LanczosNet: Multi-Scale Deep Graph Convolutional Networks

Abstract

We propose the Lanczos network (LanczosNet), which uses the Lanczos algorithm to construct low rank approximations of the graph Laplacian for graph convolution. Relying on the tridiagonal decomposition of the Lanczos algorithm, we not only efficiently exploit multi-scale information via fast approximated computation of matrix power but also design learnable spectral filters. Being fully differentiable, LanczosNet facilitates both graph kernel learning as well as learning node embeddings. We show the connection between our LanczosNet and graph based manifold learning methods, especially the diffusion maps. We benchmark our model against several recent deep graph networks on citation networks and QM8 quantum chemistry dataset. Experimental results show that our model achieves the state-of-the-art performance in most tasks. Code is released at: \url{https://github.com/lrjconan/LanczosNetwork}.

The International Conference on Learning Representations (ICLR) 2019

References in corpus (1)

• FastGCN: Fast Learning with Graph Convolutional Networks via Importance Sampling

Cited by in corpus (67)

• Simplifying Graph Convolutional Networks
• Graph Neural Networks with convolutional ARMA filters
• Gated Graph Recurrent Neural Networks
• Graph Neural Networks: Taxonomy, Advances and Trends
• Cross-modal Knowledge Reasoning for Knowledge-based Visual Question Answering
• Masked Label Prediction: Unified Message Passing Model for Semi-Supervised Classification
• Unifying Graph Convolutional Neural Networks and Label Propagation
• Disentangling and Unifying Graph Convolutions for Skeleton-Based Action Recognition
• Representation Learning for Dynamic Graphs: A Survey
• Should Graph Convolution Trust Neighbors? A Simple Causal Inference Method
• Directed Graph Convolutional Network
• Alchemy: A Quantum Chemistry Dataset for Benchmarking AI Models
• Graph-MLP: Node Classification without Message Passing in Graph
• Are Powerful Graph Neural Nets Necessary? A Dissection on Graph Classification
• Multi-hop Attention Graph Neural Network
• Path Integral Based Convolution and Pooling for Graph Neural Networks
• AdaGCN: Adaboosting Graph Convolutional Networks into Deep Models
• Sub-graph Contrast for Scalable Self-Supervised Graph Representation Learning
• Break the Ceiling: Stronger Multi-scale Deep Graph Convolutional Networks
• What graph neural networks cannot learn: depth vs width
• Permutation Invariant Graph Generation via Score-Based Generative Modeling
• Spatial Pyramid Based Graph Reasoning for Semantic Segmentation
• Label Efficient Semi-Supervised Learning via Graph Filtering
• Local Augmentation for Graph Neural Networks
• Complete the Missing Half: Augmenting Aggregation Filtering with Diversification for Graph Convolutional Networks
• PAN: Path Integral Based Convolution for Deep Graph Neural Networks
• Utilizing Edge Features in Graph Neural Networks via Variational Information Maximization
• Beltrami Flow and Neural Diffusion on Graphs
• SAC: Accelerating and Structuring Self-Attention via Sparse Adaptive Connection
• Bilinear Graph Neural Network with Neighbor Interactions
• Higher-order Weighted Graph Convolutional Networks
• Ripple Walk Training: A Subgraph-based training framework for Large and Deep Graph Neural Network
• Fisher-Bures Adversary Graph Convolutional Networks
• How Framelets Enhance Graph Neural Networks
• Training Matters: Unlocking Potentials of Deeper Graph Convolutional Neural Networks
• DFNets: Spectral CNNs for Graphs with Feedback-Looped Filters
• Why Propagate Alone? Parallel Use of Labels and Features on Graphs
• Graph Neural Network Architecture Search for Molecular Property Prediction
• Graph Convolution with Low-rank Learnable Local Filters
• Optimization and Generalization Analysis of Transduction through Gradient Boosting and Application to Multi-scale Graph Neural Networks
• Spectral Graph Attention Network with Fast Eigen-approximation
• Graph Cross Networks with Vertex Infomax Pooling
• $p$-Laplacian Based Graph Neural Networks
• Hop-Hop Relation-aware Graph Neural Networks
• PushNet: Efficient and Adaptive Neural Message Passing
• Towards Scale-Invariant Graph-related Problem Solving by Iterative Homogeneous Graph Neural Networks
• PINE: Universal Deep Embedding for Graph Nodes via Partial Permutation Invariant Set Functions
• Enhance Information Propagation for Graph Neural Network by Heterogeneous Aggregations
• SoGCN: Second-Order Graph Convolutional Networks
• Stacked Graph Filter
• Graph Attention Networks with Positional Embeddings
• Fast and Accurate: Structure Coherence Component for Face Alignment
• Data-Driven Learning of Geometric Scattering Networks
• Unsupervised Learning of Graph Hierarchical Abstractions with Differentiable Coarsening and Optimal Transport
• Graphs, Entities, and Step Mixture
• Which Hyperparameters to Optimise? An Investigation of Evolutionary Hyperparameter Optimisation in Graph Neural Network For Molecular Property Prediction
• Graph Denoising with Framelet Regularizer
• Graph Prolongation Convolutional Networks: Explicitly Multiscale Machine Learning on Graphs with Applications to Modeling of Cytoskeleton
• Adversarial Stein Training for Graph Energy Models
• Simple Graph Convolutional Networks
• Image-Like Graph Representations for Improved Molecular Property Prediction
• Co-embedding of Nodes and Edges with Graph Neural Networks
• Graphs for deep learning representations
• Fully Linear Graph Convolutional Networks for Semi-Supervised Learning and Clustering
• Hierarchic Neighbors Embedding
• Fast Haar Transforms for Graph Neural Networks
• Community-preserving Graph Convolutions for Structural and Functional Joint Embedding of Brain Networks