NeurIPS 2024 Papers — Page 16
Conference on Neural Information Processing Systems · 4035 papers
Geometric Analysis of Nonlinear Manifold Clustering
Nimita Shinde (Lehigh University), Rene Vidal
OptimizationImage
🎯 What it does: A nonlinear manifold clustering model based on self-expression is proposed, along with its corresponding convex optimization formulation.
Geometric Exploitation for Indoor Panoramic Semantic Segmentation
Duc Cao Dinh (MAXST), Kyusung Cho (MAXST)
SegmentationDepth EstimationOptimizationKnowledge DistillationTransformerPoint Cloud
🎯 What it does: A new framework is proposed that splits indoor panoramic semantic segmentation into oversampled areas (floor/ceiling) and undersampled areas, and jointly optimizes them using geometric information.
Geometric Trajectory Diffusion Models
Jiaqi Han (Stanford University), Stefano Ermon (Stanford University)
GenerationData SynthesisGraph Neural NetworkDiffusion modelTime SeriesSequentialPhysics Related
🎯 What it does: This paper proposes a geometric trajectory diffusion model called GeoTDM, designed to generate 3D trajectories that satisfy physical symmetry.
Geometric-Averaged Preference Optimization for Soft Preference Labels
Hiroki Furuta (University of Tokyo), Izzeddin Gur (Google DeepMind)
Recommendation SystemOptimizationTransformerLarge Language ModelText
🎯 What it does: Proposes soft preference labels and incorporates weighted geometric mean into preference optimization algorithms such as DPO.
Geometry Awakening: Cross-Geometry Learning Exhibits Superiority over Individual Structures
Yadong Sun (Jilin University), Qing Guo (Agency for Science Technology and Research)
Knowledge DistillationRepresentation LearningGraph Neural NetworkGraph
🎯 What it does: This paper proposes a cross-geometric knowledge distillation framework that combines teacher models from Euclidean space and hyperbolic space. It adaptively selects appropriate geometric embeddings based on the δ-hyperbolicity of subgraphs and achieves efficient distillation of the student model through two modules: Structured Knowledge Transfer (SWKT) and Geometric Embedding Optimization (GEO).
Geometry Cloak: Preventing TGS-based 3D Reconstruction from Copyrighted Images
Qi Song (Hong Kong Baptist University), Renjie Wan (Hong Kong Baptist University)
OptimizationAdversarial AttackGaussian SplattingPoint Cloud
🎯 What it does: A 'geometric invisibility' method is proposed in the field of image copyright protection, which injects imperceptible perturbations into single-view images, forcing Triplane Gaussian Splatting (TGS) to generate erroneous 3D models with recognizable watermarks, thereby preventing unauthorized 3D reconstruction.
Geometry of naturalistic object representations in recurrent neural network models of working memory
Xiaoxuan Lei (McGill University), Pouya Bashivan (Mila)
Representation LearningConvolutional Neural NetworkRecurrent Neural NetworkImage
🎯 What it does: This paper trains a perception-cognition model that combines CNN and RNN to learn the multidimensional features of natural image objects in a multi-task N-back working memory task, and studies the geometric properties of its latent space.
Geometry-aware training of factorized layers in tensor Tucker format
Emanuele Zangrando (Gran Sasso Science Institute), Francesco Tudisco (University of Edinburgh)
CompressionOptimizationConvolutional Neural NetworkSupervised Fine-TuningImage
🎯 What it does: A geometric-aware tensor Tucker format layer training algorithm is proposed, which can dynamically and adaptively adjust the layer rank during the training process, achieving a high compression rate for network compression.
GeoNLF: Geometry guided Pose-Free Neural LiDAR Fields
Weiyi Xue (Tongji University), changjun jiang
Pose EstimationAutonomous DrivingOptimizationNeural Radiance FieldSimultaneous Localization and MappingPoint Cloud
🎯 What it does: Proposes the GeoNLF framework, which combines multi-view point cloud registration and pose-free LiDAR-NeRF for 3D scene reconstruction and viewpoint synthesis.
Get rich quick: exact solutions reveal how unbalanced initializations promote rapid feature learning
Daniel Kunin (Stanford University), Surya Ganguli (Stanford University)
OptimizationRepresentation LearningConvolutional Neural NetworkImageOrdinary Differential Equation
🎯 What it does: This paper derives an analytical solution that can accurately describe the transition from lazy to rich learning by solving the gradient flow equation of a minimal linear network. The method is extended to wide linear networks and shallow nonlinear networks, further validating the acceleration effect of unbalanced initialization on feature learning across various tasks.
Get Rid of Isolation: A Continuous Multi-task Spatio-Temporal Learning Framework
Zhongchao Yi (University of Science and Technology of China), Yang Wang (University of Science and Technology of China)
TransformerAuto EncoderTime Series
🎯 What it does: The CMuST (Continuous Multi-task Spatio-Temporal) framework is proposed, which enables joint learning of multiple spatio-temporal prediction tasks within the same urban system and achieves rapid adaptation to new tasks through continuous rolling training.
Getting More Juice Out of the SFT Data: Reward Learning from Human Demonstration Improves SFT for LLM Alignment
Jiaxiang Li (University of Minnesota), Mingyi Hong (University of Minnesota)
OptimizationReinforcement Learning from Human FeedbackTransformerLarge Language ModelSupervised Fine-TuningReinforcement LearningText
🎯 What it does: A reward learning and policy fine-tuning framework based on inverse reinforcement learning is proposed, which constructs a reward model using demonstration data and improves LLM alignment during the SFT phase.
GFlowNet Assisted Biological Sequence Editing
Pouya M. Ghari (University of California Irvine), Ehsan Hajiramezanali (Genentech)
Drug DiscoveryFlow-based ModelSequentialBiomedical Data
🎯 What it does: A sequence editing method based on GFlowNet, GFNSeqEditor, is proposed. It first uses a pre-trained flow function to identify editable sites in the seed sequence, and then edits these sites according to a random strategy to enhance target attributes while controlling the number of edits.
GFT: Graph Foundation Model with Transferable Tree Vocabulary
Zehong Wang (University of Notre Dame), Yanfang Ye (University of Notre Dame)
ClassificationDomain AdaptationDrug DiscoveryGraph Neural NetworkContrastive LearningGraphBiomedical Data
🎯 What it does: This paper proposes a cross-task and cross-domain graph-based model GFT, utilizing computational trees (subtrees obtained from the message passing process) as transferable vocabulary to construct a pre-training and fine-tuning process.
GIC: Gaussian-Informed Continuum for Physical Property Identification and Simulation
Junhao Cai (Alibaba Group), Qifeng Chen (Alibaba Group)
Gaussian SplattingPoint CloudPhysics Related
🎯 What it does: A hybrid framework combining 3D Gaussian representation with MPM physical simulation is constructed, using dynamic Gaussian reconstruction to obtain accurate geometry and employing Gaussian-generated continuous rendering of 2D masks to assist in physical property estimation.
GITA: Graph to Visual and Textual Integration for Vision-Language Graph Reasoning
Yanbin Wei (Southern University of Science and Technology), Yu Zhang (Hong Kong University of Science and Technology)
Graph Neural NetworkLarge Language ModelVision Language ModelMultimodalityGraph
🎯 What it does: An end-to-end GITA framework is proposed, which renders graph structures into visual graphs and combines them with textual descriptions for graph reasoning.
GL-NeRF: Gauss-Laguerre Quadrature Enables Training-Free NeRF Acceleration
Silong Yong (Carnegie Mellon University), Katia P. Sycara
Data SynthesisComputational EfficiencyNeural Radiance FieldImage
🎯 What it does: This paper proposes GL-NeRF, which directly rewrites the NeRF volume rendering integral using Gauss-Laguerre quadrature of order four, significantly reducing the number of calls to the color MLP without requiring additional networks or training.
Gliding over the Pareto Front with Uniform Designs
Xiaoyuan Zhang (City University of Hong Kong), Qingfu Zhang (City University of Hong Kong)
OptimizationTabular
🎯 What it does: A unified multi-objective optimization method UMOD based on maximum packing design is proposed, which directly maximizes the minimum dual distance to achieve uniform coverage of the Pareto front.
GLinSAT: The General Linear Satisfiability Neural Network Layer By Accelerated Gradient Descent
Hongtai Zeng (Tsinghua University), Qinglai Guo (Tsinghua University)
OptimizationReinforcement Learning
🎯 What it does: A differentiable linear constraint satisfaction layer, GLinSAT, is proposed to project the output of neural networks into a feasible domain that satisfies general linear and bounded constraints.
Global Convergence in Training Large-Scale Transformers
Cheng Gao (Princeton University), Jianqing Fan (Princeton University)
OptimizationTransformer
🎯 What it does: This paper proves through mean field analysis that as the width and depth of the Transformer approach infinity, gradient flow training can globally converge to the optimal solution, and provides an upper bound on the weak convergence error of discrete gradient flow to continuous Wasserstein gradient flow.
Global Distortions from Local Rewards: Neural Coding Strategies in Path-Integrating Neural Systems
Francisco Acosta (University of California Santa Barbara), Nina Miolane (University of California Santa Barbara)
Recurrent Neural NetworkReinforcement LearningSequential
🎯 What it does: This paper studies how local rewards can distort the global firing patterns of grid cells by training a path integral recurrent neural network (piRNN) with a reward-guided loss. It connects the geometric deformation of the firing field with the topological geometry of the neural network (two-dimensional torus); it also verifies that the toroidal topology of grid cells can be preserved when the readout weights are frozen, and the resulting distortion is a global property.
Global Lyapunov functions: a long-standing open problem in mathematics, with symbolic transformers
Alberto Alfarano (Meta), Amaury Hayat (Institut Polytechnique de Paris)
Transformer
🎯 What it does: This paper proposes the use of generated Lyapunov function data to train a sequence-to-sequence transformer, automatically discovering global Lyapunov functions for stable dynamical systems.
Global Rewards in Restless Multi-Armed Bandits
Naveen Janaki Raman (Carnegie Mellon University), Fei Fang (Carnegie Mellon University)
OptimizationReinforcement LearningTabular
🎯 What it does: The RMAB-G model is proposed to address the issue of traditional RMABs being unable to handle non-separable global rewards, and introduces linear and Shapley Whittle indices, as well as iterative and MCTS adaptive strategies;
Globally Convergent Variational Inference
Declan McNamara (University of Michigan), Jeffrey Regier (University of Michigan)
OptimizationTabular
🎯 What it does: This paper studies the forward KL objective in variational inference and proves that gradient descent can converge to a unique global optimal solution under infinitely wide neural networks.
Globally Q-linear Gauss-Newton Method for Overparameterized Non-convex Matrix Sensing
Xixi Jia (Xidian University), Defeng Sun (Hong Kong Polytechnic University)
Optimization
🎯 What it does: An Approximate Gauss-Newton (AGN) method is proposed to solve the over-parameterized non-convex low-rank matrix sensing problem, and it is proven to achieve global Q-linear convergence from random initialization.
GO4Align: Group Optimization for Multi-Task Alignment
Jiayi Shen (University of Amsterdam), Marcel Worring (University of Amsterdam)
OptimizationImage
🎯 What it does: This paper proposes GO4Align, a loss-oriented optimization method for aligning the learning progress of various tasks in multi-task learning through dynamic group allocation guided by task grouping and risk, aimed at alleviating the task imbalance problem.
Goal Conditioned Reinforcement Learning for Photo Finishing Tuning
Jiarui Wu (Shanghai AI Laboratory), Tianfan Xue (Shanghai AI Laboratory)
Image TranslationConvolutional Neural NetworkReinforcement LearningImage
🎯 What it does: A photo completion adjustment algorithm based on target-conditioned reinforcement learning is proposed, which can quickly approach the target image with only 10 queries in a black-box image processing pipeline.
Goal Reduction with Loop-Removal Accelerates RL and Models Human Brain Activity in Goal-Directed Learning
Huzi Cheng (Indiana University), Joshua W Brown
Robotic IntelligenceReinforcement LearningAgentic AIBiomedical DataMagnetic Resonance Imaging
🎯 What it does: This paper proposes a goal degradation mechanism called goal-reducer based on cyclic removal, which is used to decompose a distant ultimate goal into approximately optimal sub-goals.
Goal-Conditioned On-Policy Reinforcement Learning
Gong Xudong, Huaimin Wang (National University of Defense Technology)
Reinforcement Learning
🎯 What it does: A multi-objective reinforcement learning framework based on on-policy, called GCPO, is proposed to address the learning bottleneck in non-Markov reward (NMR) scenarios.
Going Beyond Heuristics by Imposing Policy Improvement as a Constraint
Chi-Chang Lee (National Taiwan University), Pulkit Agrawal (Improbable AI Lab)
Robotic IntelligenceReinforcement Learning
🎯 What it does: A new reinforcement learning method is proposed, which surpasses traditional heuristic reward methods by treating policy improvement as a constraint to enhance task performance in limited data situations.
GOMAA-Geo: GOal Modality Agnostic Active Geo-localization
Anindya Sarkar (Washington University in St. Louis), Yevgeniy Vorobeychik (Washington University in St. Louis)
Object DetectionTransformerLarge Language ModelReinforcement LearningContrastive LearningImageTextMultimodality
🎯 What it does: This paper proposes a cross-modal, zero-shot active geographic localization (AGL) agent called GOMAA-Geo, which can achieve target localization using only aerial images for training, leveraging textual or ground image descriptions.
GoMatching: A Simple Baseline for Video Text Spotting via Long and Short Term Matching
Haibin He (Wuhan University), Dacheng Tao (Nanyang Technological University)
RecognitionObject DetectionObject TrackingTransformerVideoBenchmark
🎯 What it does: A simple and efficient baseline GoMatching is constructed, using the frozen image text detector DeepSolo for text detection and recognition, and tracking is performed through a lightweight rescoring head and a long-short term matching module LST-Matcher.
Gorilla: Large Language Model Connected with Massive APIs
Shishir G Patil, Joseph E. Gonzalez (University of California Berkeley)
RetrievalAI Code AssistantTransformerLarge Language ModelSupervised Fine-TuningTextBenchmarkRetrieval-Augmented Generation
🎯 What it does: The Gorilla system has been constructed, combining large-scale API documentation retrieval with LLM, capable of automatically generating accurate API calls based on user instructions; it also provides the APIBench benchmark and AST subtree matching evaluation method.
Gradient Cuff: Detecting Jailbreak Attacks on Large Language Models by Exploring Refusal Loss Landscapes
Xiaomeng Hu (Chinese University of Hong Kong), Tsung-Yi Ho (Chinese University of Hong Kong)
Anomaly DetectionTransformerLarge Language ModelText
🎯 What it does: A two-stage 'Gradient Cuff' detection method based on the rejection loss function and its gradient features is proposed, which significantly improves the detection rate of various jailbreak attacks by large language models while maintaining a low false positive rate.
Gradient Guidance for Diffusion Models: An Optimization Perspective
Yingqing Guo (Princeton University), Mengdi Wang (Princeton University)
GenerationOptimizationDiffusion modelScore-based ModelImageStochastic Differential Equation
🎯 What it does: This paper proposes a gradient-guided optimization framework for diffusion models, utilizing a pre-trained diffusion model and the gradient of the objective function to generate samples that meet the objectives while preserving the latent low-dimensional structure of the data.
Gradient Methods for Online DR-Submodular Maximization with Stochastic Long-Term Constraints
Guanyu Nie (Iowa State University), Christopher John Quinn (Iowa State University)
Optimization
🎯 What it does: A gradient ascent algorithm is proposed for the online DR-submodular function maximization problem with random long-term constraints, which can achieve sublinear loss under both semi-bandit and full information feedback.
Gradient Rewiring for Editable Graph Neural Network Training
Zhimeng Jiang (Texas A&M University), Xia Hu (Rice University)
ClassificationOptimizationGraph Neural NetworkGraph
🎯 What it does: To address the model editing problem in graph neural networks, we propose Gradient Rewiring (GRE) and its improved version GRE+, which achieve rapid correction of prediction errors for single or consecutive nodes by making local adjustments using only the gradients of the target nodes, while ensuring that the training error does not increase.
Gradient-based Discrete Sampling with Automatic Cyclical Scheduling
Patrick Pynadath (Purdue University), Ruqi Zhang (Purdue University)
OptimizationTransformerLarge Language ModelTextMultimodality
🎯 What it does: This paper proposes an Automatic Cyclical Sampler (ACS) for automatic periodic scheduling of gradient discrete sampling, which achieves efficient exploration and accurate sampling of multimodal discrete distributions by periodically adjusting the step size and balance parameters.
Gradient-free Decoder Inversion in Latent Diffusion Models
Seongmin Hong (Seoul National University), Se Young Chun (Seoul National University)
GenerationOptimizationComputational EfficiencyDiffusion modelImageVideo
🎯 What it does: A gradient-independent decoder inversion method is proposed in the latent diffusion model, utilizing the forward step method and momentum extension to achieve efficient inverse mapping.
Gradient-Free Methods for Nonconvex Nonsmooth Stochastic Compositional Optimization
Zhuanghua Liu (National University of Singapore), Bryan Kian Hsiang Low (National University of Singapore)
OptimizationReinforcement LearningTabularSequentialFinance Related
🎯 What it does: This paper proposes two gradient-free stochastic methods (GFCOM and GFCOM+) to solve non-convex non-smooth stochastic combinatorial optimization (SCO) problems, and provides corresponding non-asymptotic convergence analysis; it then extends to convex non-smooth cases through a warm-start strategy, achieving better convergence rates.
Gradient-Variation Online Learning under Generalized Smoothness
Yan-Feng Xie (Nanjing University), Zhi-Hua Zhou (Nanjing University)
Optimization
🎯 What it does: In the framework of online convex optimization, a gradient variation learning theory under generalized smoothness conditions is proposed, and a unified online learning algorithm is constructed that can handle both convex and strongly convex functions.
Gradients of Functions of Large Matrices
Nicholas Krämer (Technical University of Denmark), Søren Hauberg (Technical University of Denmark)
OptimizationTabularOrdinary Differential Equation
🎯 What it does: Achieved differentiable Lanczos and Arnoldi iterations for matrix functions (such as logarithmic determinants, matrix exponentials, etc.), providing closed-form adjoint systems;
Gradual Domain Adaptation via Manifold-Constrained Distributionally Robust Optimization
seyed amir hossein saberi, Babak Khalaj
Domain AdaptationOptimizationImage
🎯 What it does: This paper proposes a progressive domain adaptation method based on manifold-constrained distributionally robust optimization, which can maintain model performance and suppress error propagation under continuous domain shifts.
Grammar-Aligned Decoding
Kanghee Park (University of Wisconsin-Madison), Loris D'Antoni (University of Wisconsin-Madison)
GenerationTransformerLarge Language ModelText
🎯 What it does: This paper studies how to maintain the original probability distribution when generating text constrained by context-free grammar in large language models, proposing the Grammar-Aligned Decoding (GAD) problem;
GRANOLA: Adaptive Normalization for Graph Neural Networks
Moshe Eliasof (University of Cambridge), Haggai Maron (Technion and NVIDIA Research)
Graph Neural NetworkGraph
🎯 What it does: A new graph neural network normalization layer called GRANOLA is proposed, which can adaptively normalize node features based on the graph structure.
Graph Classification via Reference Distribution Learning: Theory and Practice
Zixiao Wang (Chinese University of Hong Kong), Jicong Fan (Chinese University of Hong Kong)
ClassificationGraph Neural NetworkGraph
🎯 What it does: A new graph classification framework GRDL is proposed, which achieves efficient and accurate graph classification by directly comparing the distribution of node embeddings without using global pooling.
Graph Coarsening with Message-Passing Guarantees
Antonin Joly (Institute for Research in Computer Science and Random Systems), Nicolas Keriven (National Centre for Scientific Research)
ClassificationGraph Neural NetworkGraph
🎯 What it does: This paper studies the interaction between graph coarsening and message propagation in graph neural networks (GNNs) and proposes a new propagation matrix for coarsened graphs to ensure that the results of GNNs trained on coarsened graphs are close to those trained on the original graphs.
Graph Convolutions Enrich the Self-Attention in Transformers!
Jeongwhan Choi (Yonsei University), Noseong Park (KAIST)
TransformerImageTextAudio
🎯 What it does: Redesign the self-attention of the Transformer as a graph filter and propose the GFSA mechanism to replace the attention layers in various Transformer models.
Graph Diffusion Policy Optimization
Yijing Liu (Zhejiang University), Wei Chen (Renmin University of China)
OptimizationDrug DiscoveryGraph Neural NetworkReinforcement LearningDiffusion modelGraph
🎯 What it does: This paper proposes Graph Diffusion Policy Optimization (GDPO), a method that treats the reverse sampling of the Graph Diffusion Probabilistic Model (Graph DPM) as a Markov Decision Process and optimizes arbitrary (especially non-differentiable) reward objectives through reinforcement learning.
Graph Diffusion Transformers for Multi-Conditional Molecular Generation
Gang Liu (University of Notre Dame), Meng Jiang (University of Notre Dame)
GenerationData SynthesisDrug DiscoveryGraph Neural NetworkTransformerGraphTabular
🎯 What it does: Proposes the Graph DiT model, achieving multi-conditional molecular generation that can simultaneously control various attributes;
Graph Edit Distance with General Costs Using Neural Set Divergence
Eeshaan Jain (École Polytechnique Fédérale de Lausanne), Abir De (Indian Institute of Technology Bombay)
Graph Neural NetworkGraph
🎯 What it does: A neural network framework is proposed that can approximate the graph edit distance (GED) while considering the general costs of any four types of edit operations.
Graph Learning for Numeric Planning
Dillon Ze Chen, Sylvie Thiebaux
OptimizationGraph Neural NetworkGraph
🎯 What it does: This paper proposes a graph learning framework for numerical planning tasks, achieving efficient planning solutions by constructing interpretable graph features and learning heuristic functions.
Graph Neural Flows for Unveiling Systemic Interactions Among Irregularly Sampled Time Series
Giangiacomo Mercatali (HES-SO Geneva University of Manchester), Jie Chen (IBM Research)
ClassificationOptimizationExplainability and InterpretabilityRecurrent Neural NetworkGraph Neural NetworkFlow-based ModelTime SeriesSequentialBiomedical DataOrdinary Differential Equation
🎯 What it does: A Graph Neural Flow (GNeuralFlow) model is proposed to learn systematic interactions in irregularly sampled time series while simultaneously learning the corresponding Directed Acyclic Graph (DAG) structure.
Graph Neural Networks and Arithmetic Circuits
Timon Barlag (Institute for Theoretical Computer Science Leibniz University Hanover), Heribert Vollmer (Institute for Theoretical Computer Science Leibniz University Hanover)
Graph Neural Network
🎯 What it does: This paper establishes an equivalence between Graph Neural Networks (C-GNN) and real arithmetic circuits through theoretical analysis, proving that both have a complete match in expressive power.
Graph neural networks and non-commuting operators
Mauricio Velasco (Universidad Católica del Uruguay), Soledad Villar (Johns Hopkins University)
Recommendation SystemGraph Neural NetworkGraph
🎯 What it does: This paper proposes Graph-tuple Neural Networks (GtNN), a multimodal graph neural network model that jointly processes multiple graphs sharing a common set of vertices.
Graph Neural Networks Do Not Always Oversmooth
Bastian Epping (RWTH Aachen University), Michael T Schaub
Graph Neural NetworkGraph
🎯 What it does: This paper utilizes the equivalence of GCN and high-dimensional Gaussian processes (GP) to analyze the over-smoothing problem (features gradually converging to the same vector) that occurs in deep graph convolutional networks (GCN). It proves that by adjusting the variance of the initialized weights, one can enter a new non-over-smoothing (chaotic) phase, thereby supporting deeper networks while maintaining expressive power.
Graph Neural Networks Need Cluster-Normalize-Activate Modules
Arseny Skryagin (Technical University of Darmstadt), Kristian Kersting (Technical University of Darmstadt)
Graph Neural NetworkGraph
🎯 What it does: A pluggable Cluster-Normalize-Activate (CNA) module is proposed to enhance the expressive power of Graph Neural Networks (GNNs) and suppress the over-smoothing phenomenon.
Graph Structure Inference with BAM: Neural Dependency Processing via Bilinear Attention
Philipp Froehlich (Technische Universitaet Darmstadt), Heinz Koeppl (Technische Universitaet Darmstadt)
Graph Neural NetworkTransformerSupervised Fine-TuningGraph
🎯 What it does: Learn how to infer graph structures from observational data, using supervised neural networks to predict causal dependencies between variables.
Graph-based Uncertainty Metrics for Long-form Language Model Generations
Mingjian Jiang (Stanford University), Tatsunori Hashimoto (Stanford University)
GenerationGraph Neural NetworkTransformerLarge Language ModelText
🎯 What it does: This paper proposes the Graph Uncertainty framework, which implements fine-grained uncertainty assessment of each claim in long text generation using semantic entailment graphs and graph centrality measures. This assessment is used to filter low-confidence claims during decoding, enhancing the factuality and informativeness of the text.
Graph-based Unsupervised Disentangled Representation Learning via Multimodal Large Language Models
Baao Xie (Ningbo Institute of Digital Twin), Wenjun Zeng (Ningbo Institute of Digital Twin)
GenerationRepresentation LearningGraph Neural NetworkLarge Language ModelAuto EncoderImageMultimodality
🎯 What it does: This paper proposes an unsupervised representation learning framework (GEM) based on β-VAE and multimodal large language models (MLLM), achieving attribute extraction and learning of inter-attribute relationships simultaneously through a bidirectional weighted graph (DisGraph).
Graph-enhanced Optimizers for Structure-aware Recommendation Embedding Evolution
Cong Xu (East China Normal University), Wei Zhang (East China Normal University)
Recommendation SystemOptimizationGraph Neural NetworkReinforcement LearningTabularSequential
🎯 What it does: A structure-aware embedding evolution mechanism SEvo is proposed, which directly injects graph structure information into the embedding updates without the need for an explicit GNN module;
Graphcode: Learning from multiparameter persistent homology using graph neural networks
Florian Russold (Graz University of Technology), Michael Kerber (Graz University of Technology)
ClassificationComputational EfficiencyGraph Neural NetworkPoint CloudGraph
🎯 What it does: A new multi-scale topological summary called Graphcode is proposed and implemented to describe the multi-parameter persistent homology of datasets filtered along two scale parameters, and it is used as an embedded graph input for graph neural networks for classification.
GraphCroc: Cross-Correlation Autoencoder for Graph Structural Reconstruction
Shijin Duan (Northeastern University), Xiaolin Xu (Northeastern University)
Representation LearningAdversarial AttackGraph Neural NetworkAuto EncoderGraph
🎯 What it does: A graph autoencoder called GraphCroc based on cross-correlation decoding is proposed for structural reconstruction and downstream tasks in multi-graph scenarios.
GraphMETRO: Mitigating Complex Graph Distribution Shifts via Mixture of Aligned Experts
Shirley Wu (Stanford University), Jure Leskovec (Stanford University)
Domain AdaptationGraph Neural NetworkMixture of ExpertsGraph
🎯 What it does: A Mixture-of-Experts based graph neural network, GraphMETRO, is proposed to address complex and unknown graph distribution shifts in the real world.
GraphMorph: Tubular Structure Extraction by Morphing Predicted Graphs
Zhao Zhang (Peking University), Liwei Wang (Peking University)
Object DetectionSegmentationGraph Neural NetworkImageBiomedical DataMagnetic Resonance Imaging
🎯 What it does: The GraphMorph framework is proposed, which utilizes a graph decoder to predict the shapes of tubular structures and transforms the shapes into topologically accurate centerline masks using the SkeletonDijkstra algorithm, thereby achieving precise extraction of vascular or road networks.
GraphTrail: Translating GNN Predictions into Human-Interpretable Logical Rules
Burouj Armgaan (Indian Institute of Technology Delhi), Sayan Ranu (Indian Institute of Technology Delhi)
Explainability and InterpretabilityGraph Neural NetworkGraph
🎯 What it does: Designed and implemented GRAPHTRAIL, a full-process global GNN interpreter that maps the graph-level predictions of black-box GNNs to Boolean logic rules based on subgraph concepts.
GraphVis: Boosting LLMs with Visual Knowledge Graph Integration
Yihe Deng (University of California), Wei Wang (University of California)
Graph Neural NetworkTransformerLarge Language ModelSupervised Fine-TuningVision Language ModelImageTextGraph
🎯 What it does: Visualize the subgraphs of knowledge graphs as images, allowing visual language models to first learn basic graphical features through curriculum-based fine-tuning, and then apply this knowledge to question-answer reasoning, thereby enhancing the understanding of knowledge graphs (KG) and the performance of text/visual question answering (QA) in large language models (LLM).
Grasp as You Say: Language-guided Dexterous Grasp Generation
Yi-Lin Wei (Sun Yat-sen University), Wei-Shi Zheng (Sun Yat-sen University)
GenerationRobotic IntelligenceLarge Language ModelDiffusion modelTextPoint Cloud
🎯 What it does: This paper proposes a multi-finger robotic grasping task based on natural language instructions (DexGYS), constructs the DexGYSNet dataset, and introduces the DexGYSGrasp framework to achieve intent alignment, diversity, and high-quality grasp pose generation.
Great Minds Think Alike: The Universal Convergence Trend of Input Salience
Yipei Wang (Purdue University), Xiaoqian Wang (Purdue University)
OptimizationAdversarial AttackConvolutional Neural NetworkRecurrent Neural NetworkImage
🎯 What it does: This study investigates the input gradient (saliency) distribution of deep neural networks obtained through stochastic optimization and proposes that this distribution tends to converge in the same direction as the model capacity increases.
GREAT Score: Global Robustness Evaluation of Adversarial Perturbation using Generative Models
ZAITANG LI (Chinese University of Hong Kong), Tsung-Yi Ho (Chinese University of Hong Kong)
GenerationAdversarial AttackDiffusion modelGenerative Adversarial NetworkImage
🎯 What it does: A GREAT Score is proposed, a global adversarial robustness evaluation metric based on generative models;
GREATS: Online Selection of High-Quality Data for LLM Training in Every Iteration
Jiachen T. Wang (Princeton University), Ruoxi Jia (Virginia Tech)
OptimizationComputational EfficiencyData-Centric LearningTransformerLarge Language ModelSupervised Fine-TuningText
🎯 What it does: This paper proposes an online batch selection algorithm called GREATS, which utilizes the Taylor expansion of gradients and greedy optimization to dynamically select the data that can most improve validation set performance during training, thereby accelerating the convergence of LLM training and enhancing generalization.
Grid4D: 4D Decomposed Hash Encoding for High-Fidelity Dynamic Gaussian Splatting
Jiawei Xu (Nankai University), Jin Xie (Nanjing University)
Gaussian SplattingPoint Cloud
🎯 What it does: This paper presents Grid4D, a dynamic scene rendering model based on Gaussian Splatting;
Grokking of Implicit Reasoning in Transformers: A Mechanistic Journey to the Edge of Generalization
Boshi Wang (Ohio State University), Huan Sun (Ohio State University)
TransformerLarge Language Model
🎯 What it does: This study investigates the implicit reasoning ability of transformers on parameterized knowledge without explicit reasoning steps, observing the grokking phenomenon and systematic differences through synthetic combination and comparison tasks, revealing the formation of general reasoning circuits using logit lens and causal tracing.
Grounded Answers for Multi-agent Decision-making Problem through Generative World Model
Zeyang Liu (Xi'an Jiaotong University), Xuguang Lan (Xi'an Jiaotong University)
TransformerReinforcement LearningWorld ModelImage
🎯 What it does: This paper proposes a language-guided simulator that combines multi-agent reinforcement learning with a world model to generate interpretable multi-agent decision answers.
Grounding Multimodal Large Language Models in Actions
Andrew Szot (Apple), Alexander T Toshev
Robotic IntelligenceReinforcement Learning from Human FeedbackTransformerLarge Language ModelReinforcement LearningVision-Language-Action ModelMultimodality
🎯 What it does: This study investigates how to connect multimodal large language models (MLLM) to different embodied agents through various Action Space Adapters (ASA) to achieve control strategies under language conditions.
GrounDiT: Grounding Diffusion Transformers via Noisy Patch Transplantation
Yuseung Lee (KAIST), Minhyuk Sung (KAIST)
Object DetectionGenerationTransformerDiffusion modelImage
🎯 What it does: This paper presents GROUNDIT, a training-free spatial positioning method that accurately places targets within user-specified boundaries in text-image generation.
Group and Shuffle: Efficient Structured Orthogonal Parametrization
Mikhail Gorbunov (Higher School of Economics University), Maxim Rakhuba (Higher School of Economics University)
GenerationData SynthesisComputational EfficiencyConvolutional Neural NetworkTransformerDiffusion modelImageText
🎯 What it does: This paper proposes a new class of structured matrices—GS matrices—and constructs structured orthogonal parameterization based on this matrix. Its effectiveness is then validated in various tasks (text understanding, text-to-image generation, 1-Lipschitz networks).
Group Robust Preference Optimization in Reward-free RLHF
Shyam Sundhar Ramesh (University College London), Ilija Bogunovic (University College London)
OptimizationReinforcement Learning from Human FeedbackTransformerLarge Language ModelReinforcement LearningText
🎯 What it does: This paper proposes the Group Robust Preference Optimization (GRPO) method, which aims to achieve fair and robust LLM fine-tuning for different user groups within a reward-agnostic RLHF framework.
Group-wise oracle-efficient algorithms for online multi-group learning
Samuel Deng (Columbia University), Daniel Hsu (Columbia University)
Optimization
🎯 What it does: Designed an oracle-efficient online multi-group learning algorithm that can achieve sublinear logarithmic regret on large-scale or even infinite group sets;
GS-Hider: Hiding Messages into 3D Gaussian Splatting
Xuanyu Zhang (Peking University), Jian Zhang (Peking University)
Data SynthesisSafty and PrivacyGaussian SplattingPoint Cloud
🎯 What it does: A steganography framework called GS-Hider is designed for 3D Gaussian Splatting (3DGS), capable of imperceptibly hiding complete 3D scenes or images within 3DGS point clouds and achieving precise extraction.
GSDF: 3DGS Meets SDF for Improved Neural Rendering and Reconstruction
Mulin Yu (Shanghai Artificial Intelligence Laboratory), Bo Dai (Shanghai Artificial Intelligence Laboratory)
RestorationGenerationNeural Radiance FieldGaussian SplattingPoint Cloud
🎯 What it does: A dual-branch architecture is proposed, combining 3D Gaussian Splatting with Signed Distance Field to achieve a unified model that optimizes both rendering and geometric reconstruction simultaneously.
GSGAN: Adversarial Learning for Hierarchical Generation of 3D Gaussian Splats
Sangeek Hyun (Sungkyunkwan University), Jae-Pil Heo (Sungkyunkwan University)
GenerationData SynthesisTransformerGenerative Adversarial NetworkGaussian SplattingImage
🎯 What it does: A 3D Gaussian segmentation-based Generative Adversarial Network (GSGAN) is proposed, achieving efficient 3D generation through hierarchical Gaussian representation.
GTA: Generative Trajectory Augmentation with Guidance for Offline Reinforcement Learning
Jaewoo Lee (Korea Advanced Institute of Science and Technology), Jinkyoo Park (Korea Advanced Institute of Science and Technology)
Robotic IntelligenceReinforcement LearningDiffusion modelSequential
🎯 What it does: A generative trajectory augmentation method (GTA) based on conditional diffusion models is proposed, which generates high-reward and dynamically feasible new trajectories to augment offline datasets by partially adding noise to the original trajectories and denoising under the guidance of amplified rewards.
GTBench: Uncovering the Strategic Reasoning Capabilities of LLMs via Game-Theoretic Evaluations
Jinhao Duan (Drexel University), Kaidi Xu (Drexel University)
TransformerLarge Language ModelPrompt EngineeringTextBenchmarkChain-of-Thought
🎯 What it does: GTBench is designed to evaluate the strategic reasoning capabilities of LLMs in game theory tasks, and experiments are conducted for LLM-vs-LLM and LLM-vs-traditional solver matches.
GuardT2I: Defending Text-to-Image Models from Adversarial Prompts
Yijun Yang (Chinese University of Hong Kong), Qiang Xu (Chinese University of Hong Kong)
GenerationAdversarial AttackTransformerLarge Language ModelPrompt EngineeringImageText
🎯 What it does: A generative text censorship framework GUARDT2I has been designed and implemented, which converts text-guided embeddings into natural language through a conditional LLM, thereby detecting and preventing NSFW image generation caused by adversarial prompts.
GUIDE: Real-Time Human-Shaped Agents
Lingyu Zhang (Duke University), Boyuan Chen (Duke University)
Robotic IntelligenceReinforcement Learning from Human FeedbackConvolutional Neural NetworkReinforcement LearningMultimodality
🎯 What it does: The GUIDE framework is proposed, which combines real-time human continuous feedback with reinforcement learning, utilizing continuous human feedback to convert it into dense rewards to accelerate policy learning, and achieving continuous training without human intervention through a learned human feedback simulator.
Guided Trajectory Generation with Diffusion Models for Offline Model-based Optimization
Taeyoung Yun (Korea Advanced Institute of Science and Technology), Jinkyoo Park (Korea Advanced Institute of Science and Technology)
OptimizationDiffusion modelTabularBenchmark
🎯 What it does: This paper proposes using conditional diffusion models to generate trajectories leading to high partitions in offline model optimization. It first constructs diverse improved trajectories through local search, then trains the diffusion model and incorporates contextual conditions and classifier-free guidance during sampling, and finally filters candidate designs using a proxy function.
Guiding a Diffusion Model with a Bad Version of Itself
Tero Karras (NVIDIA), Samuli Laine (NVIDIA)
GenerationDiffusion modelImageOrdinary Differential Equation
🎯 What it does: An autoguidance method is proposed, which uses a low-quality version of the model itself to guide the high-quality model in the diffusion model, thereby improving image quality while maintaining diversity.
Guiding Neural Collapse: Optimising Towards the Nearest Simplex Equiangular Tight Frame
Evan Markou (Australian National University), Stephen Gould (Australian National University)
ClassificationOptimizationImage
🎯 What it does: This paper proposes a strategy for dynamically finding the nearest simplex equiangular tight frame (ETF) during the training process and using it as classifier weights, utilizing Riemannian optimization and deep declarative nodes for end-to-end training;
GVKF: Gaussian Voxel Kernel Functions for Highly Efficient Surface Reconstruction in Open Scenes
Gaochao Song (Hong Kong University of Science and Technology), Hao Wang (Hong Kong University of Science and Technology)
Autonomous DrivingComputational EfficiencyGaussian SplattingPoint CloudMesh
🎯 What it does: Proposes the Gaussian Voxel Kernel Functions (GVKF) method for efficient and high-quality 3D surface reconstruction in open scenes.
HairDiffusion: Vivid Multi-Colored Hair Editing via Latent Diffusion
Yu Zeng (Shenzhen University), Jinbao Wang (Shenzhen University)
GenerationData SynthesisDiffusion modelImage
🎯 What it does: Proposes HairDiffusion based on the latent diffusion model, supporting text, reference images, and stroke input, enabling independent editing of colorful hair.
HairFastGAN: Realistic and Robust Hair Transfer with a Fast Encoder-Based Approach
Maxim Nikolaev (Higher School of Economics), Aibek Alanov (Higher School of Economics)
Image TranslationGenerationGenerative Adversarial NetworkImage
🎯 What it does: A high-speed, high-quality hairstyle transfer method named HairFastGAN is proposed, which can perform hairstyle shape and color transfer on facial photos under near real-time conditions while preserving identity and background details.
Hallo3D: Multi-Modal Hallucination Detection and Mitigation for Consistent 3D Content Generation
Hongbo Wang (Institute of Automation Chinese Academy of Sciences), Ran He (Institute of Automation Chinese Academy of Sciences)
GenerationData SynthesisDiffusion modelScore-based ModelGaussian SplattingMultimodalityPoint Cloud
🎯 What it does: A parameter-free 3D content generation method called Hallo3D is proposed, which combines large multimodal models to detect and correct perspective inconsistencies and hallucinations produced by 2D diffusion models.
HaloScope: Harnessing Unlabeled LLM Generations for Hallucination Detection
Xuefeng Du (University of Wisconsin), Yixuan Li (University of Wisconsin)
ClassificationTransformerLarge Language ModelText
🎯 What it does: Proposes the HaloScope framework, which utilizes unlabeled LLM-generated text that naturally occurs in the real world to estimate the authenticity of samples through subspace decomposition of internal activations, and trains a binary classifier for authenticity detection based on this.
Hamba: Single-view 3D Hand Reconstruction with Graph-guided Bi-Scanning Mamba
Haoye Dong (Carnegie Mellon University), Fernando De la Torre (Carnegie Mellon University)
Pose EstimationGraph Neural NetworkTransformerImageMesh
🎯 What it does: A graph-guided bidirectional scanning framework based on Mamba (Hamba) is proposed for 3D hand mesh reconstruction from single-view RGB images.
Hamiltonian Monte Carlo Inference of Marginalized Linear Mixed-Effects Models
Jinlin Lai (University of Massachusetts), Daniel Sheldon (University of Massachusetts)
Computational EfficiencyTabular
🎯 What it does: This paper proposes an efficient random effects marginalization algorithm for linear mixed effects models (LMM), significantly improving the sampling efficiency of Hamiltonian Monte Carlo (HMC).
Hamiltonian Monte Carlo on ReLU Neural Networks is Inefficient
Vu C. Dinh (University of Delaware), Cuong V. Nguyen (Durham University)
ImageTabular
🎯 What it does: The paper analyzes the error rate of Hamiltonian Monte Carlo (HMC) using the leapfrog integrator on ReLU neural networks and demonstrates its inefficiency.
Hamiltonian Score Matching and Generative Flows
Peter Holderrieth (Massachusetts Institute of Technology), Tommi Jaakkola (Massachusetts Institute of Technology)
GenerationData SynthesisDiffusion modelScore-based ModelFlow-based ModelImageOrdinary Differential Equation
🎯 What it does: This paper proposes a Hamiltonian Velocity Predictor (HVP) and Hamiltonian Score Matching (HSM), and constructs a Hamiltonian Generative Flow (HGF) model based on these. It further introduces an oscillatory HGF using a harmonic oscillator force field; score estimation and generation are achieved through Neural ODE and minimizing velocity prediction errors. Experimental results validate the consistency of the score matching metric with traditional explicit score matching, achieving competitive generative quality with diffusion models and flow matching models on image datasets such as CIFAR-10 and FFHQ. Additionally, it outperforms traditional diffusion models at high resolutions but is slightly inferior to fine-tuned EDM.
Handling Learnwares from Heterogeneous Feature Spaces with Explicit Label Exploitation
Peng Tan (Nanjing University), Zhi-Hua Zhou (Nanjing University)
ClassificationDomain AdaptationContrastive LearningTabular
🎯 What it does: This paper studies how to utilize trained models (learnware) from different feature spaces on the learnware platform to assist users in completing tasks, specifically addressing the issues of model transfer and reuse under heterogeneous feature spaces.
Happy: A Debiased Learning Framework for Continual Generalized Category Discovery
Shijie Ma (Institute of Automation, Chinese Academy of Sciences), Cheng-Lin Liu (Institute of Automation, Chinese Academy of Sciences)
ClassificationKnowledge DistillationTransformerContrastive LearningImage
🎯 What it does: This paper proposes a continuous general category discovery (C-GCD) task without sample replay, long-term multi-stage, and unknown old-new category ratios, and provides a complete learning framework.