神经网络能用来干什么_「十项全能」图神经网络能干嘛?

前几天在群里划水,有人在问图神经网络能应用于NLP/CV吗?

回想一下这两年GNN的发展,貌似我知道的领域全都有GNN的身影. 这让我想起了中文西那把无所不能的终极杀人武器:夺命3000.

下面就梳理一下GNN活跃的十个领域.正所谓"十项全能GNN"

注: 这里每个领域只简单罗列几篇,只是为了说明GNN得到了成功应用.

自然语言处理

19AAAI Graph convolutional networks for text classification

19EMNLP Heterogeneous Graph Attention Networks for Semi-supervised Short Text Classification

计算机视觉

20CVPR Graph-FCN for image semantic segmentation

20CVPR GraphTER: Unsupervised Learning of Graph Transformation Equivariant Representations via Auto-Encoding Node-wise Transformations

20CVPR Grid-GCN for Fast and Scalable Point Cloud Learning

推荐系统

19KDD MEIRec Metapath-guided Heterogeneous Graph Neural Network for Intent Recommendation

19SIGIR NGCF Neural Graph Collaborative Filtering

18KDD PinSage Graph Convolutional Neural Networks for Web-Scale Recommender Systems

强化学习

18ICLR NerveNet: Learning Structured Policy with Graph Neural Networks

19NIPS Learning Transferable Graph Exploration

20AAAI Multi-Agent Game Abstraction via Graph Attention Neural Network

交通/城市

19AAAI Spatiotemporal Multi‐Graph Convolution Network for Ride-hailing Demand Forecasting

19NIPS Social-BiGAT: Multimodal Trajectory Forecasting using Bicycle-GAN and Graph Attention Networks. NeurIPS 2019.

20AAAI GMAN: A Graph Multi‐Attention Network for Traffic Prediction

物理

19ICML Graph Element Networks: adaptive, structured computation and memory

18ICML Graph networks as learnable physics engines for inference and control

组合优化

19ICML Learning a SAT Solver from Single-Bit Supervision

19AAAI Learning to Solve NP-Complete Problems - A Graph Neural Network for Decision TSP

19NIPS Exact Combinatorial Optimization with Graph Convolutional Neural Networks.

分子/药物设计

20ICLR Directional Message Passing for Molecular Graphs

19ICLR Learning Multimodal Graph-to-Graph Translation for Molecular Optimization

19KDD Graph Transformation Policy Network for Chemical Reaction Prediction

程序代码分析

19ICML Open Vocabulary Learning on Source Code with a Graph-Structured Cache

19NIPS Devign: Effective Vulnerability Identification by Learning Comprehensive Program Semantics via Graph Neural Networks

恶意检测

19AAAI Graph Neural Networks with Adaptive Receptive Paths

19CIKM GAS Spam Review Detection with Graph Convolutional Networks