NeurIPS 2023 Papers — Page 11
Conference on Neural Information Processing Systems · 3218 papers
Equivariant Neural Simulators for Stochastic Spatiotemporal Dynamics
Koen Minartz (Eindhoven University of Technology), Vlado Menkovski (Eindhoven University of Technology)
Graph Neural NetworkGenerative Adversarial NetworkTime SeriesSequential
🎯 What it does: This paper proposes Equivariant Probabilistic Neural Simulation (EPNS), a method for stochastic spatiotemporal dynamic autoregressive probabilistic simulation that maintains symmetry constraints.
Equivariant Single View Pose Prediction Via Induced and Restriction Representations
Owen Lewis Howell (Northeastern University), Robin Walters (Northeastern University)
Pose EstimationImage
🎯 What it does: The research constructs an SO(3) equivariant network from 2D images to 3D poses using induced and constrained representations, proposing a learnable induction/constrain layer to achieve 2D to spherical projection, satisfying SO(2) consistency constraints.
Equivariant Spatio-Temporal Attentive Graph Networks to Simulate Physical Dynamics
Liming Wu (Renmin University of China), Wenbing Huang (North Carolina State University)
Graph Neural NetworkGraphTime SeriesPhysics Related
🎯 What it does: This paper proposes an Equivariant Spatiotemporal Attention Graph Network (ESTAG) that utilizes historical trajectories for physical dynamics modeling to predict future node positions, eliminating the Markov assumption in frame-to-frame predictions.
Error Bounds for Learning with Vector-Valued Random Features
Samuel Lanthaler (California Institute of Technology), Nicholas H. Nelsen (California Institute of Technology)
Tabular
🎯 What it does: This paper provides a comprehensive error analysis of learning using vector-valued random features (RF), focusing on the theory of RF ridge regression in an infinite-dimensional input-output setting, improving upon existing finite-dimensional analyses.
Error Discovery By Clustering Influence Embeddings
Fulton Wang (Meta), Narine Kokhlikyan (Meta)
Anomaly DetectionExplainability and InterpretabilityImageTextTime SeriesBiomedical Data
🎯 What it does: This paper proposes a slice discovery method based on influence embedding, called InfEmbed, to identify subsets of samples in the test set that perform poorly and share the same error causes.
Errors-in-variables Fr\'echet Regression with Low-rank Covariate Approximation
Dogyoon Song (University of Michigan), Kyunghee Han (University of Illinois at Chicago)
Tabular
🎯 What it does: This study investigates how to improve the estimation method of global Fréchet regression using low-rank covariate approximation in the presence of measurement errors in high-dimensional covariates.
Ess-InfoGAIL: Semi-supervised Imitation Learning from Imbalanced Demonstrations
Huiqiao Fu (Nanjing University), Chunlin Chen (Moonshot AI)
Robotic IntelligenceReinforcement LearningGenerative Adversarial NetworkMultimodality
🎯 What it does: The paper proposes a semi-supervised imitation learning framework called Ess-InfoGAIL, which can learn from imbalanced demonstration data and decouple multimodal behavior representations.
ESSEN: Improving Evolution State Estimation for Temporal Networks using Von Neumann Entropy
Qiyao Huang (Xiamen University), Edwin Hancock (University of York)
Graph Neural NetworkMixture of ExpertsContrastive LearningGraphTime Series
🎯 What it does: The ESSEN framework is proposed, utilizing approximate von Neumann entropy and thermodynamic temperature to sense the evolving state of time-varying networks, and achieving node representation and link prediction through entropy-aware attention, virtual evolution node learning, and a Mixture of Thermodynamic Experts decoder.
Estimating and Controlling for Equalized Odds via Sensitive Attribute Predictors
Beepul Bharti (Johns Hopkins University), Jeremias Sulam (Johns Hopkins University)
Tabular
🎯 What it does: In situations where sensitive attributes are unobservable, this paper proposes using predicted proxy sensitive attributes to estimate and control the violation of Equalized Odds in models, providing a strict worst-case upper bound and deriving an optimal post-processing linear programming method based on that bound.
Estimating Causal Effects Identifiable from a Combination of Observations and Experiments
Yonghan Jung (Purdue University), Elias Bareinboim (Columbia University)
Tabular
🎯 What it does: This paper proposes a general framework that combines multiple sets of observations and experimental samples to estimate causal effects identifiable through g-identification.
Estimating Koopman operators with sketching to provably learn large scale dynamical systems
Giacomo Meanti (Italian Institute of Technology), Lorenzo Rosasco (Italian Institute of Technology)
Time SeriesSequentialPhysics RelatedOrdinary Differential Equation
🎯 What it does: Three Nyström random projection-based Koopman operator kernel methods (KRR, PCR, RRR) are proposed, achieving scalability for large-scale dynamical system learning.
Estimating Noise Correlations Across Continuous Conditions With Wishart Processes
Amin Nejatbakhsh (New York University), Alex H Williams (Flatiron Institute)
TabularSequential
🎯 What it does: A probabilistic model based on the Wishart process is proposed and implemented to estimate the noise covariance of neurons under continuously parameterized experimental conditions, and it is capable of interpolation predictions under unseen conditions.
Estimating Propensity for Causality-based Recommendation without Exposure Data
Zhongzhou Liu (Singapore Management University), Min Wu (A*STAR)
Recommendation SystemTabularBenchmark
🎯 What it does: Proposes the PROPCARE framework, which achieves causal recommendation by estimating exposure and propensity scores with only user-item interaction data and without actual exposure and propensity scores.
Estimating Riemannian Metric with Noise-Contaminated Intrinsic Distance
Jiaming Qiu (Fred Hutchinson Cancer Center), Xiongtao Dai (University of California)
Image
🎯 What it does: This paper proposes a method for estimating the Riemannian metric of data space using noise-affected similarity measurements, and infers the metric tensor and its derivatives based on local regression.
Estimating the Rate-Distortion Function by Wasserstein Gradient Descent
Yibo Yang (University of California), Stephan Mandt (University of California)
CompressionOptimizationImagePhysics RelatedAudio
🎯 What it does: A particle method based on Wasserstein gradient descent is proposed for estimating the distortion-rate function of continuous sources.
Evaluating and Inducing Personality in Pre-trained Language Models
Guangyuan Jiang (Peking University), Yixin Zhu (Peking University)
TransformerLarge Language ModelPrompt EngineeringTextChain-of-Thought
🎯 What it does: This paper systematically evaluates and induces the personality traits of large language models by constructing the Robot Personality Inventory (MPI) and Personality Induction Prompts (P2).
Evaluating Cognitive Maps and Planning in Large Language Models with CogEval
Ida Momennejad (Microsoft Research), Jonathan Larson (Microsoft Research)
TransformerLarge Language ModelPrompt EngineeringTextGraph
🎯 What it does: The CogEval protocol is proposed for systematically evaluating the cognitive mapping and planning abilities of large language models.
Evaluating Neuron Interpretation Methods of NLP Models
Yimin Fan (Chinese University of Hong Kong), Hassan Sajjad (Dalhousie University)
Explainability and InterpretabilityTransformerText
🎯 What it does: A systematic comparison of various neuron interpretation methods is conducted, and a voting compatibility-based evaluation framework is proposed.
Evaluating Post-hoc Explanations for Graph Neural Networks via Robustness Analysis
Junfeng Fang (University of Science and Technology of China), Xiangnan He (University of Science and Technology of China)
Anomaly DetectionExplainability and InterpretabilityAdversarial AttackGraph Neural NetworkGraph
🎯 What it does: An evaluation metric OAR based on adversarial robustness and OOD reweighting is proposed to measure the reliability of subgraph explanations in graph neural networks.
Evaluating Robustness and Uncertainty of Graph Models Under Structural Distributional Shifts
Gleb Bazhenov (HSE University), Liudmila Prokhorenkova (Yandex Research)
Domain AdaptationAnomaly DetectionGraph Neural NetworkGraph
🎯 What it does: A general method is proposed to generate node-level structural distribution shifts through graph structural attributes (such as popularity, locality, and density) to evaluate the robustness and uncertainty of graph models under distribution shifts.
Evaluating the Moral Beliefs Encoded in LLMs
Nino Scherrer (FAR AI), David Blei (Columbia University)
TransformerLarge Language ModelPrompt EngineeringText
🎯 What it does: This paper studies the internal moral beliefs of large language models (LLMs) by designing and executing a large-scale moral scenario questionnaire, and quantifying the choices and uncertainties of LLMs using statistical methods.
Evaluating the Robustness of Interpretability Methods through Explanation Invariance and Equivariance
Jonathan Crabbé (University of Cambridge), Mihaela van der Schaar (University of Cambridge)
Explainability and InterpretabilityGraph Neural NetworkMultimodalityTime SeriesElectrocardiogram
🎯 What it does: A framework for assessing the robustness of interpretability methods based on model symmetry is proposed, defining invariance and equivariance metrics for explanations and providing improvement strategies.
Every Parameter Matters: Ensuring the Convergence of Federated Learning with Dynamic Heterogeneous Models Reduction
Hanhan Zhou (George Washington University), Wenbo Ding (Tsinghua University)
Federated LearningImage
🎯 What it does: This study investigates the impact of model reduction on convergence in dynamic heterogeneous federated learning, providing a unified theoretical framework and proving that under certain conditions, it can converge to the stationary point of standard FL.
EvoFed: Leveraging Evolutionary Strategies for Communication-Efficient Federated Learning
Mohammad Mahdi Rahimi (Korea Advanced Institute of Science and Technology), Jaekyun Moon (Korea Advanced Institute of Science and Technology)
OptimizationFederated LearningConvolutional Neural NetworkImage
🎯 What it does: The EvoFed framework is proposed, which combines evolutionary strategies (ES) with federated learning (FL). It utilizes shared seeds to generate the same noise population, where clients only upload the distance (fitness) between population members and the updated model after gradient updates. The server aggregates and broadcasts, allowing clients to update the model, significantly reducing communication volume.
Evolutionary Neural Architecture Search for Transformer in Knowledge Tracing
Shangshang Yang (Anhui University), Xingyi Zhang (Anhui University)
Recommendation SystemNeural Architecture SearchTransformerTabularTime Series
🎯 What it does: A knowledge tracing model based on Transformer, which introduces convolution operations into Transformer for enhanced local context modeling, and automatically selects input features and balances local/global context through evolutionary neural architecture search;
Evolving Connectivity for Recurrent Spiking Neural Networks
Guan Wang (Tsinghua University), Sen Song (Tsinghua University)
Robotic IntelligenceRecurrent Neural NetworkSpiking Neural NetworkReinforcement LearningSequential
🎯 What it does: A framework for evolutionary connectivity that uses only inference is proposed to train recurrent spiking neural networks (RSNN) with 1-bit sparse connections.
Evolving Standardization for Continual Domain Generalization over Temporal Drift
Mixue Xie (Beijing Institute of Technology), Zehui Dai (Lazada Search and Monetisation Tech)
Domain AdaptationAdversarial AttackTransformerImageText
🎯 What it does: This paper proposes a continuous domain generalization method called EvoS, which addresses domain distribution that gradually drifts over time and achieves generalization to future domains;
EvoPrompting: Language Models for Code-Level Neural Architecture Search
Angelica Chen (New York University), David So
OptimizationNeural Architecture SearchGraph Neural NetworkLarge Language ModelPrompt EngineeringTabular
🎯 What it does: This paper proposes a framework called EVOPROMPTING that uses large language models (LM) as adaptive crossover/mutation operators, integrating evolutionary search and prompt-tuning for generating and optimizing neural network architectures at the code level.
Exact Bayesian Inference on Discrete Models via Probability Generating Functions: A Probabilistic Programming Approach
Fabian Zaiser (University of Oxford), Luke Ong
OptimizationComputational EfficiencyTabularTime Series
🎯 What it does: A precise Bayesian inference framework based on probability generating functions is proposed, and an automated tool called Genfer is implemented, capable of performing exact inference on discrete models (even supporting infinite support and continuous priors).
Exact Generalization Guarantees for (Regularized) Wasserstein Distributionally Robust Models
Waïss Azizian (University of Grenoble Alpes), Jérôme Malick (University of Grenoble Alpes)
🎯 What it does: This paper presents non-asymptotic generalization guarantees for the Wasserstein Distributionally Robust Optimization (WDRO) model and its regularized version, providing a strict upper bound on the true risk that holds with high probability, and this upper bound does not contain additional error terms.
Exact Optimality of Communication-Privacy-Utility Tradeoffs in Distributed Mean Estimation
Berivan Isik (Stanford University), Albert No (Hongik University)
OptimizationFederated LearningSafty and PrivacyTabular
🎯 What it does: In scenarios such as federated learning, this study investigates distributed mean estimation under limited communication and local differential privacy (ε-LDP) constraints, and proposes an algorithm that achieves perfectly accurate optimal error.
Exact recovery and Bregman hard clustering of node-attributed Stochastic Block Model
Maximilien Dreveton (Ecole Polytechnique Federale de Lausanne), Daniel R. Figueiredo (Federal University of Rio de Janeiro)
Graph Neural NetworkGraph
🎯 What it does: This paper provides the exact recovery threshold for networks with weighted edges and attributes under the framework of the Community Structure Block Model (CSBM), and proposes an iterative hard clustering algorithm based on Bregman divergence, which can achieve efficient community detection on sparse weighted networks.
Exact Representation of Sparse Networks with Symmetric Nonnegative Embeddings
Sudhanshu Chanpuriya (University of Illinois Urbana-Champaign), Cameron N Musco
Representation LearningGraph Neural NetworkAuto EncoderGraph
🎯 What it does: A graph generation model based on symmetric non-negative matrix factorization is proposed, which assigns two non-negative embeddings (B and C) to each node, capturing homogeneity and heterogeneity edges through a logistic regression link function, and proving that the model can achieve low-rank exact representation on graphs with bounded arboricity.
Exact Verification of ReLU Neural Control Barrier Functions
Hongchao Zhang (Washington University in St. Louis), Andrew Clark (Washington University in St. Louis)
OptimizationSafty and PrivacyTime Series
🎯 What it does: A precise safety verification method for ReLU feedforward neural networks using control barrier functions (NCBF) is proposed.
Expanding Small-Scale Datasets with Guided Imagination
Yifan Zhang (National University of Singapore), Jiashi Feng (ByteDance)
GenerationData SynthesisDiffusion modelImageBiomedical Data
🎯 What it does: This paper proposes a guided imagination framework (GIF) based on a prior generative model, aimed at automatically generating labeled new samples to expand datasets on small-scale datasets.
Experiment Planning with Function Approximation
Aldo Pacchiano (Broad Institute), Emma Brunskill (Stanford University)
OptimizationReinforcement Learning from Human FeedbackReinforcement LearningTabular
🎯 What it does: This study investigates the experimental design problem using function approximation in the context of contextual multi-armed bandit problems, proposing two experimental design strategies compatible with function approximation.
Experimental Designs for Heteroskedastic Variance
Justin David Naggar Weltz (Duke University), Lalit K Jain (University of Washington)
Optimization
🎯 What it does: In the presence of heteroscedastic noise in linear experimental design, a two-stage adaptive sampling strategy is proposed for accurately estimating noise variance and identifying the optimal arm/threshold;
Expert load matters: operating networks at high accuracy and low manual effort
Sara Sangalli (ETH Zürich), Ender Konukoglu (ETH Zürich)
ClassificationOptimizationImageBiomedical Data
🎯 What it does: A new multi-class loss function called AUCOCLoss is proposed, aimed at simultaneously improving model accuracy and reducing the number of samples requiring human expert intervention.
Explain Any Concept: Segment Anything Meets Concept-Based Explanation
Ao Sun (Hong Kong University of Science and Technology), Shuai Wang (Hong Kong University of Science and Technology)
SegmentationExplainability and InterpretabilityConvolutional Neural NetworkImage
🎯 What it does: The EAC (Explain Any Concept) method is proposed, which utilizes SAM to automatically extract instance concepts from images and provides efficient and interpretable concept-level explanations for the target DNN through Shapley values and a lightweight Per-Input Equivalence (PIE) surrogate.
Explainable and Efficient Randomized Voting Rules
Soroush Ebadian (University of Toronto), Nisarg Shah (University of Toronto)
Explainability and InterpretabilityComputational EfficiencyTabular
🎯 What it does: This study investigates the improvement of voting rule efficiency by adding simple randomization steps to deterministic voting rules while maintaining interpretability.
Explainable Brain Age Prediction using coVariance Neural Networks
Saurabh Sihag (University of Pennsylvania), Alejandro Ribeiro (University of Pennsylvania)
Anomaly DetectionExplainability and InterpretabilityGraph Neural NetworkBiomedical DataMagnetic Resonance ImagingAlzheimer's Disease
🎯 What it does: Using the Variational Neural Network (VNN) to predict age based on cortical thickness features, and by analyzing the distribution of model residuals across different brain regions, we obtain interpretable brain age differences (∆‑Age) and their association with clinical scores of Alzheimer's disease (AD).
Explaining Predictive Uncertainty with Information Theoretic Shapley Values
David Watson, Ido Guy (Meta)
Explainability and InterpretabilityTextTabular
🎯 What it does: A framework based on information theory is proposed, utilizing an improved Shapley value to explain the uncertainty of model outputs, along with corresponding inference and implementation methods.
Explaining the Uncertain: Stochastic Shapley Values for Gaussian Process Models
Siu Lun Chau (CISPA Helmholtz Center for Information Security), Dino Sejdinovic (University of Adelaide)
Explainability and InterpretabilityTabular
🎯 What it does: This paper proposes two algorithms, GP-SHAP and BayesGP-SHAP, which use random Shapley values to explain Gaussian process models and further predict the explanations of other models through Shapley prior.
Explaining V1 Properties with a Biologically Constrained Deep Learning Architecture
Galen Pogoncheff (University of California), Michael Beyeler (University of California)
ClassificationRecognitionConvolutional Neural NetworkImageBenchmark
🎯 What it does: A deep convolutional network that integrates biological constraints has been constructed, systematically embedding neurological mechanisms such as center-surround antagonism, local receptive fields, tuned normalization, and cortical magnification into ResNet50 to better simulate the neural responses and modulation characteristics of the primary visual cortex V1.
Exploiting Connections between Lipschitz Structures for Certifiably Robust Deep Equilibrium Models
Aaron J Havens, Bin Hu
ClassificationOptimizationConvolutional Neural NetworkImageStochastic Differential Equation
🎯 What it does: This paper constructs a reparameterization method that preserves Lipschitz constants, mapping various common first-order Lipschitz network structures (such as MonDEQ, SLL, CPL, AOL, etc.) to Lipschitz-bounded DEQ (LBEN), achieving an improvement in the provable robustness of DEQ models.
Exploiting Contextual Objects and Relations for 3D Visual Grounding
Li Yang (State Key Laboratory of Multimodal Artificial Intelligence Systems), Weiming Hu (State Key Laboratory of Multimodal Artificial Intelligence Systems)
RecognitionObject DetectionTransformerPoint Cloud
🎯 What it does: This paper proposes a three-stage 3D visual localization framework CORE-3DVG, which achieves 3D point cloud object localization based on natural language descriptions through text-guided detection, relationship matching, and object recognition.
Exploiting Correlated Auxiliary Feedback in Parameterized Bandits
Arun Verma (National University of Singapore), Bryan Kian Hsiang Low (National University of Singapore)
Reinforcement Learning
🎯 What it does: The study improves reward estimation in parameterized multi-armed bandits by utilizing auxiliary feedback (Hybrid Reward) related to rewards, thereby reducing the total penalty for learners.
Exploiting hidden structures in non-convex games for convergence to Nash equilibrium
Iosif Sakos (Singapore University of Technology and Design), Georgios Piliouras (Singapore University of Technology and Design)
Optimization
🎯 What it does: This paper proposes a Preprocessed Gradient Descent (PHGD) algorithm for non-convex games utilizing hidden convex structures, which achieves global convergence to Nash equilibrium under hidden monotonic structures.
Explore In-Context Learning for 3D Point Cloud Understanding
Zhongbin Fang (Sun Yat-sen University), Mengyuan Liu (Peking University)
RestorationSegmentationTransformerPrompt EngineeringPoint CloudBenchmark
🎯 What it does: A framework named Point-In-Context (PIC) is proposed, which implements context learning for 3D point clouds and constructs a benchmark dataset that includes four tasks: reconstruction, denoising, registration, and part segmentation.
Explore to Generalize in Zero-Shot RL
Ev Zisselman (Technion Israel Institute of Technology), Aviv Tamar (Technion Israel Institute of Technology)
Recurrent Neural NetworkReinforcement LearningSequentialBenchmark
🎯 What it does: The ExpGen algorithm is proposed, achieving better generalization in zero-shot reinforcement learning through the integration of maximum entropy exploration and reward aggregation.
Exploring and Interacting with the Set of Good Sparse Generalized Additive Models
Chudi Zhong (Duke University), Cynthia Rudin (Duke University)
OptimizationExplainability and InterpretabilityTabularBiomedical DataElectronic Health RecordsFinance Related
🎯 What it does: This paper studies the Rashomon set of sparse additive models (GAM), proposes an efficient approximation method based on the maximum volume ellipsoid, and utilizes this approximation to address interactive tasks such as variable importance, monotonicity constraints, and model editing.
Exploring Diverse In-Context Configurations for Image Captioning
Xu Yang (Southeast University), Xin Geng (Chinese University of Hong Kong)
GenerationRetrievalTransformerVision Language ModelImageText
🎯 What it does: In the image description task, a systematic exploration of the impact of multimodal context configurations is conducted, proposing four image selection and four caption allocation strategies to construct few-shot prompt sequences.
Exploring Geometry of Blind Spots in Vision models
Sriram Balasubramanian (University of Maryland), Soheil Feizi (University of Maryland)
ClassificationRecognitionAdversarial AttackConvolutional Neural NetworkTransformerImage
🎯 What it does: This paper proposes the Level Set Traversal (LST) algorithm, which systematically explores the significant perturbation blind spots of common CNN and ViT models in the input space by performing gradient orthogonal projection on the equal confidence level set of the visual model, and demonstrates that these blind spots exhibit a star-shaped connected structure.
Exploring Loss Functions for Time-based Training Strategy in Spiking Neural Networks
Yaoyu Zhu (Peking University), Zhaofei Yu (Peking University)
Spiking Neural NetworkImage
🎯 What it does: This paper studies the relationship of loss functions under time-based training strategies, proving that rate-coded loss can be mapped to an equivalent time-coded form, and based on this, proposes a more suitable enhanced counting loss; at the same time, it transfers the scale parameter in weight normalization to threshold learning to further stabilize training.
Exploring Question Decomposition for Zero-Shot VQA
Zaid Khan (Northeastern University), Yun Fu (NEC Laboratories America)
RecognitionGenerationTransformerLarge Language ModelVision Language ModelImageMultimodality
🎯 What it does: This paper proposes the use of question decomposition as a reasoning strategy in zero-shot visual question answering (VQA), leveraging large-scale vision-language models (VLM) to generate and utilize decompositions without additional training.
Exploring the Optimal Choice for Generative Processes in Diffusion Models: Ordinary vs Stochastic Differential Equations
Yu Cao (Shanghai Jiao Tong University), Xiang ZHOU
GenerationData SynthesisDiffusion modelImageStochastic Differential EquationOrdinary Differential Equation
🎯 What it does: Analyzed the differences in sampling quality between ODE (probability flow) and SDE (stochastic diffusion) in diffusion models, providing a theoretical derivation of error propagation under different noise intensities.
Exponential Lower Bounds for Fictitious Play in Potential Games
Ioannis Panageas (University of California, Irvine), Volkan Cevher (EPFL)
Optimization
🎯 What it does: A class of potential games is constructed, and it is proven that Fictitious Play requires exponential time to converge to Nash equilibrium in such games.
Exponentially Convergent Algorithms for Supervised Matrix Factorization
Joowon Lee (University of Wisconsin Madison), Weixin Yao (University of California Riverside)
ClassificationOptimizationTextBiomedical Data
🎯 What it does: An exponentially convergent supervised matrix factorization (SMF) training algorithm is proposed.
Exposing Attention Glitches with Flip-Flop Language Modeling
Bingbin Liu (Carnegie Mellon University), Cyril Zhang (Microsoft Research)
Recurrent Neural NetworkTransformerTextSequential
🎯 What it does: A Flip-Flop language model (FFLM) based on single-bit storage operations is proposed as a minimal synthetic benchmark to test the long-range reasoning capabilities of Transformers, revealing the frequent occurrence of the attention glitch problem in long sequence reasoning.
Exposing flaws of generative model evaluation metrics and their unfair treatment of diffusion models
George Stein (Layer 6 AI), Gabriel Loaiza-Ganem (Layer 6 AI)
GenerationData SynthesisDiffusion modelGenerative Adversarial NetworkImage
🎯 What it does: This paper evaluates 41 models based on four types of datasets (CIFAR10, ImageNet1k, FFHQ, LSUN-Bedroom) including diffusion, GAN, VAE, flow, Transformer, and consistency models through large-scale human experiments and multidimensional metrics. It systematically verifies the low correlation between traditional metrics (such as FID) and human real perception, and proposes that replacing Inception-V3 with a self-supervised feature extractor (especially DINOv2-ViT-L/14) significantly enhances the correlation between metrics and human evaluations.
Expressive probabilistic sampling in recurrent neural networks
Shirui Chen (University of Washington), Eric Todd SheaBrown
Recurrent Neural NetworkScore-based ModelSequentialStochastic Differential Equation
🎯 What it does: This paper studies the ability of recurrent neural networks to sample complex probability distributions and proposes a 'Reservoir-Sampling Network (RSN)' that can achieve sampling from any distribution, along with corresponding theoretical proofs and efficient learning algorithms.
Expressive Sign Equivariant Networks for Spectral Geometric Learning
Derek Lim (Massachusetts Institute of Technology), Haggai Maron (Technion)
OptimizationRepresentation LearningGraph Neural NetworkPoint CloudGraph
🎯 What it does: Proposed and implemented a neural network architecture for feature vector sign equivariance, addressing the expressive limitations of traditional symmetry (sign/basis invariance) in multi-node representation, link prediction, and orthogonal equivariance tasks.
Expressivity-Preserving GNN Simulation
Fabian Jogl (Vienna University of Technology), Thomas Gärtner
Graph Neural NetworkGraph
🎯 What it does: This paper studies how to simulate various non-standard message passing graph neural networks (GNNs) through graph transformations in standard message passing (MPNN) without losing expressiveness, proposing the concepts of strong/weak simulation and providing feasible automated transformation methods.
ExPT: Synthetic Pretraining for Few-Shot Experimental Design
Tung Nguyen (University of California, Los Angeles), Aditya Grover (University of California, Los Angeles)
OptimizationTransformerAuto EncoderGenerative Adversarial Network
🎯 What it does: A foundational model for few-shot experimental design, ExPT, is proposed, which utilizes unlabeled data for synthetic pre-training and quickly adapts to the target function through contextual learning.
Extending the Design Space of Graph Neural Networks by Rethinking Folklore Weisfeiler-Lehman
Jiarui Feng (Washington University in St. Louis), Yixin Chen (Washington University in St. Louis)
Graph Neural NetworkGraph
🎯 What it does: A general (k, t)-FWL+ framework is proposed, which extends the tuple aggregation method and neighborhood definition of k-FWL, and implements a practical N2-GNN under this framework, achieving an expressive power close to 3-WL with only O(n²) space.
Extensible Prompts for Language Models on Zero-shot Language Style Customization
Tao Ge (Microsoft), Furu Wei (Microsoft)
GenerationDomain AdaptationTransformerLarge Language ModelPrompt EngineeringText
🎯 What it does: A scalable prompt (X-Prompt) mechanism is proposed, allowing the addition of learnable 'virtual words' beyond natural language prompts, enabling large language models to generate text in a specified language style in a zero-shot manner.
Extracting Reward Functions from Diffusion Models
Felipe Pinto Coelho Nuti (University of Oxford), Joao F. Henriques (University of Oxford)
OptimizationReinforcement Learning from Human FeedbackReinforcement LearningDiffusion modelImage
🎯 What it does: By comparing the diffusion processes of the base model and the expert model, a relative reward function is extracted.
Extraction and Recovery of Spatio-Temporal Structure in Latent Dynamics Alignment with Diffusion Models
Yule Wang (Georgia Institute of Technology), Anqi Wu (Georgia Institute of Technology)
Domain AdaptationTransformerDiffusion modelAuto EncoderTime Series
🎯 What it does: Proposes the ERDiff method, which extracts the complete spatiotemporal structure of latent dynamics using a diffusion model in the source domain, and then recovers this structure through maximum likelihood alignment in the target domain, achieving unsupervised neural distribution alignment.
Extremal Domain Translation with Neural Optimal Transport
Milena Gazdieva (Skolkovo Institute of Science and Technology), Evgeny Burnaev (Skolkovo Institute of Science and Technology)
Image TranslationOptimizationGenerative Adversarial NetworkImage
🎯 What it does: Proposes the Extremal Transport (ET) theory for unpaired image domain translation, introducing an algorithm that can be approximated by Incomplete Transport (IT);
f-Policy Gradients: A General Framework for Goal-Conditioned RL using f-Divergences
Siddhant Agarwal (University of Texas at Austin), Amy Zhang (University of Texas at Austin)
Reinforcement Learning
🎯 What it does: A target conditional reinforcement learning framework based on f-divergence minimization, called f-Policy Gradients (f-PG), is proposed, which uses the f-divergence between the state visitation distribution and the target distribution as the optimization objective.
FABind: Fast and Accurate Protein-Ligand Binding
Qizhi Pei (Renmin University of China), Rui Yan (Renmin University of China)
Drug DiscoveryProtein Structure PredictionGraph Neural NetworkBiomedical Data
🎯 What it does: An end-to-end protein-ligand binding prediction framework called FABind is proposed, which integrates pocket prediction and docking, capable of directly providing ligand binding sites and conformations.
Face Reconstruction from Facial Templates by Learning Latent Space of a Generator Network
Hatef Otroshi Shahreza (Idiap Research Institute), Sébastien Marcel (Idiap Research Institute)
RecognitionGenerationAdversarial AttackGenerative Adversarial NetworkImage
🎯 What it does: Utilizing the pre-trained StyleGAN3 network, a mapping network is learned to map from facial templates to an intermediate latent space, thereby generating high-resolution realistic facial images;
FACE: Evaluating Natural Language Generation with Fourier Analysis of Cross-Entropy
Zuhao Yang (Nanyang Technological University), Kefan Chen (Nanyang Technological University)
GenerationTransformerLarge Language ModelText
🎯 What it does: This paper proposes the FACE (Fourier Analysis of Cross-Entropy) metric, which quantifies the similarity in the frequency domain between the cross-entropy sequences of model-generated text and human text through a fast Fourier transform, thereby assessing the quality of natural language generation.
FaceComposer: A Unified Model for Versatile Facial Content Creation
Jiayu Wang (Alibaba Group), Jingren Zhou (Ant Group)
GenerationData SynthesisDiffusion modelImageVideoTextMultimodality
🎯 What it does: A unified FaceComposer model is proposed to achieve text-driven facial generation, editing, and animation for multi-task facial content creation.
FaceDNeRF: Semantics-Driven Face Reconstruction, Prompt Editing and Relighting with Diffusion Models
Hao ZHANG, Chi-Keung Tang (Hong Kong University of Science and Technology)
GenerationData SynthesisDiffusion modelScore-based ModelGenerative Adversarial NetworkImage
🎯 What it does: Reconstruct high-quality Face NeRF from a single face image using the frozen 3D GAN generator EG3D, and achieve semantically driven 3D editing and lighting reconstruction through text prompts.
Facilitating Graph Neural Networks with Random Walk on Simplicial Complexes
Cai Zhou (Tsinghua University), Muhan Zhang (Peking University)
Graph Neural NetworkGraph
🎯 What it does: This paper proposes EdgeRWSE, a structure encoding based on edge-level random walks, and the first edge-level positional encoding Hodge1Lap based on the Hodge 1-Laplacian spectrum by extending random walks to k-dimensional simplices (including edges and higher-order simplices). It further introduces a cross-random walk framework that unifies various simplex-based GNN methods.
Facing Off World Model Backbones: RNNs, Transformers, and S4
Fei Deng (Rutgers University), Sungjin Ahn (KAIST)
Recurrent Neural NetworkTransformerWorld ModelImageSequential
🎯 What it does: A visual world model S4WM based on the Parallel Solvable State Space Model (PSSM) is proposed, and its performance is compared with RNN, Transformer (TSSM-XL), and S4 as the backbone of the world model;
Factorized Contrastive Learning: Going Beyond Multi-view Redundancy
Paul Pu Liang (Carnegie Mellon University), Russ Salakhutdinov
Representation LearningData-Centric LearningContrastive LearningMultimodalityBiomedical DataBenchmark
🎯 What it does: This paper studies the shortcomings of multimodal contrastive learning in scenarios with uneven distribution of task-related information (low shared, high unique) and proposes the FACTORCL method, which can separate and learn shared and unique task-related representations.
Failure-Aware Gaussian Process Optimization with Regret Bounds
Shogo Iwazaki (MI-6 Limited), Mitsuru Irie (MI-6 Limited)
OptimizationTabularSequentialBenchmark
🎯 What it does: A Gaussian process optimization algorithm considering unknown failure regions, F-GP-UCB, is proposed.
Fair Adaptive Experiments
Waverly Wei (University of California), Jingshen Wang (University of California)
🎯 What it does: A fair adaptive experimental design is proposed, taking into account statistical efficiency, group fairness, and subject welfare;
Fair Allocation of Indivisible Chores: Beyond Additive Costs
Bo Li (Hong Kong Polytechnic University), Yu Zhou (Hong Kong Polytechnic University)
Optimization
🎯 What it does: This paper studies the fair allocation problem based on the maximum-minimum share (MMS) standard in the context of non-additive cost functions for indivisible task allocation, and provides several existence and approximation results.
Fair Canonical Correlation Analysis
Zhuoping Zhou (University of Pennsylvania), Li Shen (University of Pennsylvania)
OptimizationTabularBiomedical DataAlzheimer's Disease
🎯 What it does: This paper studies the fairness issue in Canonical Correlation Analysis (CCA) and proposes a Fair CCA (F-CCA) framework that eliminates bias by minimizing the correlation difference error between different protected attribute groups, providing both multi-objective and single-objective optimization algorithms.
Fair Graph Distillation
Qizhang Feng (Texas A&M University), Xia Hu
OptimizationKnowledge DistillationGraph Neural NetworkGraph
🎯 What it does: In the context of difficulties in large-scale graph processing with graph neural networks, a Fair Graph Distillation (FGD) method is proposed to generate compressed graphs that maintain predictive performance while reducing group fairness bias.
Fair Streaming Principal Component Analysis: Statistical and Algorithmic Viewpoint
Junghyun Lee (Korea Advanced Institute of Science and Technology), Chulhee Yun (Korea Advanced Institute of Science and Technology)
ImageTabular
🎯 What it does: A Fair Streaming PCA that can be executed in memory-constrained streaming environments is proposed, along with a corresponding theoretical learning framework (PAFO-learnability) and its implementation algorithm FNPM.
Fair, Polylog-Approximate Low-Cost Hierarchical Clustering
Marina Knittel (University of Maryland), MohammadTaghi Hajiaghayi
OptimizationTabular
🎯 What it does: A fair low-cost hierarchical clustering algorithm based on tree operations is proposed, which can obtain a hierarchical structure that simultaneously satisfies fairness, relative balance, and low cost based on a given approximate hierarchical tree without fairness constraints.
FairLISA: Fair User Modeling with Limited Sensitive Attributes Information
Zheng Zhang (University of Science and Technology of China), Enhong Chen (University of Science and Technology of China)
Recommendation SystemGenerative Adversarial NetworkTabular
🎯 What it does: The FairLISA framework is proposed to achieve fair user modeling when only a subset of users possess sensitive attribute information.
Fairly Recommending with Social Attributes: A Flexible and Controllable Optimization Approach
Jinqiu Jin (University of Science and Technology of China), Xiangnan He (University of Science and Technology of China)
Recommendation SystemOptimizationVideoTabular
🎯 What it does: Proposed a Neighborhood SP/NEO indicator for item-side group fairness (IGF) based on social attributes, and designed a multi-objective optimization algorithm SoFA that balances direct exposure and social benefits while maintaining recommendation accuracy.
Fairness Aware Counterfactuals for Subgroups
Loukas Kavouras (Athena Research Center), Ioannis Emiris (National and Kapodistrian University of Athens)
OptimizationExplainability and InterpretabilityComputational EfficiencyTabular
🎯 What it does: The FACTS framework is proposed, which uses counterfactual explanations to assess subgroup fairness and measure the difficulty of obtaining remedies.
Fairness Continual Learning Approach to Semantic Scene Understanding in Open-World Environments
Thanh-Dat Truong (University of Arkansas), Khoa Luu (University of Arkansas)
SegmentationKnowledge DistillationConvolutional Neural NetworkTransformerContrastive LearningImage
🎯 What it does: This paper proposes a fair continuous learning framework called FairCL for semantic segmentation, addressing issues of class imbalance, catastrophic forgetting, and background shift.
Fairness-guided Few-shot Prompting for Large Language Models
Huan Ma (Tianjin University), Bingzhe Wu (Tencent)
ClassificationOptimizationTransformerLarge Language ModelPrompt EngineeringText
🎯 What it does: A few-shot prompt search method based on prediction bias (fairness) is proposed to enhance the performance of large language models with a small number of example prompts.
Faith and Fate: Limits of Transformers on Compositionality
Nouha Dziri (Allen Institute for Artificial Intelligence), Yejin Choi (University of Washington)
TransformerLarge Language ModelText
🎯 What it does: This paper systematically evaluates the performance of large Transformers on three typical combinatorial reasoning tasks (multi-digit multiplication, logic grid puzzles, dynamic programming subset sum) by constructing a computational graph of the algorithms, revealing that they often complete tasks through subgraph matching rather than true systematic reasoning during multi-step reasoning.
False Discovery Proportion control for aggregated Knockoffs
Alexandre Blain (INRIA), Pierre Neuvial (Institut de Mathématiques de Toulouse Université de Toulouse)
Biomedical DataMagnetic Resonance Imaging
🎯 What it does: This paper proposes a Knockoff-based aggregation method called KOPI, which can control the false discovery proportion (FDP) in high-dimensional conditional variable selection and improve testing power.
FAMO: Fast Adaptive Multitask Optimization
Bo Liu (University of Texas at Austin), qiang liu
OptimizationReinforcement LearningImage
🎯 What it does: An efficient multi-task learning optimizer FAMO is proposed, capable of achieving balanced loss reduction for tasks in O(1) space and time.
Fantastic Robustness Measures: The Secrets of Robust Generalization
Hoki Kim (Seoul National University), Jaewook Lee (Seoul National University)
OptimizationAdversarial AttackConvolutional Neural NetworkImage
🎯 What it does: On the CIFAR-10 and RobustBench datasets, the authors trained over 1,300 models and compared more than 100 existing robust generalization metrics, systematically evaluating the relationship between these metrics and the gap in robust generalization;
Fantastic Weights and How to Find Them: Where to Prune in Dynamic Sparse Training
Aleksandra Nowak, Jacek Tabor (Jagiellonian University)
ClassificationOptimizationConvolutional Neural NetworkImageTabular
🎯 What it does: Conduct large-scale experiments and comparisons on pruning criteria in dynamic sparse training (DST) to explore their impact on model performance, update frequency, and network structure.
Fast and Regret Optimal Best Arm Identification: Fundamental Limits and Low-Complexity Algorithms
Qining Zhang (University of Michigan), Lei Ying (University of Michigan)
OptimizationTabular
🎯 What it does: This paper proposes a class of algorithmic frameworks that simultaneously achieve fast identification of the optimal arm and minimize cumulative regret—ROBAI (Regret Optimal Best Arm Identification)—and presents three specific algorithms: EOCP with a preset stopping time, EOCP-UG with an adaptive stopping time, and KL-EOCP that achieves optimal regret under general exponential families.
Fast and Simple Spectral Clustering in Theory and Practice
Peter Macgregor (University of Edinburgh)
OptimizationComputational EfficiencyGraph
🎯 What it does: A fast spectral clustering algorithm is proposed that uses power iteration to generate O(log k) dimensional random projections, avoiding the expensive computation of k eigenvectors in traditional algorithms.
Fast Approximation of Similarity Graphs with Kernel Density Estimation
Peter Macgregor (University of Edinburgh), He Sun (University of Edinburgh)
SegmentationComputational EfficiencyGraph Neural NetworkGaussian SplattingImageGraph
🎯 What it does: A randomized algorithm utilizing kernel density estimation (KDE) as a black box has been designed and implemented, capable of quickly constructing a sparse similarity graph from an input point set while compressing the number of edges to nearly linear, all while preserving the clustering structure.
Fast Asymptotically Optimal Algorithms for Non-Parametric Stochastic Bandits
Dorian Baudry (Ecole Polytechnique), Odalric-Ambrym Maillard (University of Lille)
Recommendation SystemOptimizationComputational EfficiencyTabularAgriculture Related
🎯 What it does: This paper proposes two sets of rapidly implementable non-parametric stochastic Bandit algorithms, namely FMED/FIMED (which fully computes KL-inf only on the pulled arms) and OMED/OIMED (which approximates KL-inf using online portfolio selection algorithms), significantly reducing the computational and memory overhead of the original MED/IMED algorithms.
Fast Attention Over Long Sequences With Dynamic Sparse Flash Attention
Matteo Pagliardini (École Polytechnique Fédérale de Lausanne), François Fleuret (University of Geneva)
TransformerLarge Language ModelText
🎯 What it does: Developed a FlashAttention extension that efficiently implements dynamic sparse attention on GPUs, supporting dynamic modes such as key/query dropping and hash sparsity.