ICLR 2025 Papers — Page 14
International Conference on Learning Representations · 3704 papers
Gaussian Differentially Private Human Faces Under a Face Radial Curve Representation
Carlos J Soto, Mark Shriver (Pennsylvania State University)
Safty and PrivacyRepresentation LearningMesh
🎯 What it does: A new facial surface representation method called facial radial curves is proposed, and based on this, a mechanism that satisfies Gaussian Differential Privacy (GDP) is designed, which can publish 3D facial data while preserving facial shape.
Gaussian Ensemble Belief Propagation for Efficient Inference in High-Dimensional, Black-box Systems
Dan MacKinlay (CSIRO Data61), Petra Kuhnert (CSIRO Data61)
OptimizationComputational EfficiencyTime Series
🎯 What it does: An algorithm named Gaussian Ensemble Belief Propagation (GEnBP) is proposed for efficient inference in high-dimensional graphical models, capable of simultaneously handling nonlinear observations, unknown parameters, and deep dependencies.
Gaussian Head & Shoulders: High Fidelity Neural Upper Body Avatars with Anchor Gaussian Guided Texture Warping
Tianhao Walter Wu (University of Cambridge), Cengiz Oztireli (University of Cambridge)
RestorationGenerationData SynthesisPose EstimationGaussian SplattingVideo
🎯 What it does: Using 3D Gaussian Splatting and sparse anchor Gaussians, combined with pose-related neural texture warping, to reconstruct high-fidelity, controllable head and upper body (chest and shoulders) full-body avatars from monocular video.
Gaussian Mixture Counterfactual Generator
Jong-Hoon Ahn (Otsuka Pharmaceutical Development and Commercialization), Akshay Vashist (Otsuka Pharmaceutical Development and Commercialization)
GenerationData SynthesisTabularTime Series
🎯 What it does: A Gaussian Mixture Model-based Generator for Controls (GMCG) is proposed for estimating individualized treatment effects and generating control data for continuous, multidimensional, time-varying treatments.
Gaussian Splatting Lucas-Kanade
Liuyue Xie (Carnegie Mellon University), Laszlo Attila Jeni (Fujitsu Research of America)
RestorationDepth EstimationGaussian SplattingOptical FlowVideoOrdinary Differential Equation
🎯 What it does: This paper proposes a regularization method for analytical scene flow based on Lucas-Kanade, aimed at constraining the warp field of dynamic Gaussian splats to achieve more accurate dynamic 3D scene reconstruction.
Gaussian-Based Instance-Adaptive Intensity Modeling for Point-Supervised Facial Expression Spotting
Yicheng Deng (Osaka University), Hajime Nagahara (Osaka University)
ClassificationRecognitionTransformerContrastive LearningOptical FlowVideo
🎯 What it does: The research focuses on supervised facial expression detection, constructing a dual-branch framework and introducing Gaussian Instance Adaptive Intensity Modeling to achieve soft pseudo-labeling, further enhancing feature discrimination through intensity-aware contrastive learning.
Gaussian-Det: Learning Closed-Surface Gaussians for 3D Object Detection
Hongru Yan (Tsinghua University), Yueqi Duan (Tsinghua University)
Object DetectionGaussian SplattingPoint Cloud
🎯 What it does: This paper proposes a multi-view 3D object detection framework called Gaussian-Det, which utilizes Gaussian Splatting for continuous surface representation, and designs a Closure Inferring Module to suppress noise points and enhance detection accuracy.
GaussianAnything: Interactive Point Cloud Flow Matching for 3D Generation
Yushi LAN, Chen Change Loy (Nanyang Technological University)
GenerationData SynthesisTransformerDiffusion modelFlow-based ModelAuto EncoderImageTextPoint Cloud
🎯 What it does: A 3D generation framework called GAUSSIANANYTHING is proposed, which is based on an interactive point cloud structure latent space and can support 3D generation and editing conditioned on text, images, and point clouds simultaneously.
GaussianBlock: Building Part-Aware Compositional and Editable 3D Scene by Primitives and Gaussians
Shuyi Jiang (Singapore University of Technology and Design), Na Zhao (Singapore University of Technology and Design)
GenerationData SynthesisGaussian SplattingPoint Cloud
🎯 What it does: Proposes the GaussianBlock method to achieve editable and semantically coherent block-based 3D reconstruction from 2D images;
GDrag:Towards General-Purpose Interactive Editing with Anti-ambiguity Point Diffusion
Xiaojian Lin (Sun Yat-Sen University), Xiaodan Liang
Diffusion modelImage
🎯 What it does: A general task-aware point drag image editing framework GDrag has been developed, achieving more precise interactive editing by defining atomic tasks and utilizing ambiguity-resistant dense trajectories and adaptive motion supervision.
GenARM: Reward Guided Generation with Autoregressive Reward Model for Test-Time Alignment
Yuancheng Xu (University of Maryland), Sumitra Ganesh (JPMorgan AI Research)
GenerationReinforcement Learning from Human FeedbackTransformerLarge Language ModelReinforcement LearningText
🎯 What it does: The GenARM method is proposed, which utilizes an autoregressive reward model (Autoregressive RM) to guide the frozen LLM for human preference alignment generation during testing.
GenDataAgent: On-the-fly Dataset Augmentation with Synthetic Data
Zhiteng Li (Shanghai Jiao Tong University), Alice Xiang (Sony AI)
ClassificationData SynthesisConvolutional Neural NetworkSupervised Fine-TuningDiffusion modelImage
🎯 What it does: A generative agent called GenDataAgent is proposed to generate synthetic data in real-time during the training process, aimed at enhancing image classification training data and improving model generalization and fairness.
General Scene Adaptation for Vision-and-Language Navigation
Haodong Hong (University of Queensland), Qi Wu (University of Adelaide)
Domain AdaptationRobotic IntelligenceTransformerLarge Language ModelVision Language ModelTextMultimodality
🎯 What it does: This paper proposes the GSA-VLN task and its evaluation dataset GSA-R2R, exploring how visual-language navigation agents achieve adaptation in persistent scenes through continuous memory and parameter updates.
Generalizability of Neural Networks Minimizing Empirical Risk Based on Expressive Power
Lijia Yu (Key Laboratory of System Software of Chinese Academy of Sciences), Lijun Zhang (Institute of AI for Industries, Chinese Academy of Sciences)
🎯 What it does: This study investigates the generalization lower bound of two-layer neural networks that minimize or approximately minimize empirical risk when their expressive power is sufficient. It derives the joint effect of the number of training samples and network width on accuracy and provides a lower bound on sample complexity.
Generalizable Human Gaussians from Single-View Image
Jinnan Chen (National University of Singapore), Gim Hee Lee (National University of Singapore)
GenerationData SynthesisPose EstimationDiffusion modelGaussian SplattingImage
🎯 What it does: A method is proposed to generate a high-quality 3D human Gaussian model (HGM) that can be rendered from any viewpoint using a single portrait photo.
Generalizable Motion Planning via Operator Learning
Sharath Matada (University of California San Diego), Nikolay Atanasov (University of California San Diego)
OptimizationRobotic IntelligenceGraph
🎯 What it does: This paper proposes the Planning Neural Operator (PNO), which achieves generalized motion planning for different environments and target locations by learning the solution operator of the Eikonal PDE.
Generalization and Distributed Learning of GFlowNets
Tiago Silva, Diego Mesquita (Getulio Vargas Foundation)
Drug DiscoveryGraph Neural NetworkGraph
🎯 What it does: This paper presents, for the first time, a data-dependent GFlowNet generalization upper bound through PAC-Bayes theory, revealing the negative impact of state space size and trajectory length on generalization performance. It also proposes a distributed learning algorithm based on graph partitioning, Subgraph Asynchronous Learning (SAL), which significantly improves pattern coverage and distribution matching.
Generalization Bounds and Model Complexity for Kolmogorov–Arnold Networks
Xianyang Zhang (Texas A&M University), Huijuan Zhou (Shanghai University of Finance and Economics)
ImageTabular
🎯 What it does: This paper presents the upper bounds of generalization error for Kolmogorov–Arnold Networks (KAN) in the cases of basis function expansion and low-rank RKHS activation functions, and validates their relationship with model complexity through numerical experiments.
Generalization Bounds for Canonicalization: A Comparative Study with Group Averaging
Behrooz Tahmasebi (Massachusetts Institute of Technology), Stefanie Jegelka (Technical University of Munich)
Point Cloud
🎯 What it does: This study investigates the construction of invariant function classes' generalization bounds using the canonicalization method in learning tasks with symmetry, comparing it with the traditional group averaging method.
Generalization Guarantees for Representation Learning via Data-Dependent Gaussian Mixture Priors
Milad Sefidgaran (Huawei Technologies France), Piotr Krasnowski (Huawei Technologies France)
ClassificationRepresentation LearningConvolutional Neural NetworkRecurrent Neural NetworkImage
🎯 What it does: A generalization error upper bound for representation learning based on Minimum Description Length (MDL) is proposed, and a regularization method based on data-adaptive Gaussian mixture priors is designed using this upper bound.
Generalization in VAE and Diffusion Models: A Unified Information-Theoretic Analysis
Qi CHEN, Florian Shkurti (University of Toronto)
GenerationData SynthesisDiffusion modelAuto EncoderImage
🎯 What it does: A unified information-theoretic framework is proposed to analyze the generalization performance of VAE and diffusion models in terms of encoders and generators, providing computable error bounds;
Generalization through variance: how noise shapes inductive biases in diffusion models
John Vastola
GenerationData SynthesisDiffusion modelPoint CloudStochastic Differential Equation
🎯 What it does: A theoretical framework was constructed, using path integral methods to derive the typical learning distribution of diffusion models during training due to noise targets, and the model's generalization behavior was analyzed under different architectures and forward process settings through V-kernel analysis.
Generalization v.s. Memorization: Tracing Language Models’ Capabilities Back to Pretraining Data
Xinyi Wang (University of California), William Yang Wang (University of California)
TransformerLarge Language ModelText
🎯 What it does: This paper quantifies the distributed memory and generalization degree of large language models in different tasks by constructing task grammar tables and task grammar language models, and tracks the source of their capabilities from pre-training corpora.
Generalization, Expressivity, and Universality of Graph Neural Networks on Attributed Graphs
Levi Rauchwerger (Technion - Israel Institute of Technology), Ron Levie (Technion - Israel Institute of Technology)
Graph Neural NetworkGraph
🎯 What it does: A pseudo-metric for attributed graphs is proposed, making the graph space compact under this metric, and proving that Graph Neural Networks (MPNN) are continuous and can separate different graphs under this metric, thereby demonstrating the universal approximation and generalization bounds of MPNN.
Generalized Behavior Learning from Diverse Demonstrations
Varshith Sreeramdass (Georgia Institute of Technology), Matthew Gombolay (Georgia Institute of Technology)
Robotic IntelligenceReinforcement LearningSequential
🎯 What it does: Research on how to learn generalizable diverse strategies from diverse demonstrations.
Generalized Consistency Trajectory Models for Image Manipulation
Beomsu Kim (Korea Advanced Institute of Science and Technology), Jong Chul Ye (Korea Advanced Institute of Science and Technology)
Image TranslationRestorationGenerationFlow-based ModelImageOrdinary Differential Equation
🎯 What it does: This paper proposes Generalized Consistency Trajectory Models (GCTM), which enables one-time conversion between any two distributions, extending the original Consistency Trajectory Models (CTM) framework that could only convert from Gaussian noise to data.
Generalized Principal-Agent Problem with a Learning Agent
Tao Lin (Harvard University), Yiling Chen (Harvard University)
🎯 What it does: This paper studies the scenario of learning agents in a general principal-agent problem where the agent has no private information and the principal has no commitment power, and provides upper and lower bounds on the expected utility that the principal can obtain.
Generalized Video Moment Retrieval
You Qin (Nanjing University), Roger Zimmermann (National University of Singapore)
RetrievalTransformerContrastive LearningVideoText
🎯 What it does: A General Video Moment Retrieval (GVMR) framework is proposed, capable of handling multi-target and no-target queries.
Generalizing Reasoning Problems to Longer Lengths
Changnan Xiao, Bing Liu (University of Illinois Chicago)
TransformerSequentialChain-of-Thought
🎯 What it does: This paper studies the length generalization (LG) problem, proposes the (n,r)-consistency condition, and proves that Transformers can achieve LG on reasoning problems that satisfy this condition. Subsequently, experiments are conducted on artificially synthesized problems such as arithmetic, parity, addition, multiplication, and division for validation.
Generalizing Weisfeiler-Lehman Kernels to Subgraphs
Dongkwan Kim (Korea Advanced Institute of Science and Technology), Alice Oh (Korea Advanced Institute of Science and Technology)
ClassificationRepresentation LearningGraph Neural NetworkGraph
🎯 What it does: A subgraph kernel WLKS based on Weisfeiler-Lehman is proposed to address subgraph-level tasks.
Generating Graphs via Spectral Diffusion
Giorgia Minello (Ca Foscari University), Luca Cosmo (Ca Foscari University)
GenerationData SynthesisDrug DiscoveryGraph Neural NetworkTransformerDiffusion modelGraphBiomedical Data
🎯 What it does: A diffusion-based graph generation model GGSD is proposed, which directly samples spectral pairs in the node space using spectral decomposition and a denoising diffusion process to reconstruct the graph adjacency matrix and generate conditionally;
Generating CAD Code with Vision-Language Models for 3D Designs
Kamel Alrashedy (Georgia Institute of Technology), Matthew Gombolay (Georgia Institute of Technology)
GenerationOptimizationAI Code AssistantTransformerLarge Language ModelVision Language ModelTextPoint CloudBenchmarkChain-of-Thought
🎯 What it does: The CADCodeVerify method is proposed, which utilizes Vision-Language models to automatically generate and iteratively refine CAD scripts, enabling the generation and improvement of 3D designs without human intervention.
Generating Freeform Endoskeletal Robots
Muhan Li (Northwestern University), Sam Kriegman (Northwestern University)
Robotic IntelligenceReinforcement LearningAuto Encoder
🎯 What it does: Automatically design free-form soft robots with internal skeletons, co-evolving morphology and control.
Generating Likely Counterfactuals Using Sum-Product Networks
Jiří Němeček (Czech Technical University), Jakub Marecek
OptimizationExplainability and InterpretabilityTabularFinance Related
🎯 What it does: By embedding the Sum-Product Network (SPN) into a Mixed Integer Optimization (MIO) framework, a method for generating counterfactual explanations (LiCE) that can simultaneously optimize interpretability, similarity, sparsity, executability, and feasibility (high likelihood) is proposed.
Generating Physical Dynamics under Priors
Zihan Zhou (Chinese University of Hong Kong), Tianshu Yu (Chinese University of Hong Kong)
GenerationData SynthesisOptimizationDiffusion modelTime SeriesSequentialPhysics Related
🎯 What it does: In the task of generating physics-driven dynamics using data, this paper proposes a diffusion generative model framework that can directly generate dynamic sequences that satisfy physical constraints.
Generation and Comprehension Hand-in-Hand: Vision-guided Expression Diffusion for Boosting Referring Expression Generation and Comprehension
Jingcheng Ke (National Tsing Hua University), Yen-Yu Lin (National Yang Ming Chiao Tung University)
RecognitionGenerationTransformerDiffusion modelImageText
🎯 What it does: A visually guided diffusion model VIE-DM is proposed for generating diverse and accurate referring expressions and expanding the REC dataset.
Generative Adapter: Contextualizing Language Models in Parameters with A Single Forward Pass
Tong Chen (University of Washington), Hao Cheng (Microsoft)
GenerationOptimizationTransformerLarge Language ModelSupervised Fine-TuningText
🎯 What it does: An adaptive method based on a generative adapter is proposed, which can encode any context into parameter updates with just one forward pass during inference, thereby quickly customizing large language models.
Generative Classifiers Avoid Shortcut Solutions
Alexander Cong Li, Deepak Pathak (Carnegie Mellon University)
ClassificationDomain AdaptationTransformerDiffusion modelImage
🎯 What it does: Using class-conditional generative models to achieve classification through Bayes' theorem, avoiding the shortcuts that discriminative models easily learn.
Generative Flows on Synthetic Pathway for Drug Design
Seonghwan Seo (Korea Advanced Institute of Science and Technology), Woo Youn Kim (Korea Advanced Institute of Science and Technology)
Drug DiscoveryFlow-based ModelTabular
🎯 What it does: This work proposes RXNFLOW, a synthetic pathway generation framework based on generative flow networks, which utilizes a massive chemical building block module and reaction templates to construct synthesizable drug molecules.
Generative Inbetweening: Adapting Image-to-Video Models for Keyframe Interpolation
Xiaojuan Wang (University of Washington), Steve Seitz
GenerationData SynthesisDiffusion modelVideo
🎯 What it does: A lightweight fine-tuning method is proposed to transform the pre-trained image-to-video diffusion model (Stable Video Diffusion) into a model capable of video interpolation between two given key frames; by rotating the temporal self-attention map and only fine-tuning the value and output projection matrices, a backward motion model is generated, and during inference, bidirectional sampling is used to fuse forward and backward motion, achieving coherent video generation.
Generative Monoculture in Large Language Models
Fan Wu (University of Illinois Urbana-Champaign), Varun Chandrasekaran (University of Illinois Urbana-Champaign)
GenerationReinforcement Learning from Human FeedbackTransformerLarge Language ModelReinforcement LearningPrompt EngineeringText
🎯 What it does: This paper defines and systematically evaluates the phenomenon of 'generative monoculture' in large language models, experimentally testing the lack of diversity in book reviews and code generation tasks, and attempts various generation parameters and prompts to mitigate this issue.
Generative Representational Instruction Tuning
Niklas Muennighoff (Contextual AI), Douwe Kiela (Contextual AI)
GenerationRetrievalTransformerLarge Language ModelSupervised Fine-TuningMixture of ExpertsTextRetrieval-Augmented Generation
🎯 What it does: This paper proposes a unified model for text embedding and generation—GRIT (Generative Representational Instruction Tuning), which addresses the previous issue of focusing solely on a single task by simultaneously learning embedding tasks and generation tasks on the same large language model.
Generative Verifiers: Reward Modeling as Next-Token Prediction
Lunjun Zhang (University of Toronto), Rishabh Agarwal (Google DeepMind)
GenerationData SynthesisTransformerLarge Language ModelSupervised Fine-TuningTextChain-of-Thought
🎯 What it does: Proposes to reframe the verification process as next-word prediction, training a generative verifier (GenRM) to enhance the performance of large language models in reasoning tasks.
Generator Matching: Generative modeling with arbitrary Markov processes
Peter Holderrieth (Massachusetts Institute of Technology), Yaron Lipman (Meta)
GenerationData SynthesisProtein Structure PredictionDiffusion modelFlow-based ModelImageMultimodalityBiomedical DataStochastic Differential EquationOrdinary Differential Equation
🎯 What it does: A Generator Matching framework is proposed, utilizing the generator of any Markov process to achieve scalable generative models, allowing for the combination of different models and multimodal data.
GenEx: Generating an Explorable World
TaiMing Lu, Jieneng Chen (Johns Hopkins University)
GenerationRobotic IntelligenceLarge Language ModelDiffusion modelVideo
🎯 What it does: A generative world exploration framework called GenEx is constructed, based on a panoramic video diffusion model, enabling agents to generate continuous 360° perspectives through imagination and cognitive updates;
GenSE: Generative Speech Enhancement via Language Models using Hierarchical Modeling
Jixun Yao (Northwestern Polytechnical University), Lei Xie (Nanyang Technological University)
RestorationGenerationTransformerPrompt EngineeringGenerative Adversarial NetworkAudio
🎯 What it does: A language model-based speech enhancement framework called GenSE is proposed, which includes a single quantizer speech encoder SimCodec and a hierarchical denoising-generating two-stage process, maintaining speaker timbre consistency through token chain prompts.
GenVP: Generating Visual Puzzles with Contrastive Hierarchical VAEs
Kalliopi Basioti (Rutgers University), Vladimir Pavlovic (Rutgers University)
GenerationData SynthesisReinforcement LearningMixture of ExpertsImage
🎯 What it does: The GenVP model is proposed, which can solve RAVEN-style RPM puzzles and generate complete new puzzles based on abstract rules, supporting multiple answer generation.
GenXD: Generating Any 3D and 4D Scenes
Yuyang Zhao (National University of Singapore), Lijuan Wang (National University of Singapore)
GenerationData SynthesisTransformerDiffusion modelGaussian SplattingImageVideo
🎯 What it does: This paper presents Gen X D, a unified image-conditioned 3D and 4D generation model;
GeoILP: A Synthetic Dataset to Guide Large-Scale Rule Induction
Si Chen (Beihang University), Xu Zhang (National Computer Network Emergency Response Technical Team Coordination Center of China)
Image
🎯 What it does: This paper presents GeoILP, a large-scale synthetic geometric reasoning dataset designed to evaluate and advance research in large-scale ILP methods.
GeoLoRA: Geometric integration for parameter efficient fine-tuning
Steffen Schotthöfer (Oak Ridge National Laboratory), Jonas Kusch (Norwegian University of Life Sciences)
OptimizationTransformerSupervised Fine-TuningDiffusion modelImageText
🎯 What it does: GeoLoRA is developed, a parameter-efficient fine-tuning method based on geometric dynamics low-rank approximation;
Geometric Inductive Biases of Deep Networks: The Role of Data and Architecture
Sajad Movahedi (ELLIS Institute), Seyed-Mohsen Moosavi-Dezfooli (Apple)
ClassificationOptimizationConvolutional Neural NetworkTransformerImage
🎯 What it does: This study investigates the geometric evolution of neural networks in the input space, proposes the Geometric Invariance Hypothesis (GIH), and explores the impact of architecture and data on geometry and generalization.
Geometry Image Diffusion: Fast and Data-Efficient Text-to-3D with Image-Based Surface Representation
Slava Elizarov (Unity Technologies), Simon Donné (Unity Technologies)
GenerationData SynthesisDiffusion modelImageText
🎯 What it does: This paper proposes Geometry Image Diffusion (GIMDiffusion), a method that combines geometric images with collaborative control to generate 3D models from text descriptions, capable of producing re-lightable and editable 3D assets while also outputting UV maps.
Geometry of Lightning Self-Attention: Identifiability and Dimension
Nathan W. Henry (University of Toronto), Kathlén Kohn (Royal Institute of Technology)
Transformer
🎯 What it does: Analyzed the function space geometry of unnormalized (lightning) self-attention networks, described the fiber structure of parameter mappings, and derived the dimensionality formula for deep networks.
Geometry of Long-Tailed Representation Learning: Rebalancing Features for Skewed Distributions
Lingjie Yi (Stony Brook University), Chao Chen (University Paris 1 Pantheon-Sorbonne)
Representation LearningContrastive LearningImage
🎯 What it does: This paper first analyzes the impact of long-tail distribution on the geometry of the feature space from a theoretical perspective, proving that the centers of tail classes will shrink or collapse. It then proposes the FeatRecon method, which rebalances the feature distribution by synthesizing and filling the confidence support region in the feature space, resulting in a symmetrical and linearly separable geometric structure for various feature classes.
Geometry of Neural Reinforcement Learning in Continuous State and Action Spaces
Saket Tiwari (Brown University), George Konidaris (Brown University)
Reinforcement LearningSequential
🎯 What it does: This study investigates the geometric structure of reinforcement learning with continuous state and action spaces, proving that under the limit of wide two-layer neural networks and small learning rates, the reachable set of states approximates a low-dimensional manifold, with an upper dimension bound of 2d_a + 1, and experiments are conducted in various MuJoCo environments.
Geometry-Aware Approaches for Balancing Performance and Theoretical Guarantees in Linear Bandits
Yuwei Luo (Stanford University), Mohsen Bayati (Stanford University)
OptimizationReinforcement LearningTabular
🎯 What it does: A real-time data-driven framework based on the geometric features of confidence ellipses is proposed, which unifies the analysis of POFUL class algorithms including OFUL, LinTS, TS-Freq, and Greedy. Based on this, course correction algorithms TS-MR and Greedy-MR are designed, which retain the computational efficiency and empirical performance of the original algorithms while achieving optimal frequent theoretical regret across all instances.
Geometry-aware RL for Manipulation of Varying Shapes and Deformable Objects
Tai Hoang (Karlsruhe Institute of Technology), Gerhard Neumann (Karlsruhe Institute of Technology)
Robotic IntelligenceGraph Neural NetworkTransformerReinforcement LearningGraph
🎯 What it does: This paper proposes a geometric perception-based heterogeneous graph reinforcement learning framework and designs the HEPi model on this basis to manipulate objects with varying shapes and flexible materials.
GeoX: Geometric Problem Solving Through Unified Formalized Vision-Language Pre-training
Renqiu Xia (Shanghai Jiao Tong University), Bo Zhang (Shanghai Artificial Intelligence Laboratory)
TransformerLarge Language ModelSupervised Fine-TuningVision Language ModelTextMultimodality
🎯 What it does: A multimodal large model named GeoX has been developed specifically for automatic geometric problem solving, achieving a unified input-output format through formalized visual-language pre-training.
GeSubNet: Gene Interaction Inference for Disease Subtype Network Generation
Ziwei Yang (Kyoto University), Jimeng Sun (University of Illinois Urbana-Champaign)
GenerationData SynthesisDrug DiscoveryGraph Neural NetworkAuto EncoderGraphBiomedical Data
🎯 What it does: Proposed the GeSubNet framework, which combines patient gene expression with prior gene network learning to infer disease subtype-specific gene networks.
GETS: Ensemble Temperature Scaling for Calibration in Graph Neural Networks
Dingyi Zhuang (Massachusetts Institute of Technology), Jinhua Zhao (Massachusetts Institute of Technology)
Graph Neural NetworkMixture of ExpertsGraph
🎯 What it does: A graph mixture of experts (MoE) based ensemble temperature scaling framework GETS is proposed for node-level uncertainty calibration of graph neural networks (GNNs).
GEVRM: Goal-Expressive Video Generation Model For Robust Visual Manipulation
Hongyin Zhang (Zhejiang University), Donglin Wang (Westlake University)
GenerationRobotic IntelligenceTransformerDiffusion modelRectified FlowContrastive LearningVideoText
🎯 What it does: A closed-loop visual language action (VLA) framework GEVRM based on the principle of Internal Model Control (IMC) has been designed and implemented, using a text-guided video generator to produce highly expressive target states, and aligning states through prototype contrastive learning to enhance the robustness of the robot under external disturbances.
GI-GS: Global Illumination Decomposition on Gaussian Splatting for Inverse Rendering
HONGZE CHEN, Jun Zhang (Hong Kong University of Science and Technology)
Gaussian SplattingPoint Cloud
🎯 What it does: The inverse rendering framework GI-GS is implemented based on 3D Gaussian Splatting, combining deferred shading and differentiable path tracing to achieve direct and indirect decomposition of lighting, resulting in high-quality view synthesis and relighting.
GIFT: Unlocking Full Potential of Labels in Distilled Dataset at Near-zero Cost
Xinyi Shang (University College London), Tao Lin (Westlake University)
Data SynthesisKnowledge DistillationConvolutional Neural NetworkImage
🎯 What it does: A plugin method called GIFT is proposed in dataset distillation, which fully utilizes the complete information of soft labels and hard labels, significantly improving the training effectiveness of synthetic data.
Glad: A Streaming Scene Generator for Autonomous Driving
Bin Xie (Tianjin University), Xiangyu Zhang (MEGVII Technology)
GenerationAutonomous DrivingDiffusion modelVideo
🎯 What it does: Proposes the GLAD framework, which uses Stable Diffusion for online frame-level video generation, supporting arbitrary lengths and reference frame-driven simulations.
Glauber Generative Model: Discrete Diffusion Models via Binary Classification
Harshit Varma (Inception Labs), Karthikeyan Shanmugam (Google DeepMind)
GenerationData SynthesisTransformerDiffusion modelImageText
🎯 What it does: Proposes the Glauber Generative Model (GGM), a discrete diffusion model based on discrete Markov chains, which generates samples from the target distribution by progressively denoising.
Glimpse: Enabling White-Box Methods to Use Proprietary Models for Zero-Shot LLM-Generated Text Detection
Guangsheng Bao (Zhejiang University), Yue Zhang (Westlake University)
GenerationOptimizationTransformerLarge Language ModelText
🎯 What it does: This paper proposes the Glimpse method, which estimates the complete word distribution using only the Top-K word probabilities, enabling white-box detection techniques to operate on private LLMs.
Global Convergence in Neural ODEs: Impact of Activation Functions
Tianxiang Gao (DePaul University), Hongyang Gao (Iowa State University)
ImageTextOrdinary Differential Equation
🎯 What it does: This paper studies the impact of the smoothness and nonlinearity of activation functions on the convergence and stability of neural ODE training. It proves that smooth and non-polynomial activations can ensure the global uniqueness of forward/backward ODE solutions, and provides a global convergence proof for neural ODEs in over-parameterized cases through NTK theory.
Global Convergence of Policy Gradient in Average Reward MDPs
Navdeep Kumar (Technion - Israel Institute of Technology), Shie Mannor (Technion - Israel Institute of Technology)
OptimizationReinforcement LearningTabular
🎯 What it does: This paper provides a complete global convergence analysis for the projected policy gradient method in average reward infinite-horizon MDPs, proving that the convergence rate is sublinear O(1/T) and providing a proof of the smoothness of the average reward.
Global Identifiability of Overcomplete Dictionary Learning via L1 and Volume Minimization
Yuchen Sun (University of Florida), Kejun Huang (University of Florida)
OptimizationTabular
🎯 What it does: A novel objective function for overcomplete dictionary learning is proposed, which integrates the weighted ℓ1 norm of the sparse coefficient row and the logarithm of the volume of the dictionary matrix, achieving global identifiability;
Global Well-posedness and Convergence Analysis of Score-based Generative Models via Sharp Lipschitz Estimates
Connor Mooney (University of California at Irvine), Yifeng Yu (University of California at Irvine)
GenerationData SynthesisOptimizationDiffusion modelScore-based Model
🎯 What it does: This paper theoretically proves the global existence and convergence of the fractional generative model (SGM) under general initial data through sharp Lipschitz estimates of the fractional function, especially providing a local Lipschitz upper bound for the entire time period and achieving the KL convergence rate for discrete sampling.
GLOMA: Global Video Text Spotting with Morphological Association
Han Wang (Bytedance), Can Huang (Bytedance)
RecognitionObject DetectionObject TrackingTransformerVideo
🎯 What it does: A global association framework GLOMA based on Transformer is proposed for video text detection, recognition, and tracking.
GLoRa: A Benchmark to Evaluate the Ability to Learn Long-Range Dependencies in Graphs
Dongzhuoran Zhou (Bosch Center for AI), Egor V. Kostylev (University of Oslo)
Graph Neural NetworkTransformerGraphBenchmark
🎯 What it does: This paper proposes a synthetic benchmark (GLoRa) for evaluating the long-range dependency learning capability of graph learning models and systematically assesses various mainstream GNN and Transformer systems on this benchmark.
GlycanML: A Multi-Task and Multi-Structure Benchmark for Glycan Machine Learning
Minghao Xu (Peking University), Wentao Zhang (Peking University)
ClassificationDrug DiscoveryConvolutional Neural NetworkRecurrent Neural NetworkGraph Neural NetworkTransformerGraphBenchmark
🎯 What it does: A benchmark called GLYCANML is proposed for evaluating machine learning models on various tasks related to glycan molecules, including classification, immunogenicity, glycosylation types, and protein-glycan interactions, covering both sequence and graph representations.
GMValuator: Similarity-based Data Valuation for Generative Models
Jiaxi Yang (University of British Columbia), Xiaoxiao Li (University of British Columbia)
GenerationExplainability and InterpretabilityComputational EfficiencyData-Centric LearningImageText
🎯 What it does: GMVALUATOR is proposed, a data value assessment method that is training-free and universally applicable to any generative model, transforming the data value problem into a similarity matching between generated samples and training samples.
GNNs Getting ComFy: Community and Feature Similarity Guided Rewiring
Celia Rubio-Madrigal (CISPA Helmholtz Center for Information Security), Rebekka Burkholz (CISPA Helmholtz Center for Information Security)
ClassificationOptimizationGraph Neural NetworkGraph
🎯 What it does: Three graph re-linking strategies based on community structure and feature similarity are proposed (ComMa, FeaSt, ComFy), which enhance the node classification performance of graph neural networks (GNNs) by altering the connectivity patterns of edges.
Gnothi Seauton: Empowering Faithful Self-Interpretability in Black-Box Transformers
Shaobo Wang (Shanghai Jiao Tong University), Linfeng Zhang (Shanghai Jiao Tong University)
Explainability and InterpretabilityComputational EfficiencyTransformerImageText
🎯 What it does: The AutoGnothi method is proposed, which achieves self-explainability using a side network while keeping the original parameters of the black-box Transformer unchanged, and generates theoretically credible Shapley value explanations.
GOAL: A Generalist Combinatorial Optimization Agent Learner
Darko Drakulic, Jean-Marc Andreoli (NAVER LABS Europe)
OptimizationTransformerSupervised Fine-TuningReinforcement LearningGraph
🎯 What it does: We propose GOAL, a general combinatorial optimization learning model that can learn and solve various combinatorial optimization problems in both single-task and multi-task scenarios.
GOFA: A Generative One-For-All Model for Joint Graph Language Modeling
Lecheng Kong (Washington University in St. Louis), Muhan Zhang (Peking University)
Graph Neural NetworkTransformerLarge Language ModelGraph
🎯 What it does: A generative integrated graph foundation model called GOFA has been constructed, which embeds graph neural networks into large language models and achieves zero-shot inference on various graph tasks.
Going Beyond Feature Similarity: Effective Dataset distillation based on Class-aware Conditional Mutual Information
Xinhao Zhong (Harbin Institute of Technology), EN-HUI YANG
Data SynthesisKnowledge DistillationConvolutional Neural NetworkImage
🎯 What it does: A dataset distillation loss based on Conditional Mutual Information (CMI) is proposed, utilizing CMI constraints in the feature space of a pre-trained network to reduce the class-aware complexity of synthetic data, resulting in a more learnable synthetic dataset.
Going Beyond Static: Understanding Shifts with Time-Series Attribution
Jiashuo Liu (Tsinghua University), Mihaela van der Schaar (University of Cambridge)
Anomaly DetectionExplainability and InterpretabilityTime SeriesBiomedical DataElectronic Health Records
🎯 What it does: A TSSA framework was constructed to perform fine-grained attribution of model performance degradation using time series feature indicators, helping to understand and address distribution drift issues in time series.
GOLD: Graph Out-of-Distribution Detection via Implicit Adversarial Latent Generation
Danny Wang (University of Queensland), Zi Huang (University of Queensland)
Anomaly DetectionOptimizationGraph Neural NetworkGenerative Adversarial NetworkGraph
🎯 What it does: A graph OOD detection framework called GOLD is proposed, which utilizes only ID data to self-generate pseudo OOD embeddings and discriminates them through energy scores.
GoodDrag: Towards Good Practices for Drag Editing with Diffusion Models
Zewei Zhang (McMaster University), Xiangyu Xu (Xi'an Jiaotong University)
GenerationData SynthesisDiffusion modelImage
🎯 What it does: The GoodDrag framework is proposed, achieving high-quality drag editing.
GotenNet: Rethinking Efficient 3D Equivariant Graph Neural Networks
Sarp Aykent (Comcast AI Technologies), Tian Xia (Microsoft)
Computational EfficiencyDrug DiscoveryGraph Neural NetworkGraph
🎯 What it does: A novel three-dimensional equivariant graph neural network, GotenNet, has been designed and implemented to efficiently and accurately capture the spatial structure and symmetry of molecular graphs.
GOttack: Universal Adversarial Attacks on Graph Neural Networks via Graph Orbits Learning
Zulfikar Alom (University of Manitoba), Cuneyt Gurcan Akcora (University of Central Florida)
Adversarial AttackGraph Neural NetworkGraph
🎯 What it does: This paper proposes a global adversarial attack framework called Gottack, based on graph orbit learning, to induce structural perturbations in Graph Neural Networks (GNNs) for node classification tasks, leading to misclassification.
GPromptShield: Elevating Resilience in Graph Prompt Tuning Against Adversarial Attacks
Shuhan Song (Chinese Academy of Sciences), Xiaochun Ye (Chinese Academy of Sciences)
ClassificationAdversarial AttackGraph Neural NetworkPrompt EngineeringGraph
🎯 What it does: This study proposes GPROMPTSHIELD, aimed at enhancing the robustness of graph prompts against adversarial attacks, forming a scalable protection system.
GPS: A Probabilistic Distributional Similarity with Gumbel Priors for Set-to-Set Matching
Ziming Zhang (Worcester Polytechnic Institute), Venkatesh Saligrama (Boston University)
ClassificationAutonomous DrivingGenerative Adversarial NetworkImagePoint Cloud
🎯 What it does: Proposes a Gumbel prior-based probability distribution similarity (GPS) loss for learning set-to-set matching.
GPUDrive: Data-driven, multi-agent driving simulation at 1 million FPS
Saman Kazemkhani (New York University), Eugene Vinitsky (New York University)
Autonomous DrivingReinforcement LearningPoint Cloud
🎯 What it does: Developed a GPU-accelerated multi-agent driving simulator called GPUDrive, capable of generating over a million simulation steps per second;
GrabS: Generative Embodied Agent for 3D Object Segmentation without Scene Supervision
Zihui Zhang (Hong Kong Polytechnic University), Bo Yang (Hong Kong Polytechnic University)
SegmentationRobotic IntelligenceReinforcement LearningDiffusion modelPoint Cloud
🎯 What it does: A two-stage unsupervised 3D object segmentation framework called GrabS is proposed, which first learns a generative object prior on a single-object dataset and then discovers and segments multiple objects in complex scenes through an embodied agent.
Gradient correlation is a key ingredient to accelerate SGD with momentum
Julien Hermant (University of Bordeaux), Aude Rondepierre (University of Toulouse)
OptimizationConvolutional Neural NetworkImageStochastic Differential Equation
🎯 What it does: This paper studies the acceleration effect of Stochastic Nesterov Accelerated Gradient (SNAG) compared to Stochastic Gradient Descent (SGD) under convex or strongly convex objective functions using Gradient Autocorrelation (RACOGA).
Gradient descent with generalized Newton’s method
Zhiqi Bu, Shiyun Xu (University of Pennsylvania)
Recommendation SystemOptimizationTransformerLarge Language ModelImageTextAudio
🎯 What it does: A general second-order optimization method is proposed—Generalized Newton's Method (GeN), which can automatically and dynamically determine the optimal learning rate for any optimizer (such as SGD, Adam) and accelerate convergence without the need for manual hyperparameter tuning.
Gradient-Free Generation for Hard-Constrained Systems
Chaoran Cheng (University of Illinois Urbana-Champaign), Bernie Wang
GenerationOptimizationFlow-based ModelTime SeriesSequentialPhysics RelatedOrdinary Differential Equation
🎯 What it does: The ECI sampling framework is proposed, utilizing a pre-trained flow matching model to achieve precise generation and regression of hard constraints such as PDEs in a zero-shot, gradient-free manner.
GRAIN: Exact Graph Reconstruction from Gradients
Maria Drencheva (INSAIT), Martin Vechev (ETH Zurich)
Federated LearningAdversarial AttackGraph Neural NetworkGraph
🎯 What it does: Proposed and implemented the GRAIN attack in the context of federated learning, which accurately recovers the graph structure and node features of graph neural networks by observing client gradients.
Gramian Multimodal Representation Learning and Alignment
Giordano Cicchetti (Sapienza University of Rome), Danilo Comminiello (Sapienza University of Rome)
RetrievalRepresentation LearningContrastive LearningVideoMultimodalityAudio
🎯 What it does: A Gramian correlation metric (GRAM) based on the volume of parallel polyhedra is proposed, achieving unified geometric alignment of multimodal embedding vectors.
Grammar Reinforcement Learning: path and cycle counting in graphs with a Context-Free Grammar and Transformer approach
Jason Piquenot (University of Rouen Normandy), Sébastien Adam (University of Rouen Normandy)
OptimizationTransformerReinforcement LearningGraph
🎯 What it does: This paper proposes a deep reinforcement learning framework based on context-free grammar (CFG) and transformers (Grammar Reinforcement Learning, GRL), and utilizes this framework to automatically discover efficient matrix formulas in graph path/cycle counting tasks.
Graph Assisted Offline-Online Deep Reinforcement Learning for Dynamic Workflow Scheduling
Yifan Yang (Victoria University of Wellington), Mengjie Zhang (Singapore Management University)
OptimizationGraph Neural NetworkReinforcement LearningGraph
🎯 What it does: A dynamic workflow scheduling method for cloud computing environments, GOODRL, is proposed, which can perform online/offline learning and scheduling of workflows on heterogeneous clusters with the goal of minimizing average flow time.
Graph Neural Networks Are More Than Filters: Revisiting and Benchmarking from A Spectral Perspective
Yushun Dong (Florida State University), Jundong Li (University of Virginia)
ClassificationGraph Neural NetworkGraphBenchmark
🎯 What it does: This paper proposes a comprehensive benchmarking of Graph Neural Networks (GNNs) from a spectral perspective and reveals through experiments that GNNs can flexibly generate outputs of different frequency components under the constraints of neighborhood aggregation mechanisms.
Graph Neural Networks Can (Often) Count Substructures
Paolo Pellizzoni (Max Planck Institute of Biochemistry), Karsten Borgwardt (Max Planck Institute of Biochemistry)
Drug DiscoveryGraph Neural NetworkGraphBiomedical Data
🎯 What it does: This study investigates the practical performance and theoretical limitations of message-passing graph neural networks (GNNs) in the subgraph counting task, providing sufficient conditions for GNNs to count subgraphs on real datasets, and designing a dynamic programming algorithm that can be simulated by GNNs, verifying the universality of these conditions in real molecular graph datasets.
Graph Neural Networks for Edge Signals: Orientation Equivariance and Invariance
Dominik Fuchsgruber (Technical University Munich), Simon Geisler (Technical University Munich)
Graph Neural NetworkGraph
🎯 What it does: A graph neural network framework specifically designed for edge-level tasks, EIGN, is proposed, which can simultaneously handle orientation-equivariant and orientation-invariant edge signals, and can distinguish between directed and undirected edges.
Graph Neural Networks Gone Hogwild
Olga Solodova (Princeton University), Ryan P Adams
Graph Neural NetworkGraph
🎯 What it does: This study investigates the robustness of graph neural networks (GNNs) in distributed asynchronous inference and proposes an implicit definition GNN architecture called Energy GNN, which can converge in asynchronous environments. Its performance is validated through synthetic and benchmark datasets.
Graph Neural Preconditioners for Iterative Solutions of Sparse Linear Systems
Jie Chen (MIT IBM Watson AI Lab)
OptimizationGraph Neural NetworkGraph
🎯 What it does: This paper proposes using graph neural networks to learn the inverse of sparse linear systems as a preconditioner, and embeds it into a flexible GMRES solver.