ICLR 2024 Papers — Page 9
International Conference on Learning Representations · 2260 papers
GAFormer: Enhancing Timeseries Transformers Through Group-Aware Embeddings
Jingyun Xiao (Georgia Institute of Technology), Eva L Dyer
ClassificationTransformerTime Series
🎯 What it does: This paper proposes a Group Embedding method and constructs the GAFormer Transformer for feature extraction of multivariate time series data.
GAIA: a benchmark for General AI Assistants
Grégoire Mialon (Meta AI), Thomas Scialom (Meta AI)
Large Language ModelPrompt EngineeringTextMultimodalityBenchmark
🎯 What it does: Designed and released the GAIA benchmark, which includes 466 real-world question-and-answer tasks for evaluating the reasoning, multimodal processing, and tool usage capabilities of general AI assistants.
GAIA: Zero-shot Talking Avatar Generation
Tianyu He (Microsoft), Jiang Bian (Microsoft)
GenerationData SynthesisDiffusion modelAuto EncoderVideoAudio
🎯 What it does: This paper studies a zero-shot talking head generation model GAIA, which decouples motion and appearance using VAE and utilizes a diffusion model to predict motion based on speech, thereby generating natural and diverse videos.
Gaining Wisdom from Setbacks: Aligning Large Language Models via Mistake Analysis
Kai Chen (Hong Kong University of Science and Technology), Lifeng Shang (Huawei Noah's Ark Lab)
Safty and PrivacyReinforcement Learning from Human FeedbackTransformerLarge Language ModelSupervised Fine-TuningReinforcement LearningText
🎯 What it does: A framework based on error analysis is proposed, utilizing the erroneous responses generated by the model itself for self-alignment.
Game-Theoretic Robust Reinforcement Learning Handles Temporally-Coupled Perturbations
Yongyuan Liang (University of Maryland), Stephen Marcus McAleer
Robotic IntelligenceReinforcement LearningSequential
🎯 What it does: This paper proposes a time-coupled adversarial perturbation model and designs the GRAD method based on a two-player zero-sum game to achieve robust reinforcement learning.
Gen-Z: Generative Zero-Shot Text Classification with Contextualized Label Descriptions
Sachin Kumar (Allen Institute for AI), Yulia Tsvetkov (University of Washington)
ClassificationTransformerLarge Language ModelPrompt EngineeringText
🎯 What it does: This paper proposes GEN-Z, a zero-shot generative text classification framework that utilizes the generative probabilities of language models combined with contextualized label descriptions for text classification.
GenCorres: Consistent Shape Matching via Coupled Implicit-Explicit Shape Generative Models
Haitao Yang (University of Texas at Austin), Qixing Huang (University of Texas at Austin)
GenerationData SynthesisAuto EncoderMesh
🎯 What it does: We propose GenCorres, a method for unsupervised joint shape matching using implicit-explicit shape generators.
Gene Regulatory Network Inference in the Presence of Dropouts: a Causal View
Haoyue Dai (Carnegie Mellon University), Kun Zhang (Mohamed bin Zayed University of Artificial Intelligence)
Biomedical Data
🎯 What it does: To address the bias in gene regulatory network (GRN) inference caused by technical zeros (dropout) in single-cell RNA-seq, a 'Causal Dropout Model' based on causal graphs is proposed. This model deletes samples with all conditional variables set to zero during conditional independence (CI) testing (test-wise deletion), thereby restoring the same CI relationships as in the absence of dropout, ultimately achieving unbiased GRN structure learning.
GeneOH Diffusion: Towards Generalizable Hand-Object Interaction Denoising via Denoising Diffusion
Xueyi Liu (Tsinghua University), Li Yi (Tsinghua University)
RestorationPose EstimationDomain AdaptationDiffusion modelVideo
🎯 What it does: A general hand-object interaction noise removal method called GeneOH Diffusion is proposed, which can recover the true hand trajectory under different objects, actions, and noise patterns.
General Graph Random Features
Isaac Reid (University of Cambridge), Adrian Weller (University of Cambridge)
Graph Neural NetworkGraphOrdinary Differential Equation
🎯 What it does: A general graph random feature (g-GRF) algorithm based on random walks is proposed for unbiased estimation of arbitrary weighted adjacency matrix functions, thereby efficiently approximating the graph kernel matrix.
General Stability Analysis for Zeroth-Order Optimization Algorithms
Xinyue Liu (Huazhong Agricultural University), Hong Chen (Huazhong Agricultural University)
OptimizationTabular
🎯 What it does: A general stability analysis framework is proposed to derive the generalization error upper bounds for zero-order optimization algorithms (such as ZO-SGD, ZO-GD, ZO-SVRG), and specific generalization bounds for different gradient estimation methods (1-point, 2-point, coordinate estimation) are provided in convex, strongly convex, and non-convex scenarios.
Generalization error of spectral algorithms
Maksim Velikanov (Technology Innovation Institute), Dmitry Yarotsky (Skolkovo Institute of Science and Technology)
Tabular
🎯 What it does: This paper derives a general error upper bound for kernel methods under two data models through a spectral algorithm framework, providing a complete asymptotic error expansion.
Generalization in diffusion models arises from geometry-adaptive harmonic representations
Zahra Kadkhodaie (New York University), Stéphane Mallat (Flatiron Institute)
RestorationGenerationConvolutional Neural NetworkDiffusion modelImage
🎯 What it does: This study investigates the generalization ability of diffusion models with sufficiently large training sets and reveals the denoising mechanism induced by the internal Geometric Adaptive Harmonic Basis (GAHB).
Generalization of Scaled Deep ResNets in the Mean-Field Regime
Yihang Chen (École Polytechnique Fédérale de Lausanne), Volkan Cevher (École Polytechnique Fédérale de Lausanne)
ImageOrdinary Differential Equation
🎯 What it does: This study investigates the generalization properties of deep residual networks (Scaled ResNet) in the mean-field regime under the infinite depth and width limit, providing lower bounds on the minimum eigenvalue of the Gram matrix, control of KL divergence, and upper bounds on Rademacher complexity generalization.
Generalized Neural Sorting Networks with Error-Free Differentiable Swap Functions
Jungtaek Kim (University of Pittsburgh), Minsu Cho (POSTECH)
TransformerImage
🎯 What it does: This paper proposes an error-free differentiable swap function (error-free DSF) and a permutation-based neural sorting network using multi-head attention, aimed at mapping high-dimensional inputs to ordinals and achieving sorting.
Generalized Policy Iteration using Tensor Approximation for Hybrid Control
Suhan Shetty (Idiap Research Institute), Sylvain Calinon (Idiap Research Institute)
OptimizationReinforcement LearningSequential
🎯 What it does: An approximate dynamic programming algorithm based on Tensor Train, called TTPI, is proposed to address optimal control problems with mixed continuous and discrete action spaces.
Generalized Schrödinger Bridge Matching
Guan-Horng Liu (Georgia Institute of Technology), Ricky T. Q. Chen (Meta)
Image TranslationOptimizationDiffusion modelImageStochastic Differential Equation
🎯 What it does: Proposes the Generalized Schrödinger Bridge Matching (GSBM) algorithm, which learns diffusion models that satisfy the distributions at both ends under given task-specific state costs; reformulates the distribution matching problem as conditional stochastic optimal control and provides a scalable solution process.
Generating Images with 3D Annotations Using Diffusion Models
Wufei Ma (Johns Hopkins University), Alan Yuille (Johns Hopkins University)
Object DetectionGenerationData SynthesisPose EstimationLarge Language ModelDiffusion modelImagePoint Cloud
🎯 What it does: Utilizing ControlNet to incorporate 3D geometry-based visual prompts (edge maps) and diverse text prompts generated by large language models into diffusion models, automatically generating multi-view, high-quality images with real 3D annotations for data augmentation.
Generating Pragmatic Examples to Train Neural Program Synthesizers
Saujas Vaduguru (Carnegie Mellon University), Yewen Pu (Autodesk Research)
GenerationData SynthesisAI Code AssistantTransformerLarge Language ModelSupervised Fine-TuningText
🎯 What it does: An iterative data generation and fine-tuning framework called PRAX, based on self-dialogue (speaker-listener) and Rational Speech Acts (RSA) reasoning, has been constructed to train a neural program synthesizer, enabling it to better recognize target regular expressions given only examples.
Generative Adversarial Equilibrium Solvers
Denizalp Goktas (Brown University), Andrea Tacchetti (Google DeepMind)
GenerationOptimizationGenerative Adversarial Network
🎯 What it does: This paper proposes a Generative Adversarial Equilibrium Solver (GAES) based on Generative Adversarial Networks (GANs) to learn Generalized Nash Equilibria (GNE) and Arrow-Debreu Competitive Equilibria (CE) from pseudo-games, enabling rapid inference of equilibrium solutions on new pseudo-games after a single training session.
Generative Human Motion Stylization in Latent Space
chuan guo, Li Cheng (University of Alberta)
GenerationData SynthesisConvolutional Neural NetworkAuto EncoderVideo
🎯 What it does: A generative human motion stylization framework based on a pre-trained autoencoder latent space is proposed, capable of achieving diverse stylization through random sampling based on reference actions, labels, or unconditional priors.
Generative Judge for Evaluating Alignment
Junlong Li (Shanghai Jiao Tong University), Pengfei Liu (Shanghai Jiao Tong University)
GenerationExplainability and InterpretabilityReinforcement Learning from Human FeedbackTransformerLarge Language ModelSupervised Fine-TuningText
🎯 What it does: A generative evaluator AUTO-J with 13 billion parameters is proposed to assess the alignment performance of LLMs in multiple scenarios.
Generative Learning for Financial Time Series with Irregular and Scale-Invariant Patterns
Hongbin Huang (City University of Hong Kong), Xiao Qiao (City University of Hong Kong)
GenerationData SynthesisRecommendation SystemAnomaly DetectionOptimizationDiffusion modelTime SeriesFinance Related
🎯 What it does: Designed and implemented FTS-Diffusion, a three-stage generative framework (pattern recognition, pattern generation, pattern evolution) for synthesizing financial time series with irregular and scale-invariant features.
Generative Learning for Solving Non-Convex Problem with Multi-Valued Input-Solution Mapping
Enming Liang (City University of Hong Kong), Minghua Chen (City University of Hong Kong)
OptimizationFlow-based ModelRectified Flow
🎯 What it does: A generative learning framework is proposed, which transforms the multi-valued input-solution mapping into a mapping from input to solution distribution, utilizing Rectified Flow (RectFlow) to model the multi-solution distribution of non-convex optimization problems and generate high-quality solutions.
Generative Modeling of Regular and Irregular Time Series Data via Koopman VAEs
Ilan Naiman (Ben-Gurion University), Omri Azencot (Ben-Gurion University)
GenerationData SynthesisExplainability and InterpretabilityAuto EncoderTime SeriesSequentialFinance RelatedPhysics Related
🎯 What it does: For the tasks of generating regular and irregular time series, a Koopman VAE (KoVAE) framework is proposed, which incorporates a linear latent state prior based on Koopman theory into the VAE structure, achieving high-quality generation and interpretability.
Generative Modeling with Phase Stochastic Bridge
Tianrong Chen (Georgia Tech), Shuangfei Zhai (Apple)
GenerationData SynthesisImageMultimodalityStochastic Differential EquationOrdinary Differential Equation
🎯 What it does: An Accelerated Generative Model (AGM) based on phase space dynamics is proposed, which constructs a straighter bridge matching trajectory through SOC and utilizes velocity information to achieve early data prediction, significantly reducing the number of sampling steps.
Generative Pre-training for Speech with Flow Matching
Alexander H. Liu (Massachusetts Institute of Technology), Wei-Ning Hsu (Meta AI)
RestorationGenerationTransformerFlow-based ModelAudio
🎯 What it does: This paper proposes a general speech generation pre-training model named SpeechFlow and demonstrates its transfer performance in tasks such as denoising, speech separation, and zero-shot TTS.
Generative Sliced MMD Flows with Riesz Kernels
Johannes Hertrich (University College London), Paul Hagemann (Technische Universität Berlin)
GenerationData SynthesisConvolutional Neural NetworkFlow-based ModelGenerative Adversarial NetworkImage
🎯 What it does: Using the equivalence of the MMD of the Riesz kernel and its sliced version, a sorting-based 1D gradient computation algorithm is proposed, and this efficient gradient is utilized to implement MMD flow, which is then approximated by a neural network to generate images.
GENOME: Generative Neuro-Symbolic Visual Reasoning by Growing and Reusing Modules
Zhenfang Chen (MIT IBM Watson AI Lab), Chuang Gan (MIT IBM Watson AI Lab)
RecognitionGenerationTransformerLarge Language ModelPrompt EngineeringImage
🎯 What it does: A generative neural symbolic visual reasoning framework named GENOME is proposed, which utilizes large language models to dynamically generate, validate, and store new visual modules, thereby achieving cross-task reasoning and module reuse with a small number of examples.
GenSim: Generating Robotic Simulation Tasks via Large Language Models
Lirui Wang (Massachusetts Institute of Technology), Xiaolong Wang (University of California San Diego)
Robotic IntelligenceTransformerLarge Language ModelPrompt EngineeringTextRetrieval-Augmented Generation
🎯 What it does: Utilizing large language models (GPT-4, Code-Llama, etc.) to automatically generate rich robotic simulation tasks and expert demonstrations, and training multi-task visual-language control strategies based on these tasks to enhance the robot's generalization ability in new tasks.
GeoDiffusion: Text-Prompted Geometric Control for Object Detection Data Generation
Kai Chen (Hong Kong University of Science and Technology), Dit-Yan Yeung (Hong Kong University of Science and Technology)
Object DetectionGenerationData SynthesisAutonomous DrivingPrompt EngineeringDiffusion modelImage
🎯 What it does: The GEODIFFUSION framework is proposed, which maps various geometric conditions (such as bounding boxes and camera angles) to a pre-trained text-to-image diffusion model using text prompts to generate high-quality object detection training data.
Geographic Location Encoding with Spherical Harmonics and Sinusoidal Representation Networks
Marc Rußwurm (Wageningen University), Devis Tuia (École Polytechnique Fédérale de Lausanne)
ClassificationComputational EfficiencyRepresentation LearningImage
🎯 What it does: This paper proposes a geographic location encoding method that combines Spherical Harmonics (SH) with Sinusoidal Representation Networks (SIREN) to efficiently learn representations of geographic coordinates on a global scale.
GeoLLM: Extracting Geospatial Knowledge from Large Language Models
Rohin Manvi (Stanford University), Stefano Ermon (Stanford University)
TransformerLarge Language ModelSupervised Fine-TuningPrompt EngineeringTextMultimodality
🎯 What it does: The GeoLLM method is proposed, which extracts the geographical knowledge contained in large language models by fine-tuning prompts that include supplementary map data from OpenStreetMap;
Geometrically Aligned Transfer Encoder for Inductive Transfer in Regression Tasks
Sung Moon Ko (LG AI Research), Sehui Han (LG AI Research)
Domain AdaptationDrug DiscoveryAuto EncoderTabularBiomedical Data
🎯 What it does: This paper proposes the Geometrically Aligned Transfer Encoder (GATE), a transfer learning framework based on Riemannian geometry, aimed at addressing the source-target transfer problem in molecular property regression tasks.
Geometry-Aware Projective Mapping for Unbounded Neural Radiance Fields
Junoh Lee (Gwangju Institute of Science and Technology), Hae-Gon Jeon (Gwangju Institute of Science and Technology)
Neural Radiance FieldPoint Cloud
🎯 What it does: An adaptive p-norm projection mapping and angle ray parameterization is proposed, which can more effectively represent unbounded scenes in neural radiance fields and integrate it into various NeRF frameworks for unbounded view synthesis.
Get more for less: Principled Data Selection for Warming Up Fine-Tuning in LLMs
Feiyang Kang (Virginia Tech), Ruoxi Jia (Virginia Tech)
Domain AdaptationComputational EfficiencyTransformerLarge Language ModelSupervised Fine-TuningText
🎯 What it does: A pre-fine-tuning scheme is proposed, utilizing a large amount of unlabeled open data to select samples through the GOT-D method for lightweight preheating of pre-trained language models, followed by final fine-tuning on a small amount of labeled data, thereby improving task performance and reducing costs.
Get What You Want, Not What You Don't: Image Content Suppression for Text-to-Image Diffusion Models
Senmao Li (Nankai University), jian Yang
GenerationData SynthesisOptimizationDiffusion modelImage
🎯 What it does: To address the issue of text-to-image diffusion models struggling to suppress negative targets, a method is proposed to remove unwanted content through soft-weighted regularization and optimization of text embeddings during inference.
Ghost on the Shell: An Expressive Representation of General 3D Shapes
Zhen Liu (Mila - Quebec AI Institute), Bernhard Schölkopf (Max Planck Institute for Intelligent Systems)
GenerationData SynthesisDiffusion modelMesh
🎯 What it does: A representation method called G-SHELL is proposed, which can simultaneously represent both closed and non-closed 3D meshes, and achieves multi-view reconstruction and unconditional generation.
GIM: Learning Generalizable Image Matcher From Internet Videos
Xuelun Shen (Xiamen University), Cheng Wang (Xiamen University)
RetrievalDomain AdaptationImageVideoBenchmark
🎯 What it does: Through the self-training framework GIM, an image matching model capable of generalizing across different scenes is trained using internet videos, and a zero-shot evaluation benchmark ZEB is proposed.
GIO: Gradient Information Optimization for Training Dataset Selection
Dante Everaert (Amazon), Christopher Potts (Stanford University)
OptimizationData-Centric LearningImageText
🎯 What it does: Proposes the Gradient Information Optimization (GIO) method, which selects a subset of training data by minimizing KL divergence in an unlabeled setting;
Global Optimality for Non-linear Constrained Restoration Problems via Invexity
Samuel Pinilla (Science and Technology Facilities Council), Jeyan Thiyagalingam (Science and Technology Facilities Council)
RestorationOptimizationImageComputed Tomography
🎯 What it does: This paper studies a family of invex/quasi-invex functions, constructs a model that guarantees global optimality for non-convex signal recovery problems, and provides theoretical proofs and algorithm implementations.
GlucoBench: Curated List of Continuous Glucose Monitoring Datasets with Prediction Benchmarks
Renat Sergazinov (Texas A&M University), Irina Gaynanova (Texas A&M University)
OptimizationData-Centric LearningTransformerAuto EncoderTime SeriesBiomedical DataBenchmark
🎯 What it does: This paper organizes and publicly releases five open continuous glucose monitoring (CGM) datasets, proposes standard prediction tasks (accuracy prediction and uncertainty assessment), and provides benchmark results for various baseline models.
GNeRP: Gaussian-guided Neural Reconstruction of Reflective Objects with Noisy Polarization Priors
LI Yang, Ying-Cong Chen
RestorationNeural Radiance FieldPoint Cloud
🎯 What it does: This paper proposes a method for high-precision 3D reconstruction of reflective objects using Gaussian distribution for normal vector representation, combined with polarization information.
GNNBoundary: Towards Explaining Graph Neural Networks through the Lens of Decision Boundaries
Xiaoqi Wang (Ohio State University), Han Wei Shen
Explainability and InterpretabilityGraph Neural NetworkGraph
🎯 What it does: This paper proposes GNNBoundary, which provides model-level explanations for graph neural networks from the perspective of decision boundaries.
GNNCert: Deterministic Certification of Graph Neural Networks against Adversarial Perturbations
zaishuo xia, Jinyuan Jia (Pennsylvania State University)
ClassificationAdversarial AttackGraph Neural NetworkGraph
🎯 What it does: This paper proposes GNNCert, a verifiable robustness defense for graph classification tasks that ensures the prediction label remains unchanged when the structure and node features are perturbed a limited number of times.
GNNX-BENCH: Unravelling the Utility of Perturbation-based GNN Explainers through In-depth Benchmarking
Mert Kosan (University of California), Sayan Ranu (Indian Institute of Technology)
Explainability and InterpretabilityGraph Neural NetworkSupervised Fine-TuningGraphBenchmark
🎯 What it does: Systematically benchmark the perturbation-based interpreters of GNNs (including factual and counterfactual methods), exploring their interpretability, stability, reproducibility, and feasibility.
GOAt: Explaining Graph Neural Networks via Graph Output Attribution
Shengyao Lu (University of Alberta), Di Niu (University of Alberta)
Explainability and InterpretabilityGraph Neural NetworkGraph
🎯 What it does: The Graph Output Attribution (GOAt) method is proposed for local interpretation of pre-trained Graph Neural Networks (GNNs), decomposing model outputs into scalar products and attributing them based on input node/edge features.
Going Beyond Neural Network Feature Similarity: The Network Feature Complexity and Its Interpretation Using Category Theory
Yiting Chen (Shanghai Jiao Tong University), Junchi Yan (Shanghai Jiao Tong University)
CompressionOptimizationConvolutional Neural NetworkImage
🎯 What it does: Proposes a definition of functionally equivalent features, explains the complexity of network features using category theory, and based on this, introduces the Iterative Feature Merging (IFM) algorithm to evaluate and compress networks;
GoLLIE: Annotation Guidelines improve Zero-Shot Information-Extraction
Oscar Sainz (University of the Basque Country), Eneko Agirre (University of the Basque Country)
TransformerLarge Language ModelSupervised Fine-TuningPrompt EngineeringText
🎯 What it does: Introducing GoLLIE, a large language model (LLM) designed for information extraction tasks, specifically for zero-shot learning through refined annotation guidelines;
Goodhart's Law in Reinforcement Learning
Jacek Karwowski (University of Oxford), Joar Max Viktor Skalse (Future of Humanity Institute)
Reinforcement LearningTabular
🎯 What it does: This paper conducts a quantitative, empirical, and theoretical analysis of the Goodhart effect caused by mis-specification of proxy rewards, and proposes an early stopping strategy to maximize true rewards while ensuring that the worst-case scenario does not experience Goodhart.
GPAvatar: Generalizable and Precise Head Avatar from Image(s)
Xuangeng Chu (University of Tokyo), Tatsuya Harada (University of Tokyo)
GenerationPose EstimationGenerative Adversarial NetworkImageVideoPoint Cloud
🎯 What it does: The GPAvatar framework is proposed, capable of reconstructing an animatable 3D head avatar from one or more images in a single forward inference, achieving precise control over expressions and poses.
GPT-4 Is Too Smart To Be Safe: Stealthy Chat with LLMs via Cipher
Youliang Yuan (Chinese University of Hong Kong), Zhaopeng Tu (Tencent AI Lab)
Safty and PrivacyAdversarial AttackTransformerLarge Language ModelPrompt EngineeringText
🎯 What it does: The study verifies whether existing safety alignment mechanisms can be bypassed under non-natural language input by using various ciphers in dialogue with large language models (LLMs), and proposes a systematic evaluation framework called CipherChat.
Gradual Domain Adaptation via Gradient Flow
Zhan Zhuang (Southern University of Science and Technology), Ying Wei (Nanyang Technological University)
Domain AdaptationFlow-based ModelImage
🎯 What it does: Generate continuous intermediate domains through gradient flow and gradually fine-tune the source classifier to achieve unsupervised domain adaptation.
Gradual Optimization Learning for Conformational Energy Minimization
Artem Tsypin (AIRI), Artur Kadurin (AIRI)
OptimizationDrug DiscoveryGraph Neural NetworkTabular
🎯 What it does: Using neural networks (NNP) to learn and predict the potential energy of molecular conformations, and employing its gradients for energy minimization, improves traditional physics-based conformational optimization.
GRANDE: Gradient-Based Decision Tree Ensembles for Tabular Data
Sascha Marton (University of Mannheim), Heiner Stuckenschmidt (University of Mannheim)
ClassificationOptimizationHyperparameter SearchTabular
🎯 What it does: A gradient descent-based hard-axis alignment decision tree ensemble method called GRANDE is proposed, and end-to-end training is achieved using dense representation and straight-through operations;
Graph Generation with $K^2$-trees
Yunhui Jang (Pohang University of Science and Technology), Sungsoo Ahn (Pohang University of Science and Technology)
GenerationData SynthesisTransformerGraph
🎯 What it does: A graph generation framework HGGT based on K²-tree is proposed, which utilizes the hierarchical and compression characteristics of K²-tree to represent graphs as sequences and generate them automatically using a Transformer.
Graph Lottery Ticket Automated
Guibin Zhang (Hong Kong University of Science and Technology), Yuxuan Liang (Tongji University)
CompressionComputational EfficiencyGraph Neural NetworkGraph
🎯 What it does: This paper proposes AdaGLT, an automatic, dynamic, layer-adaptive Graph Lottery Ticket search framework for simultaneously sparsifying weights and graph structures in large-scale deep graph neural networks, balancing model compression and inference efficiency.
Graph Metanetworks for Processing Diverse Neural Architectures
Derek Lim (Massachusetts Institute of Technology), James Lucas (NVIDIA)
Graph Neural NetworkImage
🎯 What it does: This study investigates how to use graph neural networks to handle the parameters of other neural networks and proposes the Graph Meta Network (GMN) framework.
Graph Neural Networks for Learning Equivariant Representations of Neural Networks
Miltiadis Kofinas (University of Amsterdam), David W. Zhang (University of Amsterdam)
OptimizationRepresentation LearningGraph Neural NetworkTransformerReinforcement LearningImage
🎯 What it does: Unified encoding of neural network parameters and architectures into a graph structure (neural graph), and using graph neural networks/Transformers for invariant/homogeneous learning, thereby achieving generalization for networks of different architectures;
Graph Parsing Networks
Yunchong Song (Shanghai Jiao Tong University), Zhouhan Lin (Shanghai Jiao Tong University)
Computational EfficiencyGraph Neural NetworkGraph
🎯 What it does: This paper proposes an adaptive graph pooling framework based on graph parsing (Graph Parsing Network, GPN), which can learn personalized pooling trees for each graph and achieve efficient node aggregation.
Graph Transformers on EHRs: Better Representation Improves Downstream Performance
Raphael Poulain (University of Delaware), Rahmatollah Beheshti (University of Delaware)
ClassificationRepresentation LearningDrug DiscoveryGraph Neural NetworkTransformerTabularTime SeriesSequentialBiomedical DataElectronic Health Records
🎯 What it does: This paper proposes a hybrid model named GT-BEHRT, which utilizes a graph Transformer to generate embeddings for each visit, and then captures the temporal relationships in the patient's visit sequence through a BERT encoder, resulting in more robust patient representations for various prediction tasks.
Graph-based Virtual Sensing from Sparse and Partial Multivariate Observations
Giovanni De Felice (University of Liverpool), Cesare Alippi (Swiss AI Lab IDSIA and Universita della Svizzera italiana)
Graph Neural NetworkTime Series
🎯 What it does: A graph-based virtual sensing method is proposed for scenarios with sparse and partially observed multivariate data, utilizing the dependencies between multivariates and across different locations to recover the spatiotemporal sequences of unobserved channels.
GRAPH-CONSTRAINED DIFFUSION FOR END-TO-END PATH PLANNING
Dingyuan Shi (Beihang University), Jieping Ye (University of Michigan)
GenerationAutonomous DrivingOptimizationGraph Neural NetworkDiffusion modelGraphTime Series
🎯 What it does: An end-to-end path planning method based on diffusion models (GDP) is proposed, which can directly generate paths that conform to actual road network constraints given a starting point and an endpoint.
GraphCare: Enhancing Healthcare Predictions with Personalized Knowledge Graphs
Pengcheng Jiang (University of Illinois Urbana-Champaign), Jimeng Sun (University of Illinois Urbana-Champaign)
Graph Neural NetworkLarge Language ModelTabularBiomedical DataElectronic Health Records
🎯 What it does: Construct a personalized knowledge graph and use a dual attention-enhanced graph neural network for multiple medical predictions for patients.
GraphChef: Decision-Tree Recipes to Explain Graph Neural Networks
Peter Müller (ETH Zurich), Roger Wattenhofer (ETH Zurich)
Explainability and InterpretabilityKnowledge DistillationGraph Neural NetworkGraph
🎯 What it does: An interpretable graph neural network called GraphChef is proposed, which generates global dataset explanation rules (recipes) using decision trees, rather than just local graph explanations.
Graphical Multioutput Gaussian Process with Attention
Yijue Dai (Chinese University of Hong Kong), Feng Yin (Chinese University of Hong Kong)
Graph Neural NetworkTabularTime SeriesElectrocardiogram
🎯 What it does: A graphical multi-output Gaussian process (GMOGP) framework is proposed, which dynamically learns the conditional dependencies between outputs through probabilistic graphical models and attention mechanisms, supports non-Gaussian priors, and achieves interpretable output associations.
GraphPulse: Topological representations for temporal graph property prediction
Kiarash Shamsi (University of Manitoba), Cuneyt Gurcan Akcora (University of Central Florida)
Recurrent Neural NetworkGraph Neural NetworkGraphTime Series
🎯 What it does: A framework named GraphPulse is proposed for predicting future attributes of time-evolving graphs, combining Mapper to generate topological summaries and using RNN for sequence modeling.
Grokking as a First Order Phase Transition in Two Layer Networks
Noa Rubin (Hebrew University), Zohar Ringel (Hebrew University)
Physics Related
🎯 What it does: This study explores the phenomenon of 'Grokking' that occurs when deep networks learn new features, modeling it as a first-order phase transition in a two-layer network.
Grokking as the transition from lazy to rich training dynamics
Tanishq Kumar (Harvard University), Cengiz Pehlevan (Harvard University)
TransformerImageTabular
🎯 What it does: Through theoretical analysis and experiments, it is demonstrated that the phenomenon of 'grokking' is caused by the transition from 'lazy' training to 'rich' feature learning, and this mechanism is validated across various networks and datasets;
Grokking in Linear Estimators -- A Solvable Model that Groks without Understanding
Noam Itzhak Levi (Raymond and Beverly Sackler School of Physics and Astronomy), Yohai Bar-Sinai (Raymond and Beverly Sackler School of Physics and Astronomy)
Tabular
🎯 What it does: Analyzed and proved the phenomenon of 'grokking' in the high-dimensional linear teacher-student model, where training error decreases first and then test error significantly increases;
GROOT: Learning to Follow Instructions by Watching Gameplay Videos
Shaofei Cai (Peking University), Yitao Liang (Peking University)
Robotic IntelligenceTransformerAuto EncoderVideoBenchmark
🎯 What it does: Proposed the GROOT framework, which learns the target space and low-level controllers from watched game videos to achieve instruction following on the Minecraft SkillForge benchmark.
Ground-A-Video: Zero-shot Grounded Video Editing using Text-to-image Diffusion Models
Hyeonho Jeong (Korea Advanced Institute of Science and Technology), Jong Chul Ye (Korea Advanced Institute of Science and Technology)
GenerationData SynthesisPose EstimationTransformerDiffusion modelOptical FlowVideoText
🎯 What it does: This paper proposes Ground-A-Video, which utilizes grounding guidance to achieve zero-training multi-attribute video editing, style transfer, and pose-guided text-to-video generation while maintaining temporal consistency.
Grounded Object-Centric Learning
Avinash Kori (Imperial College London), Ben Glocker (Imperial College London)
Object DetectionRepresentation LearningRecurrent Neural NetworkAuto EncoderImage
🎯 What it does: An unsupervised conditional slot attention mechanism (COSA) is proposed, combined with a grounded slot dictionary (GSD) to learn object-centered representations, supporting dynamic slot number estimation and object type binding.
Grounding Language Plans in Demonstrations Through Counterfactual Perturbations
Yanwei Wang (Massachusetts Institute of Technology), Julie Shah (Massachusetts Institute of Technology)
OptimizationExplainability and InterpretabilityRobotic IntelligenceConvolutional Neural NetworkLarge Language ModelReinforcement LearningPrompt EngineeringMultimodality
🎯 What it does: The GLiDE framework is proposed, which utilizes large language models to extract patterns from multi-step demonstrations and generates positive and negative samples through synthetic perturbations, training a differentiable state-to-pattern discriminator to achieve high-level planning and low-level control for physical domain tasks.
Grounding Multimodal Large Language Models to the World
Zhiliang Peng (University of Chinese Academy of Sciences), Furu Wei (Microsoft Research)
GenerationRetrievalTransformerLarge Language ModelVision Language ModelImageTextMultimodality
🎯 What it does: A multimodal large language model KOSMOS-2 with visual localization and text reference capabilities has been designed and trained. It can directly output the corresponding image area coordinates when generating text and supports users in pointing to objects using bounding boxes or referring expressions.
Group Preference Optimization: Few-Shot Alignment of Large Language Models
Siyan Zhao (University of California), Aditya Grover (University of California)
OptimizationMeta LearningTransformerLarge Language ModelReinforcement LearningText
🎯 What it does: This paper studies a few-shot alignment framework called Group Preference Optimization (GPO), which utilizes an independent Transformer module to learn and predict group or individual preferences in the LLM embedding space, allowing for rapid alignment of large language model outputs without the costly per-group fine-tuning.
GTA: A Geometry-Aware Attention Mechanism for Multi-View Transformers
Takeru Miyato (University of Tübingen), Andreas Geiger (University of Tübingen)
TransformerImage
🎯 What it does: Geometric Perception Attention (GTA) is introduced in the Transformer, applying relative geometric transformations to queries, keys, and values to achieve natural alignment in multi-view scenes.
GTMGC: Using Graph Transformer to Predict Molecule’s Ground-State Conformation
Guikun Xu (Southwest Jiaotong University), Jim Chen
Drug DiscoveryGraph Neural NetworkTransformerGraph
🎯 What it does: An end-to-end model GTMGC based on Graph Transformer is proposed, which directly predicts the 3D ground state conformation from the 2D topological structure of molecules.
Guaranteed Approximation Bounds for Mixed-Precision Neural Operators
Renbo Tu (University of Toronto), Anima Anandkumar (NVIDIA)
OptimizationComputational EfficiencyTabular
🎯 What it does: A mixed precision training method for neural operators (such as FNO) is proposed to reduce memory usage and improve throughput.
Guess & Sketch: Language Model Guided Transpilation
Celine Lee (Cornell University), Alexander M Rush
AI Code AssistantTransformerLarge Language ModelText
🎯 What it does: A framework for assembly code translation called GUESS & SKETCH, which combines neural networks and symbolic reasoning, is proposed to automatically convert assembly programs between ARMv8 and RISC-V.
Guiding Instruction-based Image Editing via Multimodal Large Language Models
Tsu-Jui Fu (University of California Santa Barbara), Zhe Gan (Apple)
Image TranslationGenerationTransformerLarge Language ModelDiffusion modelImageMultimodality
🎯 What it does: This paper proposes an image editing framework MGIE that utilizes a multimodal large language model to generate concise expression instructions and is jointly trained with a diffusion model, capable of transforming natural instructions into visual guidance for more precise image editing.
Guiding Masked Representation Learning to Capture Spatio-Temporal Relationship of Electrocardiogram
Yeongyeon Na (VUNO Inc), Sunghoon Joo (VUNO Inc)
Anomaly DetectionRepresentation LearningTransformerAuto EncoderTime SeriesBiomedical DataElectrocardiogram
🎯 What it does: Through the Masked Autoencoder (MAE) framework of self-supervised learning, we reconstruct 12-lead electrocardiograms (ECGs) using spatio-temporal patching to learn general ECG representations and fine-tune them for downstream disease screening tasks.
H-GAP: Humanoid Control with a Generalist Planner
zhengyao jiang, Yuandong Tian (AI at Meta)
Robotic IntelligenceTransformerReinforcement LearningAuto EncoderTime Series
🎯 What it does: Developed the Humanoid Generalist Autoencoding Planner (H-GAP), a general-purpose humanoid robot control model based on MoCapAct data, capable of completing various downstream control tasks in a zero-copy manner without online interaction through Model Predictive Control (MPC).
H2O-SDF: Two-phase Learning for 3D Indoor Reconstruction using Object Surface Fields
Minyoung Park (LG Electronics), Chul Lee (LG Electronics)
SegmentationGenerationNeural Radiance FieldPoint Cloud
🎯 What it does: A two-stage learning framework H O‑SDF is proposed for 3D indoor scene reconstruction; it first learns the overall geometry and then refines the surface details of objects.
Habitat 3.0: A Co-Habitat for Humans, Avatars, and Robots
Xavier Puig, Roozbeh Mottaghi (Meta)
Robotic IntelligenceReinforcement LearningMesh
🎯 What it does: The Habitat 3.0 platform is proposed to achieve rapid and realistic virtual human simulation and robot-human collaboration simulation, providing human-computer interaction evaluation tools and two types of collaborative tasks (social navigation and social rearrangement).
Harnessing Density Ratios for Online Reinforcement Learning
Philip Amortila (University of Illinois at Urbana-Champaign), Tengyang Xie (Microsoft Research)
Reinforcement Learning
🎯 What it does: This paper studies the use of density ratio modeling to achieve sample efficiency in online reinforcement learning, proposing two algorithms, GLOW and HYGLOW, and providing strict theoretical sample complexity and risk upper bounds.
Harnessing Explanations: LLM-to-LM Interpreter for Enhanced Text-Attributed Graph Representation Learning
Xiaoxin He (National University of Singapore), Bryan Hooi (National University of Singapore)
ClassificationExplainability and InterpretabilityComputational EfficiencyRepresentation LearningGraph Neural NetworkLarge Language ModelSupervised Fine-TuningPrompt EngineeringTextGraph
🎯 What it does: This paper proposes a TAPE framework that enriches the node features of Text Attribute Graphs (TAG) using explanation texts generated by large language models (LLM) and inputs them into Graph Neural Networks (GNN) for node classification.
Harnessing Joint Rain-/Detail-aware Representations to Eliminate Intricate Rains
Wu Ran (Fudan University), Hong Lu (Fudan University)
RestorationConvolutional Neural NetworkTransformerContrastive LearningImage
🎯 What it does: This paper proposes an adaptive image de-raining method called CoIC, which is based on joint rain/detail perception representation. It can train CNN/Transformer models on mixed multi-source datasets and achieve dynamic adaptation to different rain intensities/background details through instance-level modulation.
HAZARD Challenge: Embodied Decision Making in Dynamically Changing Environments
Qinhong Zhou (Institute for Interdisciplinary Information Sciences), Chuang Gan (Massachusetts Institute of Technology)
Robotic IntelligenceTransformerLarge Language ModelReinforcement LearningAgentic AIMultimodalityBenchmark
🎯 What it does: A dynamic disaster simulation benchmark named HAZARD has been designed and implemented, integrating three scenarios: fire, flood, and storm, allowing embedded agents to complete rescue tasks for target objects in a 3D environment.
Headless Language Models: Learning without Predicting with Contrastive Weight Tying
Nathan Godey (Inria), Benoît Sagot (Inria)
TransformerLarge Language ModelContrastive LearningText
🎯 What it does: A pre-training method without a language model projection head is proposed—Headless Language Modeling, which directly learns word embedding representations using Contrastive Weight Tying.
Hebbian Learning based Orthogonal Projection for Continual Learning of Spiking Neural Networks
Mingqing Xiao (Peking University), Zhouchen Lin (Peking University)
Spiking Neural NetworkImage
🎯 What it does: A Hebbian learning-based orthogonal projection method (HLOP) is proposed, which achieves online extraction of the neuronal activity subspace through lateral recurrent connections, thereby protecting old task knowledge during continual learning and avoiding catastrophic forgetting.
Heterogeneous Personalized Federated Learning by Local-Global Updates Mixing via Convergence Rate
Meirui Jiang (Chinese University of Hong Kong), Qi Dou (Chinese University of Hong Kong)
Federated LearningBiomedical Data
🎯 What it does: The LG-Mix method is proposed, which dynamically mixes local and global updates through the NTK convergence rate to achieve personalized federated learning under heterogeneous features.
Hiding in Plain Sight: Disguising Data Stealing Attacks in Federated Learning
Kostadin Garov (INSAIT), Martin Vechev (ETH Zurich)
Federated LearningSafty and PrivacyAdversarial AttackImage
🎯 What it does: This paper proposes an attack framework called SEER that utilizes malicious servers to steal data from large batches of gradients through a secret decoder in federated learning.
Hierarchical Context Merging: Better Long Context Understanding for Pre-trained LLMs
Woomin Song (KAIST), Jinwoo Shin (KAIST)
RetrievalComputational EfficiencyTransformerLarge Language ModelText
🎯 What it does: A training-independent hierarchical context merging (HOMER) scheme is proposed, which expands the context window of LLMs while maintaining computational efficiency by merging layer by layer and performing token pruning on each block before merging.
HIFA: High-fidelity Text-to-3D Generation with Advanced Diffusion Guidance
Junzhe Zhu (Stanford University), Sanmi Koyejo (Stanford University)
GenerationData SynthesisDiffusion modelScore-based ModelNeural Radiance FieldImageText
🎯 What it does: Using score distillation of pre-trained text-image diffusion models, conducted simultaneously in latent space and image space, combined with timestep annealing, z-variance regularization, and convolutional smoothing importance sampling, to achieve single-stage high-quality text-to-3D asset generation.
HiGen: Hierarchical Graph Generative Networks
Mahdi Karami (University of Alberta)
GenerationData SynthesisGraph Neural NetworkPoint CloudGraph
🎯 What it does: A hierarchical graph generation network HiGen is proposed, which can first generate community subgraphs in parallel and then predict inter-community edges, thus achieving a coarse-to-fine hierarchical graph generation.
High-dimensional SGD aligns with emerging outlier eigenspaces
Gerard Ben Arous (Courant Institute of Mathematical Sciences New York University), Aukosh Jagannath (University of Waterloo)
ClassificationOptimizationStochastic Differential Equation
🎯 What it does: This paper studies the training dynamics of Stochastic Gradient Descent (SGD) in high-dimensional classification tasks and the co-evolution of the empirical Hessian and gradient matrix spectra, proving that the SGD trajectory quickly aligns with the low-rank anomalous feature space of the Hessian and gradient matrices.
Hindsight PRIORs for Reward Learning from Human Preferences
Mudit Verma (Arizona State University), Katherine Metcalf (Apple Inc.)
Robotic IntelligenceReinforcement Learning from Human FeedbackTransformerReinforcement LearningWorld ModelSequential
🎯 What it does: A reward learning method based on Hindsight Prior is proposed, utilizing the attention weights of the world model for credit assignment, thereby improving the reward function based on human preference learning.
HoloNets: Spectral Convolutions do extend to Directed Graphs
Christian Koke (Technical University Munich), Daniel Cremers (Technical University Munich)
ClassificationOptimizationGraph Neural NetworkGraph
🎯 What it does: The HoloNet framework is proposed, realizing a trainable spectral convolutional network on directed graphs, breaking free from the limitations of traditional graph Fourier transforms.
Horizon-Free Regret for Linear Markov Decision Processes
Zihan Zhang (Princeton University), Simon Shaolei Du
OptimizationReinforcement Learning
🎯 What it does: This paper presents the first time-independent regret bound for linear Markov decision processes (MDP), addressing the sample complexity issue when the state space is large.