π― What it does: This paper studies how to efficiently learn general representations through self-supervised learning (SSL) in a decentralized and non-IID unlabeled data environment, and explores its feasibility and advantages within a federated learning framework.
π― What it does: This paper proposes and systematically evaluates a zero-shot reinforcement learning method, primarily based on Success Features (SF) and Forward-Backward (FB) representations, aimed at training an agent that can immediately execute any reward task without further planning.
Domain Generalization via Heckman-type Selection Models
Hyungu Kahng (Korea University), Judy Zhong (New York University School of Medicine)
CodeDomain AdaptationTabularBenchmark
π― What it does: Modeling the domain generalization problem as a non-random sample selection problem, jointly learning the prediction model and the domain selection model, aiming to achieve more robust predictions against the true population distribution.
π― What it does: A variational inference-based domain index inference framework (VDI) is proposed, which can learn interpretable continuous domain indices from multi-domain data under the condition of only given domain identities;
Donβt fear the unlabelled: safe semi-supervised learning via debiasing
Hugo Schmutz (Universite Cote d'Azur), Pierre-Alexandre Mattei (Universite Cote d'Azur)
CodeClassificationData-Centric LearningImage
π― What it does: A debiased semi-supervised learning framework, DeSSL, is proposed, which incorporates control variables into the traditional SSL risk estimation, ensuring that the risk estimation is unbiased under the MCAR assumption and has theoretical guarantees.
π― What it does: DropIT reduces GPU memory usage and approximates gradients by dropping intermediate tensors (activations) based on absolute minimum values or randomly during training, and then filling in the missing elements with zeros during backpropagation.
Dual Diffusion Implicit Bridges for Image-to-Image Translation
Xuan Su (Stanford University), Stefano Ermon (Stanford University)
CodeImage TranslationFederated LearningSafty and PrivacyDiffusion modelScore-based ModelImageOrdinary Differential Equation
π― What it does: Proposes Dual Diffusion Implicit Bridges (DDIBs), which utilize two independently trained diffusion models to encode source domain images into latent space, and then decode using the target domain model to obtain target domain images, completing unpaired image translation.
Dynamic Update-to-Data Ratio: Minimizing World Model Overfitting
Nicolai Dorka (University of Freiburg), Wolfram Burgard (University of Technology Nuremberg)
CodeReinforcement LearningWorld ModelImage
π― What it does: A dynamic update-to-data ratio (DUTD) method is proposed and implemented for world model training in model-based reinforcement learning. It automatically detects underfitting and overfitting by monitoring validation set errors and adjusts the ratio of training steps to data steps accordingly.
π― What it does: This paper proposes an adaptive super-resolution framework DySR based on algorithm and system co-design, which can dynamically switch subgraphs on mobile devices according to real-time computing and memory resources, maximizing super-resolution quality while maintaining frame rates.
E-CRF: Embedded Conditional Random Field for Boundary-caused Class Weights Confusion in Semantic Segmentation
Jie Zhu (Peking University), Leye Wang (Peking University)
CodeSegmentationConvolutional Neural NetworkImage
π― What it does: To address the class weight confusion (BCWC) problem in semantic segmentation, an Embedded Conditional Random Field (E-CRF) model is proposed, which integrates the CRF mechanism with CNN networks to enhance class weight discrimination and improve boundary pixel prediction.
π― What it does: A joint architecture and hyperparameter search benchmark for energy consumption, EAβHASβBench, is proposed, providing complete energy consumption and performance data.
Edgeformers: Graph-Empowered Transformers for Representation Learning on Textual-Edge Networks
Bowen Jin (University of Illinois), Jiawei Han (University of Illinois)
CodeRepresentation LearningGraph Neural NetworkTransformerLarge Language ModelTextGraph
π― What it does: For text-edge networks, the Edgeformers framework is proposed, which injects virtual node tokens into the Transformer layers to deeply couple graph structures with text information, achieving edge and node representation learning.
π― What it does: This paper proposes a framework called GAINS for efficient certified training and robustness verification of high-dimensional Neural Ordinary Differential Equations (NODEs), addressing the issue of continuous step size abstraction caused by traditional adaptive ODE solvers.
Efficient Discrete Multi Marginal Optimal Transport Regularization
Ronak Mehta (University of Wisconsin Madison), Vikas Singh (University of Wisconsin Madison)
CodeImage TranslationOptimizationImageTabular
π― What it does: This paper proposes an efficient Discrete Multi-Marginal Optimal Transport (MMOT) regularization method called DEMD, which is integrated into a deep learning framework for tasks such as fairness, discernible representation, and multi-domain image translation.
π― What it does: This paper proposes an end-to-end hybrid learning framework that queries the routing model on edge devices only for hard examples that are correctly classified in the cloud, significantly reducing overall inference latency while maintaining high accuracy.
π― What it does: A lightweight single-step model-based offline reinforcement learning algorithm, ROSMO, is proposed to replace the high-cost MCTS of MuZero Unplugged for policy improvement.
π― What it does: A GRU-based event-driven RNN, called EGRU, is proposed, which achieves activity sparsity through event communication triggered by thresholds, allowing the forward and backward propagation computations to scale linearly with the number of events.
Emergent World Representations: Exploring a Sequence Model Trained on a Synthetic Task
Kenneth Li (Harvard University), Martin Wattenberg (Harvard University)
CodeExplainability and InterpretabilityRepresentation LearningTransformerLarge Language ModelSupervised Fine-TuningTextSequential
π― What it does: This study investigates a variant of GPT that predicts legal moves in Othello (Reversi) without prior rules. It finds that nonlinear board state representations emerge in its internal activations and verifies that this representation has a causal impact on model predictions through activation interventions. Consequently, it proposes an intervention-based latent saliency map to explain model decisions.
π― What it does: Proposes the GTRANS framework, which utilizes gradient descent to perturb graph structures and node features during testing, enhancing the generalization and robustness of pre-trained GNNs under suboptimal data.
π― What it does: A method for out-of-distribution (OOD) detection based on energy functions in graph neural networks (GNN), named GNNSAFE, is proposed, which can obtain an OOD discriminator using only a standard supervised trained GNN classifier.
EPISODE: Episodic Gradient Clipping with Periodic Resampled Corrections for Federated Learning with Heterogeneous Data
Michael Crawshaw (George Mason University), Mingrui Liu (Shanghai Jiao Tong University)
CodeOptimizationFederated LearningImageText
π― What it does: EPISODE is proposed, a gradient clipping algorithm for heterogeneous data in federated learning that is non-convex and satisfies relaxed smoothness ((L,L0,1)-smoothness);
Equal Improvability: A New Fairness Notion Considering the Long-term Impact
Ozgur Guldogan (University of California), Kangwook Lee (University of Wisconsin)
CodeTabularFinance Related
π― What it does: The concept of Equal Improvability (EI) fairness is proposed, along with three methods for implementing this concept through fair regularization; its effectiveness is validated in both static and dynamic scenarios.
Equiformer: Equivariant Graph Attention Transformer for 3D Atomistic Graphs
Yi-Lun Liao (Massachusetts Institute of Technology), Tess Smidt (Massachusetts Institute of Technology)
CodeGraph Neural NetworkTransformerGraphBenchmarkPhysics Related
π― What it does: This paper proposes Equiformer, a transformation-based equivariant graph neural network for predicting quantum properties of 3D atomic graphs.
π― What it does: Equivariant Descriptor Fields (EDFs) are proposed, an end-to-end energy model that utilizes SE(3) symmetry for visual robot manipulation learning with only 5 to 10 demonstration examples;
Peihao Wang (University of Texas at Austin), Pan Li (Georgia Tech)
CodeClassificationGraph Neural NetworkGraph
π― What it does: This paper proposes ED-HNN, a equivariant hypergraph neural network based on transferable hypergraph diffusion, designed to efficiently capture high-order relationships in hypergraphs and perform node classification tasks.
π― What it does: A shape-based 3D molecular generation framework called SQUID is proposed, which can automatically generate drug-like molecules with diverse chemical structures under given 3D shape constraints.
π― What it does: Proposes ERL-Re 2, which integrates evolutionary algorithms with reinforcement learning, utilizing shared nonlinear state representations and individual linear policy representations for efficient policy optimization.
π― What it does: This paper proposes an experience replay framework based on error sensitivity modulation, ESMER, which utilizes dual-mode memory (short-term experience buffer and long-term semantic model) and error history memory to alleviate representation drift and catastrophic forgetting in continual learning.
π― What it does: A trainable calibration loss without internal hyperparameters is proposed - Expected Squared Difference (ESD), which can be optimized simultaneously with negative log-likelihood loss;
π― What it does: A second-order optimizer named Eva is proposed, which constructs the curvature matrix using Kronecker-vectorization approximation and implements gradient preconditioning without explicit inverse matrices using the Sherman-Morrison formula;
π― What it does: We propose EVA3D, a human generation model based on compositional NeRF, capable of learning to generate high-resolution (512Γ256) 3D humans from a collection of 2D images.
π― What it does: A 3D maze benchmark specifically designed to measure the long-term memory capabilities of reinforcement learning agents, called Memory Maze, has been designed and evaluated, providing human baselines, offline datasets, and representation learning probes.
π― What it does: Designed and implemented an efficient single-model variable bitrate image compression scheme that balances real-time performance with excellent rate-distortion performance.
Everybody Needs Good Neighbours: An Unsupervised Locality-based Method for Bias Mitigation
Xudong Han (University of Melbourne), Trevor Cohn (University of Melbourne)
CodeTabular
π― What it does: An unsupervised neighborhood-based proxy labeling method, ULPL, is proposed for bias mitigation in models without protected attribute labels.
Evolving Populations of Diverse RL Agents with MAP-Elites
Thomas PIERROT, Arthur Flajolet (InstaDeep)
CodeRobotic IntelligenceReinforcement LearningAgentic AI
π― What it does: A PBT-MAP-ELITES framework is proposed that combines a complete RL agent (including policy parameters, other learnable parameters, and hyperparameters) with MAP-ELITES, aiming to achieve both high quality and diversity simultaneously.
π― What it does: This paper proposes ED-Pose, a fully end-to-end single-stage multi-person pose estimation framework that utilizes explicit box detection to unify global (human-level) and local (joint-level) information.
Exploring Active 3D Object Detection from a Generalization Perspective
Yadan Luo (University of Queensland), Mahsa Baktashmotlagh (University of Queensland)
CodeObject DetectionAutonomous DrivingPoint Cloud
π― What it does: An active learning framework CRB is proposed in LiDAR 3D object detection, aiming to achieve near fully supervised detection performance with extremely low annotation costs.
π― What it does: A multi-instance learning method based on low-rank properties is proposed, which enhances feature embedding through low-rank constrained contrastive learning (LRC) and designs an Iterative Low-Rank Attention Module (ILRA-MIL) to efficiently model the inter-instance correlations of whole slide images, achieving whole slide image classification.
π― What it does: This paper proposes and implements DynaAugment, a dynamic data augmentation framework designed for video recognition, capable of generating smooth and diverse augmentation magnitudes for each frame.
Exploring the Limits of Differentially Private Deep Learning with Group-wise Clipping
Jiyan He (University of Science and Technology of China), Jiang Bian (Microsoft Research)
CodeSafty and PrivacyComputational EfficiencyTransformerSupervised Fine-TuningText
π― What it does: This paper researches and implements a differential privacy deep learning method based on group clipping, proposing layer-wise clipping that can run in parallel with backpropagation and device-level clipping aimed at large models, validating its effectiveness on various tasks.
Niklas Nolte (Massachusetts Institute of Technology), Mike Williams (Massachusetts Institute of Technology)
CodeClassificationOptimizationImageTabularPhysics Related
π― What it does: This paper proposes a method that applies L1-Lipschitz constraints to fully connected layers and adds a single residual connection, which can strictly ensure that the network is monotonic for any specified input subset and possesses provable robustness.
π― What it does: A new knowledge graph embedding model called ExpressivE is proposed, which embeds entities as points and relationships as hyper-parallelograms in a virtual triple space to achieve richer reasoning patterns.
π― What it does: This paper proposes a model extraction method based on uncertain samples (UE) called Boundary Entropy Searching Thief (BEST), which can recover the clarity accuracy and robustness of the target model in a black-box Machine Learning as a Service (MLaaS) environment using only hard label queries.
Alasdair Tran (Australian National University), Cheng Soon Ong (Data61, CSIRO)
CodeComputational EfficiencyPoint CloudPhysics Related
π― What it does: An improved Fourier Neural Operator (F-FNO) is proposed for efficient learning and simulation of various PDEs (Navier-Stokes, elasticity, aerodynamics, plastic forming).
π― What it does: The FairAC framework is proposed, which can complete missing attributes in graphs while simultaneously suppressing unfairness at both feature and topological levels.
FaiREE: fair classification with finite-sample and distribution-free guarantee
Puheng Li (Peking University), Linjun Zhang (Rutgers University)
CodeClassificationTabular
π― What it does: A post-processing algorithm FaiREE is proposed to achieve group fair classification under finite samples and no distribution assumptions.
CodeClassificationOptimizationSupervised Fine-TuningTabularFinance Related
π― What it does: Proposes the FairGBM framework, which implements fairness constraints in gradient boosting trees (GBDT) through dual ascent learning.
Thai-Hoang Pham (Ohio State University), Ping Zhang (Ohio State University)
CodeDomain AdaptationRepresentation LearningGenerative Adversarial NetworkImageBiomedical DataElectronic Health Records
π― What it does: The research focuses on maintaining both model accuracy and fairness in domain generalization scenarios, and proposes a two-stage training framework based on density matching called FATDM.
π― What it does: For anomaly detection in near-distribution scenarios, this paper proposes using a non-adversarial diffusion model to generate approximate anomalous samples and fine-tuning a pre-trained feature extractor on the generated samples to enhance detection performance.
Fantastic Rewards and How to Tame Them: A Case Study on Reward Learning for Task-oriented Dialogue Systems
Yihao Feng (Salesforce Research), Huan Wang (Salesforce Research)
CodeTransformerReinforcement LearningText
π― What it does: Two learning-to-rank based reward function learning methods, RewardNet and RewardMLE, are proposed, and the learned reward functions are combined with Gumbel-softmax policy gradients for training end-to-end task-oriented dialogue systems.
π― What it does: The Adaptive Subgoal Search (AdaSubS) algorithm is proposed, which can adaptively adjust the time window for subgoal search based on state complexity, thereby planning more efficiently in complex reasoning tasks.
π― What it does: A fast sampling method based on the Exponential Integrator, DEIS, has been developed to accelerate the generation process of diffusion models.
Florian Jaeckle (University of Oxford), Hadi Pouransari (Apple)
CodeRetrievalOptimizationImage
π― What it does: This paper proposes FastFill, a method compatible with model updates and online local backfilling, which can quickly enhance retrieval performance.
Aoxiao Zhong (Harvard University), Quanzheng Li (Massachusetts General Hospital)
CodeFederated LearningRepresentation LearningBiomedical Data
π― What it does: A new method called FedDAR is proposed for domain-mixed data distribution in cross-site federated learning, achieving model personalization through learning shared encoders and domain-specific prediction heads.
Federated Learning as Variational Inference: A Scalable Expectation Propagation Approach
Han Guo (Carnegie Mellon University), Eric Xing
CodeFederated LearningImageText
π― What it does: Proposed FedEP: a federated learning framework based on expectation propagation, which improves global posterior approximation using distributed inference;
Michael Kamp (University Hospital Essen), Jilles Vreeken (CISPA Helmholtz Center for Information Security)
CodeFederated LearningSafty and PrivacyConvolutional Neural NetworkImageBiomedical DataMagnetic Resonance Imaging
π― What it does: A federated learning method called FEDDC is proposed, which combines 'daisy chaining' and aggregation to achieve effective training in scenarios where each client has very few samples.
Yichao Du (University of Science and Technology of China), Enhong Chen (University of Science and Technology of China)
CodeFederated LearningSafty and PrivacyComputational EfficiencyText
π― What it does: Proposes the FedNN framework, which uses single-round memory-based interaction to replace traditional FedAvg, achieving efficient training and knowledge sharing for federated NMT.
π― What it does: Proposes FEDFA, which achieves data augmentation in federated learning through Gaussian sampling of feature statistics to alleviate feature shift among clients.
π― What it does: Construct a small number of powerful backdoor trigger samples and achieve a high attack success rate by predicting the training process through the Neural Tangent Kernel (NTK).
π― What it does: A few-shot cross-domain image generation method is proposed for learning latent codes during the inference phase without the need to retrain the generator;
Few-Shot Domain Adaptation For End-to-End Communication
Jayaram Raghuram (University of Wisconsin - Madison), Suman Banerjee (University of Wisconsin - Madison)
CodeDomain AdaptationAuto Encoder
π― What it does: A few-shot domain adaptation method for end-to-end communication systems is proposed, achieving rapid channel model transfer through the adaptation of Gaussian Mixture Networks (MDN) without retraining the encoder/decoder.
FIGARO: Controllable Music Generation using Learned and Expert Features
Dimitri von RΓΌtte (ETH Zurich), Thomas Hofmann (ETH Zurich)
CodeGenerationTransformerAuto EncoderAudio
π― What it does: The paper proposes a controllable music generation model FIGARO based on Transformer, which combines high-level features manually extracted by experts with vectors obtained through self-supervised learning to construct a self-supervised training framework from description to sequence.
π― What it does: This paper proposes FINDE (First Integral-preserving Neural Differential Equation), a neural network framework that can automatically discover and preserve unknown conserved quantities of dynamical systems from data.
π― What it does: This study investigates the theoretical foundation of adversarial training, reveals the error in Corollary C.2 proposed by Madry et al., and introduces the Danskin descent direction (DDi) algorithm using multiple adversarial samples to obtain the true descent direction.
π― What it does: A method called FiLM Transfer (FIT) has been developed for achieving parameter-efficient image classification transfer learning in few-shot scenarios, and it can be applied in personalized models and federated learning settings.
π― What it does: This paper proposes FLIP, a provable defense framework that combines trigger inversion and adversarial training in federated learning to reduce the success rate of backdoor attacks while maintaining the accuracy of clean samples.
π― What it does: A Flow Annealed Importance Sampling Bootstrap (FAB) method is proposed, which combines Ξ±-divergence with Ξ±=2 and Annealed Importance Sampling to train Normalizing Flows for approximating multimodal target distributions.
π― What it does: A 'Rectified Flow' model based on ODE is proposed, which learns the transport mapping from one empirical distribution to another and can be used for both unsupervised generation and domain transfer.
CodeTransformerLarge Language ModelReinforcement LearningText
π― What it does: An Expert Iteration framework is proposed for training language models to complete theorem proofs in Lean formalized mathematics, achieving self-improvement through alternating search and learning.
Free Lunch for Domain Adversarial Training: Environment Label Smoothing
YiFan Zhang, Tieniu Tan (Alibaba Group)
CodeClassificationDomain AdaptationImageTextTime Series
π― What it does: In Domain Adversarial Training (DAT), Environment Label Smoothing (ELS) is introduced to reduce the overconfidence of the discriminator and alleviate environmental label noise, enhancing training stability and convergence speed;
Sebastian Damrich (Heidelberg University), Dmitry Kobak (University of TΓΌbingen)
CodeOptimizationRepresentation LearningContrastive LearningImageBiomedical Data
π― What it does: This study investigates the essential relationship between t-SNE and UMAP within the contrastive learning framework, and proposes a tunable negative sampling spectral method to adjust the clustering tightness and global structure of embeddings by correlating Noise Contrastive Estimation (NCE) with Negative Sampling (NEG).
π― What it does: The Function-Consistent Feature Distillation (FCFD) method is proposed, which enhances the performance of the student model by using the functional consistency of intermediate features between the teacher and student in the later part of the network as supervision.
π― What it does: A convolution-based continuous super-resolution network called FunkNN is proposed, which combines a discrete generator to generate continuous images at any scale and is differentiable.
Fuzzy Alignments in Directed Acyclic Graph for Non-Autoregressive Machine Translation
Zhengrui Ma (Institute of Computing Technology, Chinese Academy of Sciences), Yang Feng (Institute of Computing Technology, Chinese Academy of Sciences)
π― What it does: A target function based on fuzzy alignment (n-gram expected matching) is proposed to train a non-autoregressive translation model with a Directed Acyclic Graph (DAG), thereby better addressing multimodal issues.
π― What it does: A guided attention module named GAMR is proposed, which learns to dynamically locate and store key information in visual reasoning tasks through a serialized attention movement and memory mechanism.
π― What it does: The paper proposes a general acceleration method for diffusion models called gDDIM, which can achieve nearly perfect sampling in a limited number of steps.
Generate rather than Retrieve: Large Language Models are Strong Context Generators
Wenhao Yu (University of Notre Dame), Meng Jiang (Microsoft Cognitive Service Research)
CodeGenerationRetrievalTransformerLarge Language ModelPrompt EngineeringTextRetrieval-Augmented Generation
π― What it does: A generate-then-read (GENREAD) framework is proposed, which directly uses large language models to generate context documents related to questions, and then a reading model provides answers based on these documents, replacing the traditional retrieve-read pipeline.
π― What it does: A method for generating diverse cooperative agents by learning incompatible strategies is proposed, aimed at enhancing the cooperation ability and diversity of the agents.
π― What it does: A Generative Augmented Flow Network (GAFlowNet) framework is proposed, which introduces intermediate rewards (through edge, state, or joint methods) into the traditional GFlowNet to enhance exploration efficiency and generation diversity in sparse reward environments.
π― What it does: A generative model based on the Inverse Heat Dissipation Model (IHDM) is designed and implemented, achieving multi-scale image generation through the inverse process of the heat equation.
π― What it does: This paper connects two types of probabilistic algorithms: Variational Inference (VI) and Generative Flow Networks (GFlowNet). It proves that they are equivalent in terms of gradient expectations under certain conditions and experimentally compares the performance of the two methods in discrete structure generation tasks.
Samuel Ainsworth, Siddhartha Srinivasa (Paul G. Allen School of Computer Science and Engineering University of Washington)
CodeOptimizationFederated LearningSafty and PrivacyConvolutional Neural NetworkImage
π― What it does: This paper proposes aligning two independently trained models by finding permutations of hidden layer units, thereby achieving linear mode connectivity (LMC) in weight space and enabling model merging.
Aohan Zeng (Tsinghua University), Jie Tang (Tsinghua University)
CodeTransformerLarge Language ModelText
π― What it does: Trained and released a 130 billion parameter bilingual (English-Chinese) large language model GLM-130B, providing complete code, training logs, and quantization implementation.
Global Explainability of GNNs via Logic Combination of Learned Concepts
Steve Azzolin (University of Trento), Andrea Passerini (University of Trento)
CodeExplainability and InterpretabilityGraph Neural NetworkGraph
π― What it does: This paper proposes GLGExplainer, which can generate global explanations through logical combinations of local explanations, thereby revealing the overall decision-making process of Graph Neural Networks (GNNs).
GNNDelete: A General Strategy for Unlearning in Graph Neural Networks
Jiali Cheng (University of Massachusetts Lowell), Marinka Zitnik (Harvard University)
CodeGraph Neural NetworkGraph
π― What it does: A general graph neural network deletion operator GNNDELETE is proposed, which can effectively delete nodes, edges, and node features without retraining the model.
GoBigger: A Scalable Platform for Cooperative-Competitive Multi-Agent Interactive Simulation
Ming Zhang (SenseTime Research), Yu Liu (Shanghai AI Laboratory)
CodeRobotic IntelligenceReinforcement Learning
π― What it does: Designed and implemented GoBigger, a scalable multi-agent interactive simulation platform that supports MΓN teams and can simulate the cooperation and competition of multiple teams within the same game.
GOOD: Exploring geometric cues for detecting objects in an open world
Haiwen Huang (University of TΓΌbingen), Dan Zhang (Bosch)
CodeObject DetectionAuto EncoderImage
π― What it does: A pseudo-labeling method based on geometric cues (depth, normals) is proposed to train an open-world category-free object detector called GOOD.
π― What it does: A non-hierarchical high-resolution visual Transformer (GPViT) is proposed, which achieves low computational cost global information exchange through the Group Propagation Block while maintaining high resolution of image features.
Gradient Boosting Performs Gaussian Process Inference
Aleksei Ustimenko (ShareChat), Liudmila Prokhorenkova (Yandex Research)
CodeTabular
π― What it does: View gradient boosting trees (GBDT) as kernel ridge regression in an implicit RKHS, prove its convergence to the posterior mean of Gaussian processes, and propose a KGB method based on random tree sampling and gradient boosting for sampling from the posterior distribution and obtaining knowledge uncertainty estimates.
π― What it does: A skeleton action recognition framework called SkeletonGCL based on graph contrastive learning is proposed, achieving self-supervised contrast across sequences through graph learning in GCN.
Graph Domain Adaptation via Theory-Grounded Spectral Regularization
Yuning You (Texas A&M University), Yang Shen (Texas A&M University)
CodeDomain AdaptationGraph Neural NetworkGraph
π― What it does: A graph domain adaptation method based on spectral regularization is proposed, which enhances cross-domain graph learning performance by controlling the spectral smoothness and maximum frequency response of graph neural networks.
Graph Neural Networks are Inherently Good Generalizers: Insights by Bridging GNNs and MLPs
Chenxiao Yang (Shanghai Jiao Tong University), Junchi Yan (Shanghai Jiao Tong University)
CodeClassificationGraph Neural NetworkGraph
π― What it does: This study investigates the generalization ability of Graph Neural Networks (GNN) in node-level prediction tasks and proposes an intermediate model called Propagational MLP (PMLP), which uses a standard Multi-Layer Perceptron (MLP) during the training phase and GNN message passing during the inference phase, to reveal the source of GNN's superiority.
Graph Neural Networks for Link Prediction with Subgraph Sketching
Benjamin Paul Chamberlain, Max Hansmire (Twitter Inc.)
CodeGraph Neural NetworkGraph
π― What it does: An efficient graph neural network ELPH (and its scalable version BUDDY) is proposed for link prediction, replacing explicit subgraph construction with subgraph Sketching, addressing the limitations of traditional MPNN in triangle counting and automatic isomorphic nodes.
π― What it does: A graph signal sampling-based inductive one-bit matrix completion framework GS-IMC and its Bayesian online extension BGS-IMC are proposed.
π― What it does: A graph-based deterministic policy gradient framework, GDPG-Twin, is proposed to solve repetitive combinatorial optimization problems and significantly reduce the optimality gap of fast distributed heuristics.