Princeton University), Peter Henderson (Princeton University)
Safty and PrivacyTransformerLarge Language ModelSupervised Fine-TuningReinforcement LearningText
🎯 What it does: This study investigates how fine-tuning alignment in large language models can lead to a decrease in safety, demonstrating the impact of both malicious and benign fine-tuning on safety.
Fine-Tuning Enhances Existing Mechanisms: A Case Study on Entity Tracking
Nikhil Prakash (Northeastern University), David Bau (Northeastern University)
Object TrackingTransformerLarge Language ModelSupervised Fine-TuningText
🎯 What it does: This study investigates the impact of fine-tuning on the internal mechanisms of large language models, using entity tracking as a case study, revealing how the model retains its original circuitry while achieving higher performance through enhancement after fine-tuning.
Fine-Tuning Language Models for Factuality
Katherine Tian (Stanford University), Chelsea Finn (Stanford University)
GenerationOptimizationTransformerLarge Language ModelSupervised Fine-TuningReinforcement LearningText
🎯 What it does: This paper proposes a language model fine-tuning method for factuality that does not require manual annotation, utilizing automatically generated factual preference data for Direct Preference Optimization (DPO) fine-tuning.
Fine-tuning Multimodal LLMs to Follow Zero-shot Demonstrative Instructions
Juncheng Li (Zhejiang University), Yueting Zhuang (Zhejiang University)
RecognitionGenerationData SynthesisTransformerLarge Language ModelVision Language ModelDiffusion modelImageTextMultimodalityBenchmark
🎯 What it does: A lightweight visual prompt generator completion module VPG-C is proposed, which enhances the understanding ability of multimodal large language models on unsupervised demonstration instructions using a synthetic discriminative training strategy; simultaneously, a DEMON benchmark is constructed to evaluate the model's performance in following multimodal interactive instructions.
Finetuning Text-to-Image Diffusion Models for Fairness
Xudong Shen (National University of Singapore), Mohan Kankanhalli (National University of Singapore)
GenerationData SynthesisSupervised Fine-TuningDiffusion modelImageText
🎯 What it does: This paper fine-tunes a text-to-image diffusion model to reduce gender, racial, and intersectional biases in generated images.
Finite Scalar Quantization: VQ-VAE Made Simple
Fabian Mentzer (Google Research), Michael Tschannen (Google DeepMind)
SegmentationGenerationDepth EstimationTransformerAuto EncoderImage
🎯 What it does: Replace the vector quantization (VQ) in VQ-VAE with a simple finite scalar quantization (FSQ) and validate its effectiveness on two large-scale models, MaskGIT and UViM.
Finite-State Autoregressive Entropy Coding for Efficient Learned Lossless Compression
Yufeng Zhang (Shanghai Jiao Tong University), Weiyao Lin (Ant Group)
CompressionImage
🎯 What it does: A lossless compression framework that combines Finite State Autoregressive (FSAR) priors and Straight-Through Hard Quantization (STHQ) is proposed;
Finite-Time Analysis of On-Policy Heterogeneous Federated Reinforcement Learning
Chenyu Zhang (Columbia University), James Anderson (Columbia University)
Federated LearningReinforcement LearningTabular
🎯 What it does: A new federated reinforcement learning algorithm called FedSARSA is proposed, which combines the on-policy TD control of SARSA with multi-client collaboration in federated learning, and provides a complete finite-time convergence analysis.
First-order ANIL provably learns representations despite overparametrisation
Oğuz Kaan Yüksel (École Polytechnique Fédérale de Lausanne), Nicolas Flammarion (École Polytechnique Fédérale de Lausanne)
Representation LearningMeta LearningTabular
🎯 What it does: This paper studies the pre-training behavior of first-order ANIL (FO-ANIL) in a multi-task linear regression model with shared linear representations. It proves that under the idealization of infinite tasks, it can learn low-dimensional shared representations even in cases of excessive width (over-parameterization) and can 'forget' the orthogonal space to achieve good performance on new tasks with just one gradient step.
FITS: Modeling Time Series with $10k$ Parameters
Zhijian Xu (Chinese University of Hong Kong), Qiang Xu (Chinese University of Hong Kong)
Anomaly DetectionOptimizationTime Series
🎯 What it does: This paper proposes a lightweight temporal model FITS, which treats time series forecasting and anomaly detection as a frequency domain interpolation problem. It utilizes a single-layer complex linear layer to achieve amplitude scaling and phase shifting, thereby completing long-term and short-term forecasting and self-supervised reconstruction with only about 10k parameters.
Fixed Non-negative Orthogonal Classifier: Inducing Zero-mean Neural Collapse with Feature Dimension Separation
Hoyong Kim (Gwangju Institute of Science and Technology), Kangil Kim (Gwangju Institute of Science and Technology)
ClassificationImage
🎯 What it does: This paper proposes a Fixed Non-negative Orthogonal Classifier (FNO), theoretically proving that it can achieve zero-mean neural collapse and attain global optimality in the Layer Peeling Model (LPM); it enhances Masked Softmax in continual learning and Arc-Mixup in imbalanced learning through the feature dimension separation (FDS) generated by FNO, resulting in significant performance improvements across various benchmark datasets.
Fixed-Budget Differentially Private Best Arm Identification
Zhirui Chen (National University of Singapore), Vincent Tan
OptimizationSafty and PrivacyTabular
🎯 What it does: This study investigates the differential privacy optimal arm identification (DP-BAI) algorithm under a fixed budget in linear multi-armed bandits, providing upper and lower bounds for the error probability.
Flag Aggregator: Scalable Distributed Training under Failures and Augmented Losses using Convex Optimization
Hamidreza Almasi (University of Illinois), Sathya N. Ravi (University of Illinois)
OptimizationFederated LearningConvolutional Neural NetworkSupervised Fine-TuningImage
🎯 What it does: A robust gradient aggregator based on subspace estimation (Flag Aggregator, FA) is proposed, which treats aggregation as maximum likelihood estimation and incorporates regularization to resist Byzantine failures and noise from data augmentation.
FlashAttention-2: Faster Attention with Better Parallelism and Work Partitioning
Tri Dao (Princeton University)
OptimizationComputational EfficiencyTransformerLarge Language ModelSequential
🎯 What it does: Improved the original FlashAttention and proposed FlashAttention-2, which adopts a better workload partitioning and parallel strategy to achieve faster and more efficient attention computation for long sequence Transformers.
FlashFFTConv: Efficient Convolutions for Long Sequences with Tensor Cores
Daniel Y Fu, Christopher Re (Stanford University)
Computational EfficiencyMultimodalitySequential
🎯 What it does: Designed and implemented FLASHFFTCONV, an FFT convolution optimization system for long sequences, which shortens computation time and reduces memory usage by utilizing Monarch decomposition, Tensor Core computation, and Kernel Fusion.
FLASK: Fine-grained Language Model Evaluation based on Alignment Skill Sets
Seonghyeon Ye (Korean Advanced Institute of Science and Technology), Minjoon Seo (Korean Advanced Institute of Science and Technology)
TransformerLarge Language ModelText
🎯 What it does: This paper presents FLASK—a fine-grained language model evaluation framework based on aligned skill sets. It first annotates the required skills, domains, and difficulty for each instruction, and then allows humans or LLMs to score from 1 to 5, supporting both standard and FLASK-HARD evaluation modes.
Flat Minima in Linear Estimation and an Extended Gauss Markov Theorem
Simon Segert (Princeton University)
Optimization
🎯 What it does: This paper proposes a linear estimation framework where the bias operator can be non-zero but is constrained by the Schatten norm, deriving the corresponding optimal estimator and providing closed-form expressions for the nuclear norm, spectral norm, and Frobenius norm (i.e., ridge regression);
FLATTEN: optical FLow-guided ATTENtion for consistent text-to-video editing
Yuren Cong (Leibniz University Hannover), Sen He (Meta AI)
GenerationData SynthesisDiffusion modelOptical FlowVideo
🎯 What it does: A text-to-video editing framework called FLATTEN is proposed, which maintains visual consistency during video editing by utilizing optical flow-guided attention.
FLD: Fourier Latent Dynamics for Structured Motion Representation and Learning
Chenhao Li (Massachusetts Institute of Technology), Sang bae Kim
Robotic IntelligenceReinforcement LearningAuto EncoderVideo
🎯 What it does: A self-supervised structured motion representation and generation method based on the Fourier frequency domain is proposed—Fourier Latent Dynamics (FLD), which can capture the spatiotemporal relationships of periodic/quasi-periodic motions and implement online tracking and fallback mechanisms in motion learning control.
Flow Matching on General Geometries
Ricky T. Q. Chen (Meta), Yaron Lipman (Meta)
Score-based ModelFlow-based ModelMeshGraph
🎯 What it does: This paper proposes Riemannian Flow Matching (RFM), a framework for training continuous normalizing flows on Riemannian manifolds;
Flow to Better: Offline Preference-based Reinforcement Learning via Preferred Trajectory Generation
Zhilong Zhang (Nanjing University), Yang Yu (Nanjing University)
Reinforcement LearningDiffusion modelTabularSequential
🎯 What it does: The Flow-to-Better (FTB) framework is proposed, which improves trajectories directly at the trajectory level using diffusion models, avoiding the pitfalls of traditional reward learning and TD training.
Follow-Up Differential Descriptions: Language Models Resolve Ambiguities for Image Classification
Reza Esfandiarpoor (Brown University), Stephen Bach
ClassificationTransformerLarge Language ModelVision Language ModelImage
🎯 What it does: A zero-shot method called FuDD is proposed, which enhances the performance of visual-language models like CLIP in image classification by detecting ambiguous categories in images and using large language models to generate differentiated class descriptions.
Forward $\chi^2$ Divergence Based Variational Importance Sampling
Chengrui Li (Georgia Institute of Technology), Anqi Wu (Georgia Institute of Technology)
Tabular
🎯 What it does: This paper proposes a new Variational Importance Sampling (VIS) method that directly estimates and maximizes the marginal log-likelihood of latent variable models using importance sampling, thereby achieving parameter learning.
Forward Learning of Graph Neural Networks
Namyong Park (Meta AI), Nesreen K. Ahmed
Graph Neural NetworkContrastive LearningGraph
🎯 What it does: A forward learning framework called FORWARDGNN is proposed, which trains graph neural networks using unidirectional forward propagation, avoiding the limitations of backpropagation.
Forward Learning with Top-Down Feedback: Empirical and Analytical Characterization
Ravi Francesco Srinivasan (IBM Research Europe), Giorgia Dellaferrera (Institute of Neuroinformatics University of Zurich and ETH Zurich)
ClassificationOptimizationConvolutional Neural NetworkImageOrdinary Differential Equation
🎯 What it does: Research and improve the 'Forward Learning' algorithms (Forward-Forward and PEPITA), providing their theoretical dynamics, experimental validation, and demonstrating their feasibility in deeper networks.
FOSI: Hybrid First and Second Order Optimization
Hadar Sivan (Technion), Assaf Schuster (Technion)
OptimizationConvolutional Neural NetworkRecurrent Neural NetworkAuto EncoderImageTextAudio
🎯 What it does: This paper proposes a meta-optimizer FOSI, which enhances the convergence speed of first-order optimizers by splitting the objective function in orthogonal subspaces during each iteration, accelerating in one subspace using Newton's method and optimizing in another subspace with a baseline first-order optimizer.
Foundation Model-oriented Robustness: Robust Image Model Evaluation with Pretrained Models
Peiyan Zhang (Hong Kong University of Science and Technology), Haohan Wang (University of Illinois at Urbana-Champaign)
ClassificationAdversarial AttackTransformerGenerative Adversarial NetworkImage
🎯 What it does: This paper proposes a robustness evaluation method based on Foundation Models, which measures the robustness of image classification models by generating adversarial samples that maintain label consistency with the original images but are significantly different.
Fourier Transporter: Bi-Equivariant Robotic Manipulation in 3D
Haojie Huang (Northeastern University), Robin Walters (Northeastern University)
Robotic IntelligenceConvolutional Neural NetworkReinforcement LearningPoint CloudBenchmark
🎯 What it does: A novel end-to-end behavior cloning model named FOURTRAN is proposed to address 2D and 3D pick-place robotic manipulation tasks, enhancing sampling efficiency through SE(3) equivariance.
Free from Bellman Completeness: Trajectory Stitching via Model-based Return-conditioned Supervised Learning
Zhaoyi Zhou (Tsinghua University), Simon Shaolei Du
Robotic IntelligenceTransformerSupervised Fine-TuningReinforcement LearningSequential
🎯 What it does: A model-based reward conditional supervised learning method (MBRCSL) is proposed, which solves the problem of trajectory stitching and the need for Bellman completeness in traditional RCSL by learning a dynamics model and behavior policy for forward sampling.
FreeDyG: Frequency Enhanced Continuous-Time Dynamic Graph Model for Link Prediction
Yuxing Tian (IDEA Research, International Digital Economy Academy), Fan Guo (Jiangxi Normal University)
Recommendation SystemOptimizationGraph Neural NetworkGraphTime SeriesOrdinary Differential Equation
🎯 What it does: The FreeDyG model is proposed for link prediction in continuous-time dynamic graphs.
FreeNoise: Tuning-Free Longer Video Diffusion via Noise Rescheduling
Haonan Qiu (Nanyang Technological University), Ziwei Liu (Nanyang Technological University)
GenerationData SynthesisTransformerDiffusion modelVideo
🎯 What it does: The FreeNoise method is proposed, which utilizes noise rescheduling (local noise shuffling + window-based attention fusion) to achieve untuned long video generation, and introduces Motion Injection to support continuous action generation with multiple text conditions.
FreeReg: Image-to-Point Cloud Registration Leveraging Pretrained Diffusion Models and Monocular Depth Estimators
Haiping Wang (Wuhan University), Bisheng Yang (Wuhan University)
Object DetectionDepth EstimationAutonomous DrivingDiffusion modelImagePoint Cloud
🎯 What it does: The FreeReg method is proposed, which achieves unsupervised image-to-point cloud registration by unifying image and point cloud modalities through a pre-trained diffusion model and a monocular depth estimator.
Frequency-Aware Transformer for Learned Image Compression
Han Li (Shanghai Jiao Tong University), Hongkai Xiong (Shanghai Jiao Tong University)
CompressionTransformerImage
🎯 What it does: A frequency-aware Transformer is proposed for end-to-end learning of image compression.
From Bricks to Bridges: Product of Invariances to Enhance Latent Space Communication
Irene Cannistraci (Sapienza University of Rome), Emanuele Rodolà (Sapienza University of Rome)
Data SynthesisRepresentation LearningGraph Neural NetworkContrastive LearningImageTextMultimodalityGraph
🎯 What it does: Combining multiple invariances into the representation, a product space is constructed to enhance communication between the latent spaces of different models, achieving zero-shot stitching in a multimodal context.
From Graphs to Hypergraphs: Hypergraph Projection and its Reconstruction
Yanbang Wang (Cornell University), Jon Kleinberg (Cornell University)
OptimizationGraph Neural NetworkGraph
🎯 What it does: This paper studies the theoretical flaws of graph-to-hypergraph projection and the inverse reconstruction problem, proposing a learning-based hypergraph reconstruction framework called SHyRe, and provides a complete algorithmic process and theoretical analysis.
From Latent Graph to Latent Topology Inference: Differentiable Cell Complex Module
Claudio Battiloro (Sapienza University of Rome), Paolo Di Lorenzo (Sapienza University of Rome)
ClassificationGraph Neural NetworkGraph
🎯 What it does: This paper proposes a Differentiable Cell Complex Module (DCM) that enables adaptive inference of high-order cell complex topology from unlabeled point clouds or incomplete graphs for downstream tasks in graph neural networks.
From Molecules to Materials: Pre-training Large Generalizable Models for Atomic Property Prediction
Nima Shoghi (Meta), Brandon M Wood
Graph Neural NetworkSupervised Fine-TuningTabular
🎯 What it does: Proposes a Joint Multi-domain Pre-training (JMP) method that utilizes multi-task supervised pre-training to learn universal atomic-level representations across various chemical domains (molecules, catalysts, materials, etc.) and achieves multi-task performance improvement through fine-tuning.
From Posterior Sampling to Meaningful Diversity in Image Restoration
Noa Cohen (Technion Israel Institute of Technology), Tomer Michaeli (Technion Israel Institute of Technology)
RestorationSuper ResolutionDiffusion modelImage
🎯 What it does: This paper proposes a method for generating meaningfully diverse reconstruction results in image restoration tasks, surpassing traditional posterior sampling.
From Sparse to Soft Mixtures of Experts
Joan Puigcerver (Google DeepMind), Neil Houlsby (Google DeepMind)
TransformerMixture of ExpertsImage
🎯 What it does: A fully differentiable sparse Transformer called Soft MoE is proposed, which allows each expert to handle only a portion of the weighted average tokens through soft allocation, achieving a high-capacity model with no significant additional computational cost.
From Zero to Turbulence: Generative Modeling for 3D Flow Simulation
Marten Lienen (Technical University of Munich), Stephan Günnemann (Technical University of Munich)
GenerationData SynthesisTransformerDiffusion model
🎯 What it does: This paper proposes using a 3D diffusion generative model to directly learn the distribution of turbulent states, generating high-quality turbulent samples without the need for an initial turbulent state.
FROSTER: Frozen CLIP is A Strong Teacher for Open-Vocabulary Action Recognition
Xiaohu Huang (University of Hong Kong), Kai Han (University of Hong Kong)
RecognitionKnowledge DistillationTransformerPrompt EngineeringVision Language ModelVideoText
🎯 What it does: The FROSTER framework is proposed, using a frozen CLIP model as a teacher to achieve joint learning of video-specific features and general features through residual feature distillation, thereby completing the open vocabulary action recognition task.
Frozen Transformers in Language Models Are Effective Visual Encoder Layers
Ziqi Pang (University of Illinois Urbana-Champaign), Yu-Xiong Wang (University of Illinois Urbana-Champaign)
RecognitionObject DetectionSegmentationAutonomous DrivingTransformerLarge Language ModelImageVideoMultimodalityPoint Cloud
🎯 What it does: Inserting a frozen Transformer block from a pre-trained large language model (LLM) into a visual encoder as a general visual encoding layer to enhance the performance of various visual tasks.
Fully Hyperbolic Convolutional Neural Networks for Computer Vision
Ahmad Bdeir (University of Hildesheim), Niels Landwehr (University of Hildesheim)
ClassificationGenerationAdversarial AttackConvolutional Neural NetworkAuto EncoderImage
🎯 What it does: A fully hypergeometric convolutional neural network (HCNN) is proposed, which performs convolution, batch normalization, and polynomial logistic regression entirely within the Lorentz model of Riemannian manifolds, achieving a full-link hypergeometric representation from input to output.
Function Vectors in Large Language Models
Eric Todd (Northeastern University), David Bau (Northeastern University)
TransformerLarge Language ModelText
🎯 What it does: The study found that there are 'Function Vectors' (FV) in autoregressive transformer language models, which can be obtained by identifying key attention heads and summing them. This vector can induce the model to perform specific tasks across different contexts and model scales.
Function-space Parameterization of Neural Networks for Sequential Learning
Aidan Scannell (Aalto University), Arno Solin (Aalto University)
ClassificationOptimizationImageTabular
🎯 What it does: This paper proposes a Sparse Function Representation (SFR) technique that transforms a trained neural network from weight space to the function space of Gaussian processes, utilizing bi-parameterization to achieve efficient sparse representation.
Functional Bayesian Tucker Decomposition for Continuous-indexed Tensor Data
Shikai Fang (University of Utah), Shandian Zhe (University of Utah)
Time SeriesStochastic Differential EquationOrdinary Differential Equation
🎯 What it does: This paper proposes a Functional Bayesian Tucker decomposition (FunBaT) that maps continuous indexed tensor data into the interaction of a core tensor and a set of latent functions, using Gaussian processes as function priors and achieving scalable inference through state space models.
Functional Interpolation for Relative Positions improves Long Context Transformers
Shanda Li (Carnegie Mellon University), Srinadh Bhojanapalli (Google DeepMind)
TransformerLarge Language ModelText
🎯 What it does: This paper proposes FIRE, a functional relative position encoding method that enhances the generalization ability of Transformers on long texts through progressive interpolation.
Fusing Models with Complementary Expertise
Hongyi Wang (Carnegie Mellon University), Mikhail Yurochkin (MIT IBM Watson AI Lab)
ClassificationGenerationMixture of ExpertsImageText
🎯 What it does: A framework called FoE is proposed, which integrates models from different domain experts to train a simple fusion model using expert outputs, achieving efficient inference on mixed-domain data. At the same time, FrugalFoE is introduced to reduce the number of expert calls during inference.
Fusion Is Not Enough: Single Modal Attacks on Fusion Models for 3D Object Detection
Zhiyuan Cheng (Purdue University), Xiangyu Zhang (Purdue University)
Object DetectionAutonomous DrivingAdversarial AttackImagePoint Cloud
🎯 What it does: For the camera-lidar multi-sensor fusion model, an attack is proposed using only adversarial stickers from the camera modality. A two-stage optimization framework is introduced: first, identify sensitive areas in the image, and then customize scene-based or target-based attacks according to the model's global or target sensitivity.
Future Language Modeling from Temporal Document History
Changmao Li (University of California), Jeffrey Flanigan (University of California)
GenerationRecurrent Neural NetworkTransformerLarge Language ModelText
🎯 What it does: Proposes a future language modeling task and constructs various temporal language models based on historical texts to predict future texts.
G$^2$N$^2$ : Weisfeiler and Lehman go grammatical
Jason Piquenot (University of Rouen Normandy), Sébastien Adam (University of Rouen Normandy)
ClassificationData-Centric LearningGraph Neural NetworkGraph
🎯 What it does: A general framework is proposed that uses context-free grammars to convert algebraic language fragments into graph neural networks (GNNs) that are provably equivalent to the third-order Weisfeiler–Lehman (3-WL), and within this framework, a new GN² model is derived;