Revisiting Graph Neural Networks: All We Have is Low-Pass Filters
arXiv:1905.09550
Abstract
Graph neural networks have become one of the most important techniques to solve machine learning problems on graph-structured data. Recent work on vertex classification proposed deep and distributed learning models to achieve high performance and scalability. However, we find that the feature vectors of benchmark datasets are already quite informative for the classification task, and the graph structure only provides a means to denoise the data. In this paper, we develop a theoretical framework based on graph signal processing for analyzing graph neural networks. Our results indicate that graph neural networks only perform low-pass filtering on feature vectors and do not have the non-linear manifold learning property. We further investigate their resilience to feature noise and propose some insights on GCN-based graph neural network design.
12 pages, 5 figures, 2 tables
References in corpus (2)
Cited by in corpus (26)
- Towards Deeper Graph Neural Networks with Differentiable Group Normalization
- Self-supervised Consensus Representation Learning for Attributed Graph
- Is Heterophily A Real Nightmare For Graph Neural Networks To Do Node Classification?
- Personalized Graph Signal Processing for Collaborative Filtering
- Dirichlet Energy Constrained Learning for Deep Graph Neural Networks
- Robust Mid-Pass Filtering Graph Convolutional Networks
- Graph Neural Networks: Methods, Applications, and Opportunities
- Interpreting and Unifying Graph Neural Networks with An Optimization Framework
- Scalable Graph Neural Networks for Heterogeneous Graphs
- Coupled Layer-wise Graph Convolution for Transportation Demand Prediction
- Graph Representation Learning Beyond Node and Homophily
- Evaluating Deep Graph Neural Networks
- Beltrami Flow and Neural Diffusion on Graphs
- Understanding Graph Neural Networks from Graph Signal Denoising Perspectives
- Dirichlet Graph Variational Autoencoder
- On Graph Neural Networks versus Graph-Augmented MLPs
- Maximum Entropy Weighted Independent Set Pooling for Graph Neural Networks
- Stacked Graph Filter
- Message Passing in Graph Convolution Networks via Adaptive Filter Banks
- SoGCN: Second-Order Graph Convolutional Networks
- Eigen-GNN: A Graph Structure Preserving Plug-in for GNNs
- Graph Denoising with Framelet Regularizer
- Multi-hop Graph Convolutional Network with High-order Chebyshev Approximation for Text Reasoning
- Meta-Aggregator: Learning to Aggregate for 1-bit Graph Neural Networks
- Learning on Random Balls is Sufficient for Estimating (Some) Graph Parameters
- Efficient Graph based Recommender System with Weighted Averaging of Messages