ICLR 2026 Papers — Page 17
International Conference on Learning Representations · 5356 papers
Expressiveness of Multi-Neuron Convex Relaxations in Neural Network Certification
Yuhao Mao (ETH Zürich), Martin Vechev (ETH Zürich)
OptimizationExplainability and Interpretability
🎯 What it does: This paper conducts a systematic theoretical analysis of the expressiveness and completeness of multi-neuron convex relaxations in neural network certification, proving that both intra-layer and cross-layer multi-neuron convex relaxations exhibit incompleteness (i.e., universal convex barriers), and provides feasible pathways to achieve completeness through equivalence-preserving network transformations or convex polytope partitions;
ExpVid: A Benchmark for Experiment Video Understanding & Reasoning
Yicheng Xu (Shanghai AI Laboratory), Yi Wang (Shanghai AI Laboratory)
TransformerLarge Language ModelVision Language ModelVideoTextMultimodalityBenchmark
🎯 What it does: Proposed the ExpVid benchmark specifically designed for scientific experiment videos, covering three-tier tasks of fine-grained perception, procedural understanding, and scientific reasoning.
Extending Fourier Neural Operators for Modeling Parameterized and Coupled PDEs
Cheng Jing (Arizona State University), Kookjin Lee (Arizona State University)
Convolutional Neural NetworkTime SeriesBenchmarkPhysics Related
🎯 What it does: Extend Fourier Neural Operators (FNO) to support parameterized and coupled time-varying partial differential equations, proposing a lightweight hypernetwork modulation and systematic Fourier layer coupling design;
Extending Sequence Length is Not All You Need: Effective Integration of Multimodal Signals for Gene Expression Prediction
Zhao Yang (Renmin University of China), Bing Su (Renmin University of China)
MultimodalityBiomedical Data
🎯 What it does: This paper proposes the Prism framework, which can predict gene expression using short sequences by effectively integrating proximal multi-modal epigenetic signals.
Extending the Context of Pretrained LLMs by Dropping Their Positional Embedding
Yoav Gelberg (University of Oxford), Edoardo Cetin (Sakana AI)
Representation LearningTransformerLarge Language ModelText
🎯 What it does: Propose the DroPE method, which removes position embeddings after pretraining and performs short-term recalibration to achieve zero-shot long context extension in language models.
Extreme Weather Nowcasting via Local Precipitation Pattern Prediction
Chang hoon Song, Youngjoon Hong (Seoul National University)
Computational EfficiencyTransformerVideo
🎯 What it does: Proposes exPreCast, a deterministic radar forecasting framework based on video Swin Transformer, aiming to efficiently and finely predict extreme precipitation events;
f-INE: A Hypothesis Testing Framework for Estimating Influence under Training Randomness
Subhodip Panda (Indian Institute of Science), Sai Praneeth Karimireddy (University of Southern California)
Anomaly DetectionSafty and PrivacyExplainability and InterpretabilityImageText
🎯 What it does: Proposed a hypothesis-testing-based influence estimation framework (f-influence) and designed an efficient algorithm f-INE that can be completed in a single training pass.
FACET: A Fragment-Aware Conformer Ensemble Transformer
Duy Minh Ho Nguyen (Max Planck Research School for Intelligent Systems), Jonathan E Allen (Lawrence Livermore National Labs)
Drug DiscoveryGraph Neural NetworkTransformerGraphBenchmark
🎯 What it does: Proposed FACET, a scalable structure-aware hybrid Transformer model that integrates 2D molecular graphs with 3D conformation ensembles, to efficiently aggregate multi-conformation information and enhance molecular property prediction performance.
FACM: Flow-Anchored Consistency Models
Yansong Peng (University of Science and Technology of China), Feng Wu (University of Science and Technology of China)
GenerationDiffusion modelFlow-based ModelImageTextStochastic Differential Equation
🎯 What it does: Proposed and trained the Flow-Anchored Consistency Model (FACM), achieving efficient few-step generation by combining flow matching with a hybrid objective of continuous-time consistency models.
FACT: a first-principles alternative to the Neural Feature Ansatz for how networks learn representations
Enric Boix-Adserà, Mikhail Belkin (University of California San Diego)
OptimizationRepresentation LearningImageTabular
🎯 What it does: This paper derives the feature matrix relationship (FACT) during neural network training convergence from first principles, using it to replace the empirical hypothesis of neural feature conjecture (NFA).
FACT: Fine-grained Across-variable Convolution for Multivariate Time Series Forecasting
Huiqiang Wang (Hong Kong Polytechnic University), Li Qing (Hong Kong Polytechnic University)
Convolutional Neural NetworkTime SeriesBenchmark
🎯 What it does: Propose a multi-scale time-frequency model FACT based on depthwise separable convolution for high-dimensional multivariate time series forecasting.
Factuality Matters: When Image Generation and Editing Meet Structured Visuals
Le Zhuo (Chinese University of Hong Kong), Hongsheng Li (Chinese University of Hong Kong)
GenerationVision Language ModelDiffusion modelImageMultimodalityBenchmarkChain-of-Thought
🎯 What it does: Constructed a code-aligned structural image dataset with 1.3M scale, and proposed a unified model and evaluation benchmark to address the generation and editing of structural visuals.
Fair Classification by Direct Intervention on Operating Characteristics
Kevin Jiang (University of Pennsylvania), Edgar Dobriban (University of Pennsylvania)
ClassificationTabular
🎯 What it does: This paper proposes a novel classifier that directly intervenes in the operational characteristics (ROC curve) of pre-trained models under an attribute-aware binary classification setting to achieve multiple fairness constraints (e.g., demographic parity, equal opportunity, and predictive parity);
Fair Conformal Classification via Learning Representation-Based Groups
Senrong Xu (Nanjing University), Xiaoxing Ma (Nanjing University)
ClassificationRepresentation LearningAuto EncoderTabularBiomedical Data
🎯 What it does: Propose the FAREG method, using a variational information bottleneck encoder-decoder to learn subgroups in the latent representation space, and constructing an adaptive balanced coverage synthetic prediction set based on this.
Fair Decision Utility in Human-AI Collaboration: Interpretable Confidence Adjustment for Humans with Cognitive Disparities
Jiashi Gao (Southern University of Science and Technology), Xuetao Wei (Southern University of Science and Technology)
Explainability and InterpretabilityImageTextTabular
🎯 What it does: This paper investigates the utility unfairness problem in human-AI collaborative decision-making caused by differences in human decision-makers' cognitive abilities, and proposes a new AI confidence adjustment objective called inter-group-alignment, along with a cognition-aware multicalibration method to achieve dual goals of fairness and optimal utility.
Fair Graph Machine Learning under Adversarial Missingness Processes
Debolina Halder Lina (Rice University), Arlei Silva (Rice University)
OptimizationSafty and PrivacyGraph Neural NetworkGraph
🎯 What it does: To address the fairness issue of graph neural networks in scenarios with missing sensitive attributes (especially adversarial missingness), we propose the BFtS (Better Fair than Sorry) three-player adversarial framework, which improves the balance between model fairness and accuracy through worst-case sensitive attribute imputation.
Fair in Mind, Fair in Action? A Synchronous Benchmark for Understanding and Generation in UMLLMs
Yiran Zhao (Nanjing University of Aeronautics and Astronautics), Liming Fang (Nanjing University of Aeronautics and Astronautics)
Large Language ModelMixture of ExpertsImageMultimodalityBenchmark
🎯 What it does: Propose the IRIS benchmark, constructing a unified 3D fair evaluation framework that uses a high-dimensional fairness space and the IRIS-MBTI diagnostic to simultaneously assess fairness of UMLLMs in generation and understanding tasks.
Fair Reinforcement Learning for Just AI
Ezgi Korkmaz
Reinforcement Learning
🎯 What it does: Proposes an oracle-efficient max quantile fair reinforcement learning algorithm to integrate multiple agents' preferences in multi-objective MDPs, ensuring each agent's return lies at the highest quantile of its return distribution.
Fairness via Independence: A General Regularization Framework for Machine Learning
Yezi Liu (University of California, Irvine), Mohsen Imani (University of California, Irvine)
ClassificationConvolutional Neural NetworkImageTabular
🎯 What it does: Propose a fairness regularization method based on Cauchy-Schwarz divergence to enhance the fairness of machine learning models under multiple fairness metrics across populations.
Fairness-Aware Multi-view Evidential Learning with Adaptive Prior
Haishun Chen (Xidian University), Xin Yang (Xi'an Jiaotong University)
ClassificationSafty and PrivacyImageMultimodality
🎯 What it does: Propose Fairness-Aware Multi-view Evidential Learning (FAML), addressing bias and unreliable uncertainty estimation in multi-view evidential learning through adaptive prior, fairness constraints, and opinion alignment mechanisms.
FaithCoT-Bench: Benchmarking Instance-Level Faithfulness of Chain-of-Thought Reasoning
Xu Shen (Jilin University), Tianlong Chen (University of North Carolina at Chapel Hill)
Explainability and InterpretabilityTextBenchmarkChain-of-Thought
🎯 What it does: Propose the FAITHCOT-BENCH framework, construct the expert-annotated FINE-COT dataset, and systematically evaluate instance-level CoT untrustworthiness detection methods.
Faithful Bi-Directional Model Steering via Distribution Matching and Distributed Interchange Interventions
Yuntai Bao (Zhejiang Normal University), Jianwei Yin (Zhejiang Normal University)
Explainability and InterpretabilityComputational EfficiencyTransformerLarge Language ModelPrompt EngineeringTextBenchmark
🎯 What it does: This paper proposes a lightweight, interpretable bidirectional control method for LLMs based on distribution matching and distributed interactive intervention (CDAS).
Faithfulness Under the Distribution: A New Look at Attribution Evaluation
Zhiyu Zhu (University of Technology Sydney), Jianlong Zhou (University of Technology Sydney)
Explainability and InterpretabilityConvolutional Neural NetworkTransformerDiffusion modelScore-based ModelImageStochastic Differential Equation
🎯 What it does: To assess the credibility of explanation methods, a novel distribution-aware evaluation framework named FUD is proposed. It utilizes a score diffusion model to reconstruct masked regions into data distribution samples while preserving important features, thereby eliminating biases caused by out-of-distribution (OOD) scenarios in traditional methods.
FakeXplain: AI-Generated Image Detection via Human-Aligned Grounded Reasoning
Yikun Ji (Shanghai Jiao Tong University), Jianfu Zhang (Shanghai Jiao Tong University)
Object DetectionAnomaly DetectionExplainability and InterpretabilityLarge Language ModelSupervised Fine-TuningReinforcement LearningPrompt EngineeringVision Language ModelImageMultimodalityChain-of-Thought
🎯 What it does: Developed FakeXplainer, an explainable AI-generated image detection system capable of detecting, locating, and explaining forged images.
Falcon: Fast Proximal Linearization of Normalized Cuts for Unsupervised Image Segmentation
Xiao Zhang (University of Pennsylvania), Konrad Kording (University of Pennsylvania)
SegmentationImageBenchmark
🎯 What it does: Propose Falcon, a primal gradient solver based on first-order approximation, directly optimizing the discrete K-way NCut objective, eliminating recursive bipartitioning and spectral relaxation processes, achieving unsupervised image segmentation.
FALCON: Few-step Accurate Likelihoods for Continuous Flows
Danyal Rehman (Mila - Québec AI Institute), Alexander Tong (Mila - Québec AI Institute)
Drug DiscoveryTransformerFlow-based ModelBiomedical Data
🎯 What it does: Proposed FALCON, a few-step reversible continuous flow model, for efficient and accurate sampling from Boltzmann distributions.
FaLW: A Forgetting-aware Loss Reweighting for Long-tailed Unlearning
Liheng Yu, Yang Wang (University of Science and Technology of China)
OptimizationData-Centric LearningConvolutional Neural NetworkImage
🎯 What it does: This paper studies the long-tailed forgetting problem in machine unlearning and proposes an instance-level dynamic loss reweighting method called FaLW;
FAME: Formal Abstract Minimal Explanation for Neural Networks
Ryma Boumazouza (Airbus SAS), Guy Katz (Hebrew University of Jerusalem)
Explainability and InterpretabilityConvolutional Neural NetworkImage
🎯 What it does: Propose the FAME framework, which utilizes abstract interpretation to generate scalable formal minimal explanations;
Fantastic Pretraining Optimizers and Where to Find Them
Kaiyue Wen (Stanford University), Percy Liang (Stanford University)
OptimizationComputational EfficiencyHyperparameter SearchTransformerTextBenchmark
🎯 What it does: Systematically evaluate the performance of 11 pre-training optimizers across multiple model scales (0.1B-1.2B parameters) and varying data-model ratios (1×–8× Chinchilla), emphasizing the importance of thorough parameter tuning and final evaluation.
Fantastic Tractor-Dogs and How Not to Find Them With Open-Vocabulary Detectors
Frank Ruis (TNO), Hugo Kuijf (TNO)
Object DetectionAnomaly DetectionPrompt EngineeringImage
🎯 What it does: This paper identifies the high false alarm rate of early fusion open vocabulary detectors on images without target backgrounds, and proposes a training-agnostic attention sink technique to suppress false alarms.
FantasyWorld: Geometry-Consistent World Modeling via Unified Video and 3D Prediction
Yixiang Dai (AMAP Alibaba Group), Yonggang Qi (Beijing University Of Posts And Telecommunications)
GenerationDiffusion modelWorld ModelVideoPoint Cloud
🎯 What it does: Proposed FANTASYWORLD, a unified forward network that can simultaneously generate video frames and implicit 3D representations in a single inference.
FAPO: Flawed-Aware Policy Optimization for Efficient and Reliable Reasoning
Yuyang Ding (Soochow University), Min Zhang (Soochow University)
OptimizationComputational EfficiencyReinforcement Learning from Human FeedbackTransformerLarge Language ModelReinforcement LearningTextBenchmark
🎯 What it does: Propose the FAPO method, which uses a generative reward model to accurately detect and penalize reasoning errors within the RLVR framework, thereby improving the reasoning quality and training stability of large language models.
FARI: Robust One-Step Inversion for Watermarking in Diffusion Models
Jindong Yang (University of Science and Technology of China), Kejiang Chen (University of Science and Technology of China)
GenerationSafty and PrivacyDiffusion modelScore-based ModelImage
🎯 What it does: Propose a first-order fast and robust reverse watermark extraction framework called FARI, which can complete the reverse generation of diffusion models and extract embedded watermarks in a single step.
FARTrack: Fast Autoregressive Visual Tracking with High Performance
Guijie Wang (Xi'an Jiaotong University), Xing Wei (Xi'an Jiaotong University)
Object TrackingKnowledge DistillationTransformerVideo
🎯 What it does: Designed and implemented a fast, high-performance multi-template autoregressive visual tracking framework called FARTrack for real-time tracking.
FASA: FREQUENCY-AWARE SPARSE ATTENTION
Yifei Wang (Alibaba Group), Julian McAuley (University of California San Diego)
CompressionComputational EfficiencyTransformerLarge Language ModelTextSequential
🎯 What it does: This paper proposes FASA, a training-free, query-aware KV cache compression framework that dynamically predicts and removes unimportant tokens by leveraging the sparsity of frequency blocks (FC) in RoPE, followed by performing full attention computation only on the retained, limited number of tokens;
Fast and Interpretable Protein Substructure Alignment via Optimal Transport
Zhiyu Wang (Shanghai Jiao Tong University), Liang Hong (Shanghai Jiao Tong University)
Explainability and InterpretabilityComputational EfficiencyProtein Structure PredictionGraph Neural NetworkBiomedical Data
🎯 What it does: Propose a protein local structure substructure alignment framework called PLASMA based on regularized optimal transport, which can efficiently and interpretable identify and align local functional structures of proteins.
Fast and Stable Riemannian Metrics on SPD Manifolds via Cholesky Product Geometry
Ziheng Chen, Nicu Sebe (University of Trento)
ClassificationRecognition
🎯 What it does: Leverage Cholesky decomposition to reveal the product structure of SPD matrix space, and design two new Riemannian metrics—Power-Cholesky Metric (PCM) and Bures-Wasserstein-Cholesky Metric (BWCM)—to provide closed-form, fast, and numerically stable geometric operations for SPD neural networks.
Fast Catch-Up, Late Switching: Optimal Batch Size Scheduling via Functional Scaling Laws
Jinbo Wang (Peking University), Lei Wu (Peking University)
OptimizationComputational EfficiencyTransformerLarge Language ModelMixture of ExpertsTextStochastic Differential Equation
🎯 What it does: Explores the theory and practice of batch size scheduling (BSS) in large language model pretraining, proposing an optimal batch scheduling scheme based on the functional scaling law (FSL);
Fast Convergence of Natural Gradient Descent for Over-parameterized Physics-Informed Neural Networks
Xianliang Xu (Tsinghua University), Ye Li (Nanjing University of Aeronautics and Astronautics)
OptimizationPhysics Related
🎯 What it does: This paper investigates the convergence of gradient descent and natural gradient descent for over-parameterized two-layer PINNs, proposes improved learning rate and width requirements, and provides theoretical convergence rates along with experimental validation.
Fast Data Mixture Optimization via Gradient Descent
Haoru Tan (University of Hong Kong), Xiaojuan Qi (University of Hong Kong)
OptimizationComputational EfficiencyText
🎯 What it does: Propose the FASTMIX framework, which uses a single-agent model to automatically find the optimal data mixing ratio through gradient descent.
Fast Escape, Slow Convergence: Learning Dynamics of Phase Retrieval under Power-Law Data
Guillaume Braun (RIKEN AIP), Masaaki Imaizumi (University of Tokyo)
OptimizationPhysics RelatedOrdinary Differential Equation
🎯 What it does: Study the gradient flow dynamics of phase retrieval under non-homogeneous Gaussian inputs with power-law covariance, revealing a three-stage learning trajectory and deriving the error scaling law.
Fast Estimation of Wasserstein Distances via Regression on Sliced Wasserstein Distances
Khai Nguyen (University of Texas at Austin), Nhat Ho (University of Texas at Austin)
OptimizationComputational EfficiencyPoint CloudBiomedical Data
🎯 What it does: Propose a framework for fast estimation of Wasserstein distance by regressing Wasserstein distance on sliced Wasserstein distance (SW);
Fast Frank–Wolfe Algorithms with Adaptive Bregman Step-Size for Weakly Convex Functions
Shota Takahashi (University of Tokyo), Akiko Takeda (University of Tokyo)
OptimizationComputational Efficiency
🎯 What it does: Proposed an adaptive Bregman step size Frank-Wolfe algorithm applicable to L-smooth adaptable (L-smad) and weakly convex objective functions, providing linear convergence analysis for both convex and non-convex scenarios.
Fast Proteome-Scale Protein Interaction Retrieval via Residue-Level Factorization
Jianan Zhao (Mila - Québec AI Institute), Jian Tang (Mila - Québec AI Institute)
RetrievalTransformerBiomedical Data
🎯 What it does: A fast protein interaction retrieval was achieved by constructing a factorizable residue-level interaction model.
Fast training of accurate physics-informed neural networks without gradient descent
Chinmay Datar (Technical University of Munich), Felix Dietrich (Technical University of Munich)
OptimizationComputational EfficiencyPhysics RelatedOrdinary Differential Equation
🎯 What it does: Propose Frozen-PINNs, utilizing space-time separation, random features, and ODE solving to achieve gradient-free high-precision physics-informed neural network training.
Fast-dLLM v2: Efficient Block-Diffusion LLM
Chengyue Wu (NVIDIA), Enze Xie (NVIDIA)
Computational EfficiencyTransformerLarge Language ModelSupervised Fine-TuningDiffusion modelTextBenchmark
🎯 What it does: Convert pre-trained autoregressive LLMs into the block diffusion model Fast-dLLM v2, achieving parallel text generation with only about 1B tokens of fine-tuning through block-level diffusion and complementary masks.
Fast-dLLM: Training-free Acceleration of Diffusion LLM by Enabling KV Cache and Parallel Decoding
Chengyue Wu (University of Hong Kong), Enze Xie (NVIDIA)
Computational EfficiencyLarge Language ModelDiffusion modelText
🎯 What it does: This paper proposes the Fast-dLLM framework, which significantly improves the inference speed of diffusion-based large language models while maintaining output quality by utilizing block-level approximate KV cache and confidence-aware parallel decoding.
FAST‑DIPS: Adjoint‑Free Analytic Steps and Hard‑Constrained Likelihood Correction for Diffusion‑Prior Inverse Problems
Minwoo Kim (Inha University), Hongki Lim (Inha University)
RestorationOptimizationComputational EfficiencyDiffusion modelImage
🎯 What it does: Proposes a training-free diffusion prior inverse problem solver, FAST-DIPS, which achieves data consistency at each noise level through hard constraint measurement space correction and resampling.
FaSTA*: Fast-Slow Toolpath Agent with Subroutine Mining for Efficient Multi-turn Image Editing
Advait Gupta (University of Maryland), Tianyi Zhou (Mohamed bin Zayed University of Artificial Intelligence)
Computational EfficiencyTransformerLarge Language ModelVision Language ModelImage
🎯 What it does: Propose a neuro-symbolic multi-step image editing agent called FaSTA ...
FastAvatar: Towards Unified and Fast 3D Avatar Reconstruction with Large Gaussian Reconstruction Transformers
Yue Wu (Tongji University), Kairui Feng (Tongji University)
GenerationPose EstimationTransformerDiffusion modelGaussian SplattingImageVideoPoint Cloud
🎯 What it does: Propose an end-to-end feed-forward 3D avatar reconstruction framework called FastAvatar, which can rapidly generate high-quality animatable 3D Gaussian Splatting (3DGS) models from any number of images or video frames, and supports progressive incremental reconstruction.
Fastcar: Cache Attentive Replay for Fast Auto-Regressive Video Generation on the Edge
Xuan Shen (Northeastern University), Jiuxiang Gu (Adobe)
GenerationTransformerVideo
🎯 What it does: Propose the FastCar framework, which utilizes temporal redundancy in autoregressive video generation by caching and replaying the MLP output of the previous frame to significantly reduce computational costs; and implement a dynamic resource scheduling hardware accelerator on FPGA;
Faster Gradient Methods for Highly-smooth Stochastic Bilevel Optimization
Lesi Chen (Tsinghua University & Shanghai Qizhi Institute), Jingzhao Zhang (Tsinghua University & Shanghai Qizhi Institute)
OptimizationText
🎯 What it does: This paper studies fully gradient-based stochastic bilevel optimization algorithms and proposes the FSA^p method, which achieves faster convergence speeds by approximating the supergradient using high-order finite differences.
Faster Vision Transformers with Adaptive Patches
Rohan Choudhury (Carnegie Mellon University), Kris Kitani
ClassificationObject DetectionSegmentationTransformerImage
🎯 What it does: Propose Adaptive Patch Transformer (APT), which accelerates Vision Transformer training and inference by adaptively partitioning images into patches (large patches for homogeneous regions, small patches for complex regions).
FASTer: Toward Powerful and Efficient Autoregressive Vision–Language–Action Models with Learnable Action Tokenizer and Block-wise Decoding
Yicheng Liu (Tsinghua University), Hang Zhao (Tsinghua University)
Computational EfficiencyData-Centric LearningRobotic IntelligenceTransformerMixture of ExpertsVision-Language-Action ModelAuto EncoderMultimodalitySequential
🎯 What it does: Proposed the FASTer framework, combining the learnable action tokenizer FASTerVQ with the block-level autoregressive decoder FASTerVLA, achieving efficient visual-language-action (VLA) learning.
FastFlow: Accelerating The Generative Flow Matching Models with Bandit Inference
Divya Jyoti Bajpai (Indian Institute of Technology Bombay), Manjesh Kumar Hanawal (Indian Institute of Technology Bombay)
GenerationComputational EfficiencyFlow-based ModelImageVideoTextOrdinary Differential Equation
🎯 What it does: Significantly accelerate generation by adaptively skipping redundant denoising steps during the inference of flow matching models, using a multi-armed bandit to determine when to approximate.
FastGHA: Generalized Few-Shot 3D Gaussian Head Avatars with Real-Time Animation
Xinya Ji (Nanjing University), Derek Bradley
GenerationTransformerDiffusion modelAuto EncoderGaussian SplattingVideoPoint Cloud
🎯 What it does: Propose the FastGHA method, which rapidly generates high-quality 3D Gaussian head avatars capable of real-time animation using a small number of view images.
FastGRPO: Accelerating Policy Optimization via Concurrency-aware Speculative Decoding and Online Draft Learning
Yizhou Zhang (Lanzhou University), Jisheng Dang (Lanzhou University)
OptimizationComputational EfficiencyLarge Language ModelReinforcement LearningText
🎯 What it does: Significantly improved the throughput during the generation phase by introducing concurrent-aware speculative decoding and online draft model learning in GRPO training, thereby accelerating the overall training process.
FastVGGT: Fast Visual Geometry Transformer
You Shen (Xiamen University), Liujuan Cao (Xiamen University)
RestorationOptimizationComputational EfficiencyTransformerImagePoint CloudMeshBenchmark
🎯 What it does: Propose FastVGGT, a training-free framework that significantly accelerates long-sequence 3D reconstruction inference by performing high-ratio merging of tokens in the global attention of VGGT.
FastVMT: Eliminating Redundancy in Video Motion Transfer
Yue Ma (Hong Kong University of Science and Technology), Linfeng Zhang (Shanghai Jiao Tong University)
GenerationTransformerDiffusion modelVideoText
🎯 What it does: This paper proposes FastVMT, a training-free video motion transfer framework that utilizes Diffusion Transformer for high-quality video synthesis, significantly improving inference speed through sliding window motion extraction and gradient skip optimization.
FATE: A Formal Benchmark Series for Frontier Algebra of Multiple Difficulty Levels
Jiedong Jiang (Westlake University), Bin Dong (Peking University)
AI Code AssistantLarge Language ModelTextBenchmark
🎯 What it does: This study proposes the FATE (Formal Algebra Theorem Evaluation) benchmark series, focusing on constructing two new difficulty levels—FATE-H (undergraduate to graduate level) and FATE-X (doctoral qualifying exams and above)—to assess the capability of large language models in formal algebraic proofs; and reveals the bottlenecks in the formalization process through a two-phase evaluation (first natural language reasoning, then conversion to Lean code).
Fathom-DeepResearch: Unlocking Long Horizon Information Retrieval and Synthesis for SLMs
Shreyas Singh (Fractal AI Research), Pradeep Moturi (Fractal AI Research)
GenerationData SynthesisRetrievalTransformerLarge Language ModelSupervised Fine-TuningReinforcement LearningTextRetrieval-Augmented GenerationChain-of-Thought
🎯 What it does: Built a two-model proxy system Fathom-DeepResearch, including a specialized search inference model Fathom-Search-4B and a structured report generation model Fathom-Synthesizer-4B;
Feature compression is the root cause of adversarial fragility in neural networks
Jingchao Gao (University of Iowa), Weiyu Xu (University of Iowa)
ClassificationAdversarial AttackImage
🎯 What it does: Studied adversarial robustness of neural network classifiers, proposing a matrix theory explanation for their vulnerability and comparing with optimal classifiers.
Feature segregation by signed weights in artificial vision systems and biological models
Giordano Ramos-Traslosheros (Harvard Medical School), Carlos R Ponce (Harvard Medical School)
Representation LearningConvolutional Neural NetworkGenerative Adversarial NetworkImageBiomedical Data
🎯 What it does: By performing positive and negative weight ablation on ImageNet-pretrained convolutional neural networks, using GAN-driven zeroth-order optimization for feature visualization, and constructing linear response models and closed-loop experimental validation in the gorilla visual cortex (V1/V4/IT), this study investigates the functional segregation of positive and negative weights in visual representations.
FeatureBench: Benchmarking Agentic Coding for Complex Feature Development
Qixing Zhou (Institute of Automation Chinese Academy of Sciences), Zhaoxiang Zhang (Institute of Automation Chinese Academy of Sciences)
AI Code AssistantLarge Language ModelAgentic AITextBenchmark
🎯 What it does: Built a feature-based agent coding benchmark called FeatureBench to evaluate the performance of LLM-driven code agents in real-world software feature development.
Features Emerge as Discrete States: The First Application of SAEs to 3D Representations
Albert Miao (University of Cambridge), Cengiz Oztireli (University of Cambridge)
Explainability and InterpretabilityRepresentation LearningAuto EncoderMesh
🎯 What it does: First applied sparse autoencoder (SAE) to the latent space of 3D reconstruction models, analyzed and explained the extracted discrete state features and their roles in reconstruction.
Fed-Duet: Dual Expert-Orchestrated Framework for Continual Federated Vision-Language Learning
Tao GUO, Laizhong Cui (Shenzhen University)
Federated LearningKnowledge DistillationPrompt EngineeringMixture of ExpertsVision Language ModelContrastive LearningMultimodality
🎯 What it does: Propose Fed-Duet, combining a server-side knowledge orchestrator with client-side dual experts (semantic prompts + parameter adapters) to achieve federated continual learning for vision-language models.
FeDaL: Federated Dataset Learning for General Time Series Foundation Models
Shengchao Chen (University of Technology Sydney), Jing Jiang (University of Technology Sydney)
Anomaly DetectionFederated LearningTransformerTime Series
🎯 What it does: Proposes FeDaL, a federated learning framework for time series foundation models, capable of training models from scratch that can achieve cross-domain generalization on distributed and heterogeneous datasets.
FedDAG: Clustered Federated Learning via Global Data and Gradient Integration for Heterogeneous Environments
Anik Pramanik (New Jersey Institute of Technology), Shantanu Sharma (Virginia Tech)
ClassificationFederated LearningImage
🎯 What it does: Proposed the FEDDAG framework, which achieves clustering federated learning by integrating data similarity and gradient similarity through a dual encoder.
Federated ADMM from Bayesian Duality
Thomas Möllenhoff (RIKEN Center for AI Project), Mohammad Emtiyaz Khan (RIKEN Center for AI Project)
OptimizationFederated LearningImage
🎯 What it does: Proposed a variational Bayesian Bayesian ADMM (Bayesian-ADMM) framework that can optimize parameter distributions in federated learning and naturally generalize traditional ADMM.
Federated Graph-Level Clustering Network with Dual Knowledge Separation
Xiaobao Wang (Tianjin University), Di Jin (Tianjin University)
Federated LearningRepresentation LearningGraph Neural NetworkGraph
🎯 What it does: Propose a new framework called FGCN-DKS for graph-level clustering in a federated environment, which can achieve effective consensus among different clients while preserving privacy.
Federated Learning of Quantile Inference under Local Differential Privacy
Leheng Cai (Tsinghua University), Shuyuan Wu (Shanghai University of Finance and Economics)
Federated LearningSafty and PrivacyTabular
🎯 What it does: This paper proposes a federated learning algorithm under the local differential privacy (LDP) framework, using local stochastic gradient descent (SGD) to achieve quantile estimation and inference.
Federated Learning with Profile Mapping under Distribution Shifts and Drifts
Mohan Li (Università della Svizzera italiana), Marc Langheinrich (Università della Svizzera italiana)
Federated LearningSafty and PrivacyImageBiomedical Data
🎯 What it does: Propose the FEROMA framework, which utilizes differential privacy distribution summaries to map federated learning distributions, dynamically selects aggregation strategies, handles distribution shifts and drifts during both training and testing phases, and supports model allocation for unlabelled clients.
FedMC: Federated Manifold Calibration
Yanbiao Ma (Renmin University of China), Andi Zhang (University of Warwick)
Federated LearningSafty and PrivacyTransformerVision Language ModelImageBenchmark
🎯 What it does: Propose a new federated learning framework FedMC, utilizing local nonlinear manifold geometry calibration to alleviate bias caused by data heterogeneity
FedMuon: Federated Learning with Bias-corrected LMO-based Optimization
Yuki Takezawa (Kyoto University), Sebastian U Stich
OptimizationFederated LearningConvolutional Neural NetworkImage
🎯 What it does: Propose FEDMUON, a federated learning optimizer, to address the convergence failure of LocalMuon caused by the bias in Muon's linear minimization operator (LMO), and prove that any number of Newton-Schulz iterations guarantees convergence.
FedOpenMatch: Towards Semi-Supervised Federated Learning in Open-Set Environments
Hongquan Liu (Fudan University), Shuigeng Zhou (Fudan University)
ClassificationAnomaly DetectionFederated LearningImageBenchmark
🎯 What it does: Proposed and implemented the first framework for open-set semi-supervised federated learning (OSSFL), FedOpenMatch, which can simultaneously handle samples from both known classes and unknown classes in the unlabeled data distributed across clients;
Feedback-driven recurrent quantum neural network universality
Lukas Gonon (University of St. Gallen), Juan-Pablo Ortega (Nanyang Technological University)
Recurrent Neural NetworkPhysics Related
🎯 What it does: Investigate the theoretical approximation capabilities of feedback-driven quantum recurrent neural networks (RQNN), providing error bounds and universality theorems.
FERD: Fairness-Enhanced Data-Free Adversarial Robustness Distillation
Zhengxiao Li (Nanjing University of Science and Technology), Shuchao Pang (Nanjing University of Science and Technology)
Knowledge DistillationConvolutional Neural NetworkImage
🎯 What it does: Studied the fairness issue in data-free robustness distillation and proposed the FERD framework to enhance the robust fairness of student models
FETAL-GAUGE: A BENCHMARK FOR ASSESSING VISION-LANGUAGE MODELS IN FETAL ULTRASOUND
Hussain Alasmawi (Mohamed bin Zayed University of Artificial Intelligence), Mohammad Yaqub (Mohamed bin Zayed University of Artificial Intelligence)
TransformerSupervised Fine-TuningImageBiomedical DataUltrasoundBenchmark
🎯 What it does: Proposed Fetal-Gauge, the first visual-language question-answering benchmark for fetal ultrasound, containing 42,036 images and 93,451 multiple-choice questions; systematically evaluated 15 leading visual-language models (including medical-specialized and general-purpose models) on five clinical tasks; revealed model limitations in fetal ultrasound through fine-grained view identification, localization, and diagnostic tasks; and performed domain adaptation fine-tuning to observe performance improvements.
Fewer Battles, More Gain: An Information-Efficient Framework for Arena-based LLM Evaluation
Zirui Liu (University of Science and Technology of China), Shijin Wang (iFLYTEK Co., Ltd)
OptimizationComputational EfficiencyText
🎯 What it does: Propose an adaptive model pair selection framework based on Fisher information, using A-optimality and D-optimality to iteratively select the most informative model pairs, significantly reducing the number of required matches for evaluation;
Fewer Weights, More Problems: A Practical Attack on LLM Pruning
Kazuki Egashira (ETH Zurich), Martin Vechev (ETH Zurich)
Adversarial AttackTransformerLarge Language ModelSupervised Fine-TuningText
🎯 What it does: Proposes an attack method that leverages LLM pruning to activate malicious behaviors;
FHE-Coder: Benchmarking Secure Agentic Code Generation for Fully Homomorphic Encryption
Mayank Kumar (University of Central Florida), Qian Lou (University of Central Florida)
Safty and PrivacyAI Code AssistantLarge Language ModelAgentic AIPrompt EngineeringTextBenchmarkRetrieval-Augmented GenerationChain-of-Thought
🎯 What it does: Propose the FHE-Coder framework, which uses LLMs to automatically generate FHE (TFHE and CKKS) code compliant with security standards.
FictionalQA: A Dataset for Studying Memorization and Knowledge Acquisition
John Kirchenbauer (University of Maryland), Daphne Ippolito (Carnegie Mellon University)
Data SynthesisExplainability and InterpretabilityData-Centric LearningTransformerLarge Language ModelSupervised Fine-TuningTextBenchmark
🎯 What it does: This paper constructs a generative model-based fictional event QA dataset to study the memory mechanisms of language models for facts and sequences during training.
FideDiff: Efficient Diffusion Model for High-Fidelity Image Motion Deblurring
Xiaoyang Liu (Shanghai Jiao Tong University), Yulun Zhang (Shanghai Jiao Tong University)
RestorationDiffusion modelAuto EncoderGenerative Adversarial NetworkImage
🎯 What it does: Proposed a single-step high-fidelity motion deblurring diffusion model called FideDiff, achieving one-time restoration during inference.
FieryGS: In-the-Wild Fire Synthesis with Physics-Integrated Gaussian Splatting
Qianfan Shen (Peking University), Baoquan Chen (Peking University)
GenerationData SynthesisLarge Language ModelDiffusion modelGaussian SplattingImagePhysics Related
🎯 What it does: This paper proposes FieryGS, a flame synthesis framework based on 3D Gaussian Splatting, which automatically generates controllable and physically consistent flames from multi-view images.
Figma2Code: Automating Multimodal Design to Code in the Wild
Yi Gui (Huazhong University of Science and Technology), Philip S. Yu (University of Illinois Chicago)
AI Code AssistantTransformerLarge Language ModelAgentic AIVision Language ModelImageTextMultimodality
🎯 What it does: Proposed the FIGMA2CODE task, which investigates how to automatically convert Figma design files (containing images, JSON metadata, and resources) into executable frontend code;
FilMaster: Bridging Cinematic Principles and Generative AI for Automated Film Generation
Kaiyi Huang (University of Hong Kong), Xihui Liu (University of Hong Kong)
GenerationRetrievalTransformerLarge Language ModelVideoTextMultimodalityRetrieval-Augmented Generation
🎯 What it does: Built FilMaster, an end-to-end automated generation system from scripts to editable movies.
Fine-Grained Activation Steering: Steering Less, Achieving More
Zijian Feng (Nanyang Technological University), Kezhi Mao (Nanyang Technological University)
TransformerLarge Language ModelContrastive LearningText
🎯 What it does: Investigated and demonstrated the heterogeneity of block-level activation in large language models (LLMs), proposing a fine-grained atomic unit (AU)-level activation modulation method called AUSteer;
Fine-Grained Class-Conditional Distribution Balancing for Debiased Learning
Miaoyun Zhao (Dalian University of Technology), Qiang Zhang (Dalian University of Technology)
Domain AdaptationImageText
🎯 What it does: Proposes an unbiased annotation bias exploration and fine-grained class-conditional distribution balancing (FG-CCDB) method to achieve robust group-level generalization in the presence of spurious correlations.
Fine-Grained Iterative Adversarial Attacks with Limited Computation Budget
Zhichao Hou (North Carolina State University), Xiaorui Liu (North Carolina State University)
Adversarial AttackSpiking Neural NetworkImageGraph
🎯 What it does: Propose a fine-grained activation reuse Spiking-PGD attack under limited computational budget to maximize the strength of iterative adversarial attacks.
Fine-Grained Privacy Extraction from Retrieval-Augmented Generation Systems by Exploiting Knowledge Asymmetry
Yufei Chen (Xidian University), Tao Gu (Macquarie University)
Safty and PrivacyAdversarial AttackTransformerTextBiomedical DataRetrieval-Augmented Generation
🎯 What it does: This study proposes a black-box attack framework capable of achieving fine-grained privacy leakage localization and cross-domain generalization within retrieval-augmented generation (RAG) systems;
Fine-R1: Make Multi-modal LLMs Excel in Fine-Grained Visual Recognition by Chain-of-Thought Reasoning
Hulingxiao He (Peking University), Yuxin Peng (Peking University)
RecognitionTransformerLarge Language ModelSupervised Fine-TuningReinforcement LearningVision Language ModelContrastive LearningImageTextMultimodalityChain-of-Thought
🎯 What it does: Designed and trained a multimodal large language model, Fine-R1, for fine-grained visual recognition (FGVR), enabling it to identify both known and unknown subcategories with a small number of training samples.
Fine-tuning Behavioral Cloning Policies with Preference‑Based Reinforcement Learning
Mael Macuglia (University of Zurich), Giorgia Ramponi (University of Zurich)
OptimizationReinforcement Learning from Human FeedbackSupervised Fine-TuningReinforcement LearningSequential
🎯 What it does: Integrate offline expert demonstrations with online preference feedback, proposing the BRIDGE algorithm. By constructing a Hellinger confidence ball to constrain online search, it achieves safe and efficient behavior cloning fine-tuning.
Fine-Tuning Diffusion Models via Intermediate Distribution Shaping
Gautham Govind Anil (Google DeepMind), Karthikeyan Shanmugam (Google DeepMind)
GenerationData SynthesisSupervised Fine-TuningReinforcement LearningDiffusion modelFlow-based ModelImageText
🎯 What it does: This paper proposes to fine-tune diffusion models by shaping intermediate distributions, unifying rejection sampling fine-tuning methods into GRAFT, and further introducing P-GRAFT, which fine-tunes only at the intermediate noise layer, along with designing inverse noise correction to improve flow models.
Fine-tuning Done Right in Model Editing
Wanli Yang (Institute of Computing Technology, Chinese Academy of Sciences), Fei Sun (Institute of Computing Technology, Chinese Academy of Sciences)
Computational EfficiencyTransformerLarge Language ModelSupervised Fine-TuningText
🎯 What it does: Re-evaluated and corrected the implementation of fine-tuning in model editing, transforming it into a standard breadth-first (BF) training pipeline combined with local parameter fine-tuning, and proposed the LocFT-BF method;
Fine-tuning Quantized Neural Networks with Zeroth-order Optimization
Sifeng SHANG, Kaiyang Zhou (Hong Kong Baptist University)
OptimizationComputational EfficiencyTransformerSupervised Fine-TuningDiffusion modelImageText
🎯 What it does: Propose a method called Quantized Zeroth-order Optimization (QZO) for fine-tuning quantized large language models (LLMs), achieving extreme memory compression by only updating the quantization scale and eliminating gradients and optimizer states.
Fingerprinting Deep Neural Networks for Ownership Protection: An Analytical Approach
Guang Yang (Virginia Commonwealth University), Changqing Luo (University of Houston)
Safty and PrivacyAdversarial AttackHyperparameter SearchConvolutional Neural NetworkGraph Neural NetworkImageGraph
🎯 What it does: Proposes a theoretical analysis-based fingerprint generation method called AnaFP, which uses a stretching factor to control the distance between the fingerprint and the decision boundary, thereby achieving uniqueness while maintaining robustness.
FingerTip 20K: A Benchmark for Proactive and Personalized Mobile LLM Agents
Qinglong Yang (Tsinghua University), Yong Li (Tsinghua University)
TransformerLarge Language ModelSupervised Fine-TuningAgentic AIPrompt EngineeringImageTextSequentialBenchmark
🎯 What it does: This paper proposes the FingerTip 20K benchmark to evaluate the capabilities of mobile large language model agents in proactive task suggestion and personalized task execution, and constructs the dataset by collecting 20,000 real user multi-step Android interaction examples.
Finite-Time Analysis of Actor-Critic Methods with Deep Neural Network Approximation
Xuyang Chen (National University of Singapore), Lin Zhao (National University of Singapore)
Reinforcement LearningBenchmark
🎯 What it does: Propose a single-time-scale Actor-Critic algorithm that uses deep neural networks to approximate value functions and policies in continuous state-action spaces, and provides theoretical analysis of its finite-time convergence to a stationary point; meanwhile, experimental validation is conducted on environments such as Pendulum and MuJoCo.
Finite-Time Convergence Analysis of ODE-based Generative Models for Stochastic Interpolants
Yuhao Liu (Tsinghua University), Longbo Huang (Tsinghua University)
GenerationImageOrdinary Differential Equation
🎯 What it does: Perform finite-time convergence analysis of ODE generative models based on random interpolation, providing upper bounds on total variation error for forward Euler and Heun numerical solvers, and deriving iteration complexity from these bounds; subsequently validate the results through experiments on 2D datasets, high-dimensional Gaussian mixtures, and image generation tasks.
FinSearchComp: Towards a Realistic, Expert-Level Evaluation of Financial Search and Reasoning
LIANG HU, Wenhao Huang (ByteDance Seed)
TransformerLarge Language ModelTextBenchmarkFinance RelatedRetrieval-Augmented GenerationChain-of-Thought
🎯 What it does: This paper proposes and publicly releases the FINSEARCHCOMP benchmark to evaluate the end-to-end retrieval and reasoning capabilities of large language models in real financial scenarios.