ICLR 2025 Papers — Page 12
International Conference on Learning Representations · 3704 papers
Entropy-based Activation Function Optimization: A Method on Searching Better Activation Functions
Haoyuan Sun (Tsinghua University), Xueqian Wang (Tsinghua University)
OptimizationTransformerSupervised Fine-TuningImage
🎯 What it does: An activation function optimization framework based on information entropy theory (EAFO) is proposed, and a new activation function CRReLU is derived within this framework.
Episodic Memories Generation and Evaluation Benchmark for Large Language Models
Alexis Huet (Huawei Technologies), Dario Rossi (Huawei Technologies)
GenerationData SynthesisRetrievalTransformerLarge Language ModelPrompt EngineeringTextBenchmarkRetrieval-Augmented Generation
🎯 What it does: A cognitive science-based framework for episodic memory is proposed, generating an uncontaminated episodic memory benchmark to evaluate large language models' abilities in recall, entity state tracking, and spatiotemporal relationships.
Episodic Novelty Through Temporal Distance
Yuhua Jiang (Tsinghua University), Qianchuan Zhao (Tsinghua University)
Robotic IntelligenceReinforcement LearningContrastive LearningSequential
🎯 What it does: A distance-based intra-episode reward mechanism named ETD is proposed, which drives the agent's exploration by learning distances in the context of Markov decision processes;
Epistemic Monte Carlo Tree Search
Yaniv Oren (Delft University of Technology), Wendelin Boehmer
Reinforcement LearningSequential
🎯 What it does: This paper proposes Epistemic Monte Carlo Tree Search (EMCTS), which incorporates the ontological uncertainty generated by learning models into MCTS to achieve deep exploration.
eQMARL: Entangled Quantum Multi-Agent Reinforcement Learning for Distributed Cooperation over Quantum Channels
Alexander DeRieux (Virginia Tech), Walid Saad (Virginia Tech)
Reinforcement LearningPhysics Related
🎯 What it does: A distributed actor-critic framework based on quantum entanglement, eQMARL, is proposed for cooperation and learning in multi-agent reinforcement learning.
EqNIO: Subequivariant Neural Inertial Odometry
Royina Karegoudra Jayanth (University of Pennsylvania), Daniel Gehrig (University of Pennsylvania)
Autonomous DrivingRobotic IntelligenceSimultaneous Localization and MappingTime Series
🎯 What it does: A method for Neural Inertial Odometry (EqNIO) based on an equivariant framework is proposed. This method first projects IMU data into a learned, gravity-aligned, and equivariant 'canonical frame', then trains a neural displacement prior within this framework, and finally maps the predicted results back to the original frame, achieving complete invariance to changes in IMU orientation.
Equivariant Denoisers Cannot Copy Graphs: Align Your Graph Diffusion Models
Najwa Laabid (Aalto University), Vikas Garg (Massachusetts Institute of Technology)
Drug DiscoveryGraph Neural NetworkDiffusion modelGraph
🎯 What it does: Research and improve the performance of graph diffusion models in graph-to-graph translation tasks (taking chemical reaction prediction as an example).
Equivariant Masked Position Prediction for Efficient Molecular Representation
Junyi An (Shanghai Academy of Artificial Intelligence for Science), Yuan Qi (Fudan University)
Representation LearningDrug DiscoveryGraph Neural NetworkGraph
🎯 What it does: A new self-supervised learning framework is proposed—Equivariant Masked Position Prediction (EMPP), which better captures quantum mechanical features by masking atomic 3D positions and predicting their coordinates using information from neighboring atoms.
Equivariant Neural Functional Networks for Transformers
Hoang V. Tran, Tan Minh Nguyen
TransformerImageTextBenchmark
🎯 What it does: This paper systematically studies the Neural Functional Network (NFN) of the Transformer by analyzing the maximum symmetric group of multi-head attention and constructing the weight space and group action. It proposes the Transformer-NFN, a polynomial NFN that is equivariant to this group action, and releases the Small Transformer Zoo dataset containing over 12,500 Transformer checkpoints.
Erasing Concept Combination from Text-to-Image Diffusion Model
hongyi nie, Yatao Bian (Tencent AI Lab)
GenerationData SynthesisLarge Language ModelDiffusion modelImageText
🎯 What it does: This paper proposes a new Concept Combination Elimination (CCE) framework that can eliminate inappropriate content concept combinations in text-to-image diffusion models without compromising the quality of individual concept generation.
Error-quantified Conformal Inference for Time Series
Junxi Wu (Nankai University), Changliang Zou (Nankai University)
Time SeriesFinance Related
🎯 What it does: This paper proposes an online consistency inference method based on error quantification (Error-Quantified Conformal Inference, ECI) for uncertainty quantification in time series forecasting.
ESE: Espresso Sentence Embeddings
Xianming LI, Qing Li (Hong Kong Polytechnic University)
CompressionRepresentation LearningTransformerSupervised Fine-TuningTextRetrieval-Augmented Generation
🎯 What it does: This paper proposes a sentence embedding method called ESE (Espresso Sentence Embeddings) that can simultaneously achieve scalability in model depth and embedding dimensions.
Estimating the Probabilities of Rare Outputs in Language Models
Gabriel Wu (Alignment Research Center), Jacob Hilton (Alignment Research Center)
TransformerLarge Language ModelText
🎯 What it does: This paper proposes a task of estimating extremely low probability outputs (such as the occurrence probability of rare words) in language models and conducts experiments on a small Transformer.
Estimation of single-cell and tissue perturbation effect in spatial transcriptomics via Spatial Causal Disentanglement
Stathis Megas (University of Cambridge), Sarah A Teichmann
Graph Neural NetworkBiomedical Data
🎯 What it does: This paper proposes Celcomen, an interpretable generative graph neural network, for achieving causal separation and spatial counterfactual prediction at the cellular and tissue levels in spatial transcriptomics data.
ET-SEED: EFFICIENT TRAJECTORY-LEVEL SE(3) EQUIVARIANT DIFFUSION POLICY
Chenrui Tie (National University of Singapore), Hao Dong (Peking University)
Robotic IntelligenceReinforcement Learning from Human FeedbackTransformerReinforcement LearningDiffusion modelPoint Cloud
🎯 What it does: This paper proposes a trajectory-level robot manipulation model ET-SEED based on SE(3) equivariant diffusion strategies, achieving efficient pose trajectory generation through a small number of demonstrations.
ETA: Evaluating Then Aligning Safety of Vision Language Models at Inference Time
Yi Ding (Purdue University), Ruqi Zhang (Purdue University)
Safty and PrivacyTransformerLarge Language ModelVision Language ModelImageTextMultimodality
🎯 What it does: A two-stage inference-time safety alignment framework named ETA is proposed, which first assesses the safety of input and output through visual and textual evaluations, and then achieves safe and useful responses through shallow interventions (interference prefixes) and deep searches (sentence-level best-of-N).
EvA: Erasing Spurious Correlations with Activations
Qiyuan He (National University of Singapore), Angela Yao (National University of Singapore)
ClassificationRecognitionComputational EfficiencyConvolutional Neural NetworkImage
🎯 What it does: The EvA method is proposed, which detects and erases pseudo-correlated channels in activations, reweights the last layer to eliminate the model's pseudo-correlation.
EVA: Geometric Inverse Design for Fast Protein Motif-Scaffolding with Coupled Flow
Yufei Huang (Zhejiang University), Stan Z. Li (Westlake University)
OptimizationDrug DiscoveryFlow-based ModelBiomedical DataBenchmarkOrdinary Differential Equation
🎯 What it does: A fast ligand construction framework called EVA based on geometric inverse design is proposed, utilizing a pre-trained flow model to generate protein backbones, and introducing ligand alignment priors and spatial interpolation during the sampling process to achieve conditional constraints.
Evaluating Large Language Models through Role-Guide and Self-Reflection: A Comparative Study
Lili Zhao (University of Science and Technology of China), Shijin Wang (iFLYTEK Co. Ltd)
TransformerLarge Language ModelSupervised Fine-TuningPrompt EngineeringText
🎯 What it does: This paper proposes the RoSe (Role-guided and Self-reflection) strategy, which evaluates self-awareness and self-correction in large language models (LLMs) and extracts high-quality data from closed-source LLMs through dual calibration (accuracy and confidence) for fine-tuning open-source LLMs.
Evaluating Semantic Variation in Text-to-Image Synthesis: A Causal Perspective
Xiangru Zhu (Fudan University), Xiaoxiao Xu (Renmin University of China)
GenerationData SynthesisTransformerLarge Language ModelSupervised Fine-TuningImageTextBenchmark
🎯 What it does: This paper studies the causal relationship between word order changes and semantic differences in text-to-image (T2I) synthesis, and proposes a new SemVarEffect metric and SemVarBench benchmark to systematically evaluate the model's ability to capture semantic changes.
Event-Driven Online Vertical Federated Learning
Ganyu Wang (Western University), Charles Ling (Western University)
Federated LearningTabular
🎯 What it does: An event-driven online vertical federated learning framework is proposed, which activates only a portion of clients when an event is triggered, while the remaining clients passively participate.
Everything is Editable: Extend Knowledge Editing to Unstructured Data in Large Language Models
Jingcheng Deng (Institute of Computing Technology, Chinese Academy of Sciences), Xueqi Cheng (Institute of Computing Technology, Chinese Academy of Sciences)
TransformerLarge Language ModelTextBenchmark
🎯 What it does: An editing method for unstructured knowledge called UnKE has been proposed, along with the release of the UnKEBench benchmark.
Everything, Everywhere, All at Once: Is Mechanistic Interpretability Identifiable?
Maxime Méloux (Université Grenoble Alpes), Maxime Peyrard (Université Grenoble Alpes)
Explainability and InterpretabilityTabular
🎯 What it does: This paper systematically investigates whether mechanistic interpretability can produce a unique explanation under given behaviors. It utilizes an enumerable small MLP to comprehensively enumerate and evaluate circuits, algorithms, and their mappings while learning Boolean functions (such as XOR), revealing a general issue of non-uniqueness.
Evidential Learning-based Certainty Estimation for Robust Dense Feature Matching
Lile Cai (Institute for Infocomm Research A*STAR), Xulei Yang (Institute for Infocomm Research A*STAR)
RecognitionAdversarial AttackImage
🎯 What it does: This paper proposes the use of evidence deep learning for confidence estimation in dense feature matching to enhance the model's robustness against image distortion and adversarial attacks.
Exact Byte-Level Probabilities from Tokenized Language Models for FIM-Tasks and Model Ensembles
Buu Phan (University of Toronto), Karen Ullrich (Meta AI)
GenerationOptimizationTransformerLarge Language ModelText
🎯 What it does: This study investigates the impact of tokenization on language models, introduces the concept of 'tokenization bias', and presents the Byte-Token Representation Lemma. Based on this theory, an O(1) algorithm is designed to convert any pre-trained tokenized language model into an equivalent byte-level model without the need for retraining, and it is applied in the Fill-in-the-Middle (FIM) task and model integration.
Exact Certification of (Graph) Neural Networks Against Label Poisoning
Mahalakshmi Sabanayagam (Technical University of Munich), Debarghya Ghoshdastidar (Technical University of Munich)
Graph Neural NetworkGraph
🎯 What it does: An exact robustness proof against label flipping attacks on Graph Neural Networks (GNN) is proposed, providing both sample-level certificates and collective certificates.
Exact Community Recovery under Side Information: Optimality of Spectral Algorithms
Julia Gaudio (Northwestern University), Nirmit Joshi (Toyota Technological Institute at Chicago)
Graph
🎯 What it does: This paper proposes a single-stage algorithm based on spectral decomposition, which can achieve precise community recovery in a two-community block model (including Bernoulli SBM and Gaussian ROS) with node attribute side information, while maintaining optimality at the information-theoretic limit.
Exact Computation of Any-Order Shapley Interactions for Graph Neural Networks
Maximilian Muschalik (Ludwig Maximilian University of Munich), Barbara Hammer (Bielefeld University)
Explainability and InterpretabilityComputational EfficiencyGraph Neural NetworkGraph
🎯 What it does: This paper proposes a precise calculation method for Shapley interactions (SIs) in graph prediction tasks of Graph Neural Networks (GNN), called GraphSHAP-IQ, and visualizes node-level interactions in the form of SI-Graph.
ExACT: Teaching AI Agents to Explore with Reflective-MCTS and Exploratory Learning
Xiao Yu (Columbia University), Zhou Yu (Columbia University)
Large Language ModelReinforcement LearningContrastive LearningTextBenchmark
🎯 What it does: A R-MCTS agent based on MCTS is proposed, which enhances exploration and evaluation capabilities through reflective learning and multi-agent debate; at the same time, exploratory learning (EL) is designed to transfer search experience back to GPT-4o;
Examining Alignment of Large Language Models through Representative Heuristics: the case of political stereotypes
Sullam Jeoung (University of Illinois at Urbana-Champaign), Jana Diesner (Technical University of Munich)
TransformerLarge Language ModelPrompt EngineeringText
🎯 What it does: From the perspective of representativeness heuristics, this paper systematically evaluates whether the outputs of large language models (LLMs) align with human values on political issues, quantitatively analyzing the extent and conditions of their deviation from empirical positions.
Execution-guided within-prompt search for programming-by-example
Gust Verbruggen (Microsoft), Sumit Gulwani (Microsoft)
AI Code AssistantTransformerLarge Language ModelPrompt EngineeringTextBenchmark
🎯 What it does: A search method for executing guidance within prompts is proposed, utilizing LLM to generate multiple lines of code and merging them into a program, continuing to expand in the next round, thus achieving search and value assessment in programmatic examples;
Expand and Compress: Exploring Tuning Principles for Continual Spatio-Temporal Graph Forecasting
Wei Chen (Hong Kong University of Science and Technology), Yuxuan Liang (Hong Kong University of Science and Technology)
Graph Neural NetworkPrompt EngineeringGraphTime Series
🎯 What it does: A continuous spatiotemporal graph prediction framework EAC based on prompt learning is proposed to address the issues of model expansion and catastrophic forgetting caused by the addition of new sensors.
Expected Return Symmetries
Darius Muglich (University of Oxford), Jakob Nicolaus Foerster
Reinforcement LearningSequential
🎯 What it does: This paper defines and studies 'Expected Return Symmetry' (ER symmetry) and combines it with the Other-Play (OP) training objective to enhance the performance of Zero-Shot Coordination (ZSC) in multi-agent settings.
Expected Sliced Transport Plans
Xinran Liu (Vanderbilt University), Soheil Kolouri (Vanderbilt University)
Point Cloud
🎯 What it does: A slice-based Expectation Slice Transport (EST) scheme and metric are constructed to obtain explicit transport plans and distances between discrete probability measures.
Explain Yourself, Briefly! Self-Explaining Neural Networks with Concise Sufficient Reasons
Shahaf Bassan (IBM Research), Shlomit Gur (IBM Research)
Explainability and InterpretabilityConvolutional Neural NetworkTransformerSupervised Fine-TuningImageText
🎯 What it does: A self-supervised training framework SST is proposed, allowing neural networks to directly provide minimal sufficient reasons in their outputs, thus avoiding post-hoc calculations.
Explaining Modern Gated-Linear RNNs via a Unified Implicit Attention Formulation
Itamar Zimerman (Tel Aviv University), Lior Wolf (Tel Aviv University)
Explainability and InterpretabilityRecurrent Neural NetworkTransformerImageText
🎯 What it does: A unified implicit attention representation is proposed, which explains and visualizes the internal mechanisms of modern gated linear RNNs (such as Mamba, RWKV, Griffin, etc.) and Transformers, and develops new interpretability methods based on this representation.
Explanations of GNN on Evolving Graphs via Axiomatic Layer edges
Yazheng Liu (Hong Kong University of Science and Technology), Sihong Xie (Hong Kong University of Science and Technology)
OptimizationExplainability and InterpretabilityGraph Neural NetworkGraph
🎯 What it does: This study investigates methods for explaining GNN predictions on evolving graphs with continuously changing edge weights, proposing a layer-edge-based explanation framework.
Exploiting Distribution Constraints for Scalable and Efficient Image Retrieval
Mohammad Omama (University of Texas at Austin), Sandeep P. Chinchali (University of Texas at Austin)
RetrievalOptimizationComputational EfficiencyKnowledge DistillationTransformerAuto EncoderImage
🎯 What it does: This paper proposes two methods, AE-SVC and SS-D2, to enhance the scalability and efficiency of foundational models (such as DINO, DINOv2, CLIP, ViT) in image retrieval tasks without the need for labeled data.
Exploiting Hidden Symmetry to Improve Objective Perturbation for DP Linear Learners with a Nonsmooth L1-Norm
Du Chen (Nanyang Technological University), Geoffrey A. Chua (Nanyang Technological University)
OptimizationSafty and PrivacyTabular
🎯 What it does: This paper proposes an Objective Perturbation algorithm C-OP that uses convolution smoothing for non-smooth differential privacy convex optimization problems with a hidden ≡ 1 structure.
Exploiting Structure in Offline Multi-Agent RL: The Benefits of Low Interaction Rank
Wenhao Zhan (Princeton University), Yonathan Efroni (Meta)
Reinforcement Learning
🎯 What it does: This study investigates the use of low-interaction rank structures in offline multi-agent reinforcement learning to approximate equilibrium learning. It proposes an algorithm that combines decentralization, regularization, and no-regret learning (DR-AC), and provides theoretical guarantees on sample complexity.
Exploratory Preference Optimization: Harnessing Implicit Q*-Approximation for Sample-Efficient RLHF
Tengyang Xie (University of Wisconsin Madison), Alexander Rakhlin (Massachusetts Institute of Technology)
OptimizationReinforcement Learning from Human FeedbackReinforcement Learning
🎯 What it does: This paper proposes Exploratory Preference Optimization (XPO), an online optimization method that incorporates exploration bonuses into RLHF, capable of learning approximately optimal policies with limited samples.
Explore Theory of Mind: program-guided adversarial data generation for theory of mind reasoning
Melanie Sclar (University of Washington), Asli Celikyilmaz (Meta)
GenerationData SynthesisTransformerLarge Language ModelSupervised Fine-TuningText
🎯 What it does: A framework called EXPLORETOM driven by A* search has been constructed to generate diverse and challenging Theory of Mind (ToM) stories and questions on a large scale.
Exploring a Principled Framework for Deep Subspace Clustering
Xianghan Meng (Beijing University of Posts and Telecommunications), Chun-Guang Li (Beijing University of Posts and Telecommunications)
Representation LearningImage
🎯 What it does: A new framework PRO-DSC is proposed, which jointly learns the subspace structure and self-expressive coefficients to address the feature collapse problem encountered in traditional deep subspace clustering;
Exploring channel distinguishability in local neighborhoods of the model space in quantum neural networks
Sabrina Herbst (TU Wien), Ivona Brandić (TU Wien)
TabularPhysics Related
🎯 What it does: This paper reveals that the channel distinguishability caused by small updates to the parameters of the ansatz in quantum neural networks leads to the existing hardware-efficient ansatz being almost indistinguishable in local neighborhoods, resulting in training difficulties.
Exploring Learning Complexity for Efficient Downstream Dataset Pruning
Wenyu Jiang (Southern University of Science and Technology), Hongxin Wei (Southern University of Science and Technology)
Computational EfficiencyData-Centric LearningConvolutional Neural NetworkTransformerSupervised Fine-TuningReinforcement LearningImageText
🎯 What it does: A training-independent sample difficulty scoring method called Distorted-based Learning Complexity (DLC) and a flexible downsampling strategy based on randomness called FlexRand are proposed for efficient pruning of downstream datasets.
Exploring Local Memorization in Diffusion Models via Bright Ending Attention
Chen Chen (University of Sydney), Chang Xu (University of Sydney)
Diffusion modelImage
🎯 What it does: This study investigates the local memory problem in diffusion models, proposing the Bright Ending (BE) mechanism for evaluating, detecting, and mitigating memory.
Exploring Prosocial Irrationality for LLM Agents: A Social Cognition View
Xuan Liu (Hong Kong Polytechnic University), Quanyan Zhu (New York University)
TransformerLarge Language ModelPrompt EngineeringText
🎯 What it does: The CogMir framework is proposed, utilizing the systematic illusion properties of LLMs to simulate human cognitive biases and assess the rationality and prosocial decision-making of LLM agents in social contexts.
Exploring the Camera Bias of Person Re-identification
Myungseo Song (mAy-I Inc.), Jong-Seok Lee (Yonsei University)
RecognitionRetrievalDomain AdaptationContrastive LearningImage
🎯 What it does: This paper systematically evaluates the impact of camera bias on unseen domains in the person re-identification task and proposes a simple yet effective debiasing method through feature vector normalization (camera-specific mean subtraction + variance normalization). It also analyzes the risks of camera bias pseudo-labels during the unsupervised learning process and provides training strategies for removing single-camera clusters and debiasing pseudo-labels.
Exploring the Design Space of Visual Context Representation in Video MLLMs
Yifan Du (Renmin University of China), Ji-Rong Wen (Renmin University of China)
OptimizationRepresentation LearningTransformerLarge Language ModelVision Language ModelVideoMultimodality
🎯 What it does: This paper systematically studies the visual context representation in video multimodal large language models (Video-MLLM), specifically how to select the number of video frames and the number of visual embeddings per frame under a fixed context window, modeling it as a constrained optimization problem to derive the optimal allocation.
Exploring the Effectiveness of Object-Centric Representations in Visual Question Answering: Comparative Insights with Foundation Models
Amir Mohammad Karimi Mamaghan (KTH Royal Institute of Technology), Andrea Dittadi (Technical University of Munich)
ClassificationObject DetectionRepresentation LearningTransformerAuto EncoderContrastive LearningImage
🎯 What it does: This paper systematically evaluates the effects and differences of Object-Centric (OC) representation and foundation models in reasoning through large-scale experiments on the multi-object visual question answering (VQA) task.
Exploring The Forgetting in Adversarial Training: A Novel Method for Enhancing Robustness
Xianglu Wang (University of Science and Technology of China), Hu Ding (University of Science and Technology of China)
ClassificationKnowledge DistillationAdversarial AttackConvolutional Neural NetworkGenerative Adversarial NetworkImage
🎯 What it does: This paper studies adversarial training from the perspective of catastrophic forgetting and proposes the Adaptive Multi-Teacher Self-Distillation (AMS) method to alleviate forgetting during the adversarial training process, thereby enhancing robustness.
Exploring The Loss Landscape Of Regularized Neural Networks Via Convex Duality
Sungyoon Kim (Stanford University), Mert Pilanci (Stanford University)
OptimizationTabular
🎯 What it does: This paper transforms the non-convex training problem of a regularized two-layer ReLU network into an equivalent convex problem, analyzes its dual, and comprehensively characterizes the loss landscape of the regularized network and the structure of the global optimal solution set.
Exponential Topology-enabled Scalable Communication in Multi-agent Reinforcement Learning
Xinran Li (Hong Kong University of Science and Technology), Jun Zhang (Hong Kong University of Science and Technology)
Recurrent Neural NetworkReinforcement LearningContrastive LearningTabular
🎯 What it does: A scalable communication protocol called ExpoComm based on exponential topology is proposed for large-scale cooperative reinforcement learning.
Exposure Bracketing Is All You Need For A High-Quality Image
Zhilu Zhang (Harbin Institute of Technology), Wangmeng Zuo (Harbin Institute of Technology)
RestorationSuper ResolutionRecurrent Neural NetworkSupervised Fine-TuningImageVideo
🎯 What it does: This paper proposes a unified high-quality night scene image restoration method by integrating denoising, deblurring, HDR reconstruction, and super-resolution of multi-exposure images through exposure segment shooting.
Expressivity of Neural Networks with Random Weights and Learned Biases
Ezekiel Williams (University of Montreal), Guillaume Lajoie (University of Montreal)
Recurrent Neural NetworkTime Series
🎯 What it does: This study proves that in feedforward and recurrent neural networks with fixed weights distributed randomly, it is possible to approximate any continuous function and finite-time dynamics with high probability by learning only the biases.
Extendable and Iterative Structure Learning Strategy for Bayesian Networks
Hamid Kalantari (University of Alberta), Pouria Ramazi (Brock University)
Score-based ModelGraph
🎯 What it does: A scalable and iterative Bayesian network structure learning method is proposed, which can efficiently add new variables based on an existing structure and gradually build a complete network through an iterative approach.
Extending Mercer's expansion to indefinite and asymmetric kernels
Sungwoo Jeong (Cornell University), Alex Townsend (Cornell University)
🎯 What it does: A general theory of Mercer expansion is proposed, proving that continuous asymmetric or indefinite kernels converge pointwise almost everywhere, unconditionally, and almost uniformly under the condition of having uniformly bounded variation, and new singular value decay bounds are provided.
F-Fidelity: A Robust Framework for Faithfulness Evaluation of Explainable AI
Xu Zheng (Florida International University), Dongsheng Luo (Florida International University)
Explainability and InterpretabilitySupervised Fine-TuningImageTextTime Series
🎯 What it does: A framework called F-Fidelity based on fine-grained tuning and random masking is proposed to evaluate the trustworthiness of explainable AI.
FaceShot: Bring Any Character into Life
Junyao Gao (Tongji University), Cairong Zhao (Tongji University)
GenerationData SynthesisDiffusion modelVideoBenchmark
🎯 What it does: FaceShot proposes a training-free portrait animation framework that can match any character with any driving video to generate realistic animations.
Facilitating Multi-turn Function Calling for LLMs via Compositional Instruction Tuning
Mingyang Chen (Baichuan Inc.), weipeng chen
TransformerLarge Language ModelSupervised Fine-TuningPrompt EngineeringText
🎯 What it does: This paper proposes the BUTTON method, which generates 8,000 multi-turn function call instruction tuning data through bottom-up task construction and top-down trajectory generation.
Factor Graph-based Interpretable Neural Networks
Yicong Li (Dalian University of Technology), Feng Xia (RMIT University)
Explainability and InterpretabilityTabularBiomedical DataElectronic Health Records
🎯 What it does: This paper proposes AGAIN, which utilizes factor graphs to encode logical rules, identifying and correcting logical errors in concept explanations during the inference phase, thereby generating understandable explanations under unknown disturbances.
FACTS: A Factored State-Space Framework for World Modelling
Li Nanbo (King Abdullah University of Science and Technology), Jürgen Schmidhuber
World ModelTime SeriesSequential
🎯 What it does: A variable memory state space model FACTS is proposed for spatio-temporal world modeling.
Failures to Find Transferable Image Jailbreaks Between Vision-Language Models
Rylan Schaeffer (Stanford University), Ethan Perez (Anthropic)
Adversarial AttackVision Language ModelImageText
🎯 What it does: This study systematically evaluates the transferability of gradient-based image jailbreak attacks across more than 40 open-source visual language models (VLMs) and compares the effectiveness of single model attacks with model ensemble attacks.
Fair Clustering in the Sliding Window Model
Vincent Cohen-Addad (Google), Samson Zhou (Texas A&M University)
TabularBenchmark
🎯 What it does: This paper studies the problem of fair clustering under a sliding window streaming model, proving that any polynomial approximation requires linear space, and proposes a sliding window algorithm that achieves (1+ε) approximate fair clustering in sublinear space.
Fair Submodular Cover
Wenjing Chen (Texas A&M University), Victoria G. Crawford (Texas A&M University)
OptimizationTabular
🎯 What it does: The Fair Submodular Cover problem is proposed, and a bi-approximation algorithm is provided that can approximately minimize the set size while satisfying fairness ratio constraints.
FairDen: Fair Density-Based Clustering
Lena Krieger (Forschungszentrum Julich), Ira Assent (Aarhus University)
TabularBenchmark
🎯 What it does: A new density-based fair clustering algorithm, FairDen, is proposed, which achieves group-level fair clustering while maintaining the data's density-connected structure.
FairMT-Bench: Benchmarking Fairness for Multi-turn Dialogue in Conversational LLMs
Zhiting Fan (Zhejiang University), Zuozhu Liu (Zhejiang University)
TransformerLarge Language ModelTextBenchmark
🎯 What it does: A multi-turn dialogue fairness evaluation benchmark, FairMT-Bench, has been constructed, which includes 10k multi-turn dialogue data (FairMT-10K) and its selected most challenging 1k samples (FairMT-1K), along with a unified evaluation framework.
FaithEval: Can Your Language Model Stay Faithful to Context, Even If "The Moon is Made of Marshmallows"
Yifei Ming (Salesforce AI Research), Shafiq Joty (Salesforce AI Research)
GenerationRetrievalTransformerLarge Language ModelPrompt EngineeringTextBenchmarkRetrieval-Augmented GenerationChain-of-Thought
🎯 What it does: The FaithEval benchmark is proposed to evaluate the fidelity of large language models in retrieval-augmented generation (RAG) scenarios, constructing three types of tasks: unanswerable, inconsistent, and counterfactual.
FakeShield: Explainable Image Forgery Detection and Localization via Multi-modal Large Language Models
Zhipei Xu (Peking University), Jian Zhang (Peking University)
Anomaly DetectionExplainability and InterpretabilityTransformerLarge Language ModelSupervised Fine-TuningImageMultimodality
🎯 What it does: Proposes a multimodal explainable image forgery detection and localization framework called FakeShield, which integrates detection, localization, and textual explanation;
Fantastic Copyrighted Beasts and How (Not) to Generate Them
Luxi He (Princeton University), Peter Henderson (Princeton University)
GenerationTransformerLarge Language ModelPrompt EngineeringDiffusion modelImageVideoText
🎯 What it does: This study investigates the risks of text-to-image/video generation models when generating copyrighted characters, proposing an evaluation framework that balances copyright protection and user intent, systematically identifying indirect anchors, and assessing and improving existing mitigation strategies (prompt rewriting, negative prompting).
Fantastic Targets for Concept Erasure in Diffusion Models and Where To Find Them
Anh Tuan Bui (Monash University), Dinh Phung (Monash University)
GenerationOptimizationDiffusion modelImage
🎯 What it does: Proposes the Adaptive Guided Erasure (AGE) method, which dynamically selects target concepts and fine-tunes the diffusion model under a minimax optimization framework to efficiently eliminate undesirable concepts while maximizing the retention of other concepts.
Fast and Accurate Blind Flexible Docking
Zizhuo Zhang (Hong Kong Baptist University), Bo Han (Hong Kong Baptist University)
Computational EfficiencyDrug DiscoveryGraph Neural NetworkBiomedical Data
🎯 What it does: Proposes FABFlex, a regression-based multi-task learning model for blind flexible docking;
Fast and Slow Streams for Online Time Series Forecasting Without Information Leakage
Ying-yee Ava Lau (Hong Kong University of Science and Technology), Dit-Yan Yeung (Hong Kong University of Science and Technology)
Time Series
🎯 What it does: A dual-stream online time series forecasting framework (DSOF) is proposed, redefining the online time series forecasting task and eliminating the information leakage problem.
Fast Direct: Query-Efficient Online Black-box Guidance for Diffusion-model Target Generation
Kim Yong Tan (Nanyang Technological University), Yew-Soon Ong (Nanyang Technological University)
GenerationOptimizationDrug DiscoveryDiffusion modelImageBiomedical Data
🎯 What it does: This paper proposes an online, query-efficient black-box target generation algorithm called Fast Direct, which utilizes a pre-trained diffusion model to guide noise sequences through pseudo-targets during inference, enabling the generation of samples that meet specific objectives (such as image alignment or molecular binding affinity) within a limited query budget.
Fast Feedforward 3D Gaussian Splatting Compression
Yihang Chen (Shanghai Jiao Tong University), Jianfei Cai (Monash University)
CompressionAuto EncoderGaussian SplattingPoint Cloud
🎯 What it does: A framework for fast compression of 3D Gaussian Splatting (3DGS) representation without optimization (FCGS) is proposed, which can complete compression in a single forward inference, significantly reducing compression time.
Fast Summation of Radial Kernels via QMC Slicing
Johannes Hertrich (University Paris Dauphine PSL), Michael Quellmalz (Technische Universitat Berlin)
OptimizationComputational EfficiencyImage
🎯 What it does: A fast kernel summation method based on random projection and one-dimensional kernel summation—slicing—is proposed, and quasi-Monte Carlo (QMC) design on the sphere is introduced to improve the accuracy of slicing, further deriving error upper bounds and variance analysis.
Fast training and sampling of Restricted Boltzmann Machines
Nicolas BEREUX, Beatriz Seoane (Universidad Complutense de Madrid)
Reinforcement LearningGenerative Adversarial NetworkImageTabular
🎯 What it does: This paper proposes the use of the natural evolution of parameters during the RBM training process (trajectory annealing) to achieve efficient log-likelihood estimation and sampling, and addresses the initialization problem of highly structured data through low-rank RBM pre-training.
Fast Training of Sinusoidal Neural Fields via Scaling Initialization
Taesun Yeom (Pohang University of Science and Technology), Jaeho Lee (Pohang University of Science and Technology)
Super ResolutionOptimizationComputational EfficiencyNeural Radiance FieldImageMultimodalityAudio
🎯 What it does: This paper studies the initialization method of Sine Neural Fields (SNF) and proposes a simple weight scaling (WS) scheme, which multiplies all weights except for the last layer by a constant, significantly improving training speed.
Fast Uncovering of Protein Sequence Diversity from Structure
luca alessandro silva, Christoph Feinauer (Bocconi University)
Protein Structure PredictionGraph Neural NetworkTransformerBiomedical Data
🎯 What it does: A reverse folding method called InvMSAFold is designed, which uses a single forward propagation to obtain lightweight Potts model parameters, and then efficiently generates diverse protein sequences on the CPU.
Fast unsupervised ground metric learning with tree-Wasserstein distance
Kira Michaela Düsterwald (University College London), Makoto Yamada (Okinawa Institute of Science and Technology)
OptimizationComputational EfficiencyBiomedical Data
🎯 What it does: A tree-Wasserstein distance-based unsupervised benchmark metric learning method, Tree-WSV, is proposed to quickly estimate the distance between samples and features.
Faster Algorithms for Structured Linear and Kernel Support Vector Machines
Yuzhou Gu (New York University), Lichen Zhang (Massachusetts Institute of Technology)
Optimization
🎯 What it does: This paper proposes a near-linear time algorithm for solving structured quadratic programming (such as linear and Gaussian kernel support vector machines), capable of quickly solving problems with high precision.
Faster Cascades via Speculative Decoding
Harikrishna Narasimhan (Google Research), Sanjiv Kumar (Google Research)
OptimizationComputational EfficiencyTransformerLarge Language ModelText
🎯 What it does: A new inference strategy combining model hierarchy (small models and large models) with speculative decoding, called Speculative Cascades, is designed to reduce inference costs while maintaining quality.
Faster Diffusion Sampling with Randomized Midpoints: Sequential and Parallel
Shivam Gupta (University of Texas at Austin), Sitan Chen (Harvard University)
Diffusion modelOrdinary Differential Equation
🎯 What it does: A new sampling scheme is proposed, based on the random midpoint method, aimed at improving the sampling efficiency of diffusion models.
Faster Inference of Flow-Based Generative Models via Improved Data-Noise Coupling
Aram Davtyan (University of Bern), Paolo Favaro (University of Bern)
GenerationData SynthesisOptimizationComputational EfficiencyFlow-based ModelImage
🎯 What it does: Accelerate the sampling of the CFM model by improving the optimal transport allocation of data-noise coupling.
FasterCache: Training-Free Video Diffusion Model Acceleration with High Quality
Zhengyao Lv (University of Hong Kong), Kwan-Yee K. Wong (University of Hong Kong)
GenerationData SynthesisComputational EfficiencyTransformerDiffusion modelVideoBenchmark
🎯 What it does: FasterCache is proposed, a training-independent acceleration strategy to enhance the inference speed of video diffusion models while maintaining high-quality generation.
Fat-to-Thin Policy Optimization: Offline Reinforcement Learning with Sparse Policies
Lingwei Zhu (University of Tokyo), Yukie Nagai (University of Tokyo)
OptimizationReinforcement LearningBiomedical Data
🎯 What it does: This paper proposes an offline reinforcement learning algorithm named FtTPO, which learns sparse continuous policies from log data using a two-stage fat-to-thin strategy.
Feast Your Eyes: Mixture-of-Resolution Adaptation for Multimodal Large Language Models
Gen Luo (Xiamen University), Rongrong Ji (Xiamen University)
RecognitionObject DetectionComputational EfficiencyTransformerLarge Language ModelSupervised Fine-TuningVision Language ModelImageTextMultimodality
🎯 What it does: This paper proposes a Mixed Resolution Adaptation (MRA) method based on dual visual channels, utilizing high-resolution information to enhance the fine-grained visual understanding capabilities of multimodal large language models, resulting in the new model LLaVA-HR.
Feature Averaging: An Implicit Bias of Gradient Descent Leading to Non-Robustness in Neural Networks
Binghui Li (Peking University), Jian Li (Tsinghua University)
OptimizationAdversarial AttackConvolutional Neural NetworkSupervised Fine-TuningImage
🎯 What it does: This study investigates the implicit bias of feature averaging that arises during the training of two-layer ReLU networks using gradient descent, and proves that this bias leads to a decrease in the robustness of the network on multi-cluster orthogonal feature data.
Feature Responsiveness Scores: Model-Agnostic Explanations for Recourse
Seung Hyun Cheon (University of California San Diego), Berk Ustun (University of California San Diego)
Recommendation SystemExplainability and InterpretabilityTabularFinance Related
🎯 What it does: This paper proposes a feature responsiveness-based explanation method for achieving actionable model explanations in fields such as consumer finance.
Feature-Based Online Bilateral Trade
Solenne Gaucher (Ecole Polytechnique), Vianney Perchet (Criteo AI Lab)
Recommendation SystemOptimizationReinforcement LearningTabularFinance Related
🎯 What it does: A feature-based online bilateral trading model is proposed, which studies how to set prices for sellers and buyers to maximize trading profits given the context vector observed at each time step.
Federated $Q$-Learning with Reference-Advantage Decomposition: Almost Optimal Regret and Logarithmic Communication Cost
Zhong Zheng (Pennsylvania State University), Lingzhou Xue (Pennsylvania State University)
Federated LearningReinforcement LearningTabular
🎯 What it does: A federated reinforcement learning algorithm named FedQ-Advantage is proposed for multi-agent collaborative learning of optimal policies in discrete episodic Markov Decision Processes (MDP) without sharing raw trajectories.
Federated Class-Incremental Learning: A Hybrid Approach Using Latent Exemplars and Data-Free Techniques to Address Local and Global Forgetting
Milad Khademi Nori (Toronto Metropolitan University), Guanghui Wang (Toronto Metropolitan University)
Federated LearningKnowledge DistillationAuto EncoderImage
🎯 What it does: A hybrid replay method is proposed in federated incremental learning through a customized autoencoder to alleviate both local and global forgetting.
Federated Continual Learning Goes Online: Uncertainty-Aware Memory Management for Vision Tasks and Beyond
Giuseppe Serra (Goethe University Frankfurt), Florian Buettner (German Cancer Research Center)
Federated LearningConvolutional Neural NetworkRecurrent Neural NetworkImageTextBiomedical Data
🎯 What it does: An online federated continual learning (online-FCL) framework is proposed, which addresses the problem of catastrophic forgetting in continuous streaming data by selecting representative samples for replay through uncertainty-aware memory management based on Bregman information (BI) at the client side.
Federated Domain Generalization with Data-free On-server Matching Gradient
Trong Binh Nguyen, Won-Joo Hwang
Domain AdaptationFederated LearningImage
🎯 What it does: This paper proposes FedOMG, an algorithm that utilizes local gradients for gradient matching on the federated learning server to achieve a domain-invariant global model, thereby addressing the Federated Domain Generalization (FDG) problem.
Federated Few-Shot Class-Incremental Learning
Muhammad Anwar Ma'sum, Ryszard Kowalczyk (University of South Australia)
Federated LearningTransformerPrompt EngineeringImageOrdinary Differential Equation
🎯 What it does: This paper defines the problem of Federated Few-Shot Class Incremental Learning (FFSCIL) and proposes a Unified Optimization Prototype Prompt (UOPP) model to address catastrophic forgetting, overfitting, and prototype bias while maintaining data privacy.
Federated Granger Causality Learning For Interdependent Clients With State Space Representation
Ayush Mohanty (Georgia Institute of Technology), Nagi Gebraeel (Georgia Institute of Technology)
Federated LearningTime Series
🎯 What it does: A federated learning framework is proposed to jointly learn the Granger causality of multiple clients using a low-dimensional state space model, aiming to achieve causal dependency detection in distributed systems.
Federated Residual Low-Rank Adaptation of Large Language Models
Yunlu Yan (Hong Kong University of Science and Technology), Lei Zhu (Hong Kong University of Science and Technology)
Federated LearningTransformerLarge Language ModelText
🎯 What it does: In response to the data heterogeneity of large-scale language models in federated learning, a Federated Residual LoRa Adaptation (FRLoRA) method based on low-rank residual updates is proposed.
FedLWS: Federated Learning with Adaptive Layer-wise Weight Shrinking
Changlong Shi (Jilin University), Yi Chang (Jilin University)
Federated LearningConvolutional Neural NetworkTransformerImageText
🎯 What it does: In federated learning, FedLWS is proposed, which utilizes adaptive hierarchical weight shrinkage on the aggregated global model at the server side to enhance the model's generalization performance.
FedTMOS: Efficient One-Shot Federated Learning with Tsetlin Machine
Shannon How Shi Qi (University of Southampton), Jonathon Hare
Federated LearningImage
🎯 What it does: Developed FedTMOS, a single-round federated learning framework that utilizes Tsetlin machines to achieve serverless model aggregation training on edge devices.
Feedback Favors the Generalization of Neural ODEs
Jindou Jia (Beihang University), Lei Guo (Beihang University)
Time SeriesOrdinary Differential Equation
🎯 What it does: This study embeds a real-time feedback mechanism into neural ODEs, forming a two-degree-of-freedom network to real-time correct the learned latent dynamics when encountering unknown disturbances or system changes, thereby significantly enhancing the generalization performance of continuous-time tasks.