Graph Neural Networks with convolutional ARMA filters

Abstract

Popular graph neural networks implement convolution operations on graphs based on polynomial spectral filters. In this paper, we propose a novel graph convolutional layer inspired by the auto-regressive moving average (ARMA) filter that, compared to polynomial ones, provides a more flexible frequency response, is more robust to noise, and better captures the global graph structure. We propose a graph neural network implementation of the ARMA filter with a recursive and distributed formulation, obtaining a convolutional layer that is efficient to train, localized in the node space, and can be transferred to new graphs at test time. We perform a spectral analysis to study the filtering effect of the proposed ARMA layer and report experiments on four downstream tasks: semi-supervised node classification, graph signal classification, graph classification, and graph regression. Results show that the proposed ARMA layer brings significant improvements over graph neural networks based on polynomial filters.

References in corpus (26)

• Efficient Estimation of Word Representations in Vector Space
• DeepWalk: Online Learning of Social Representations
• Google's Neural Machine Translation System: Bridging the Gap between Human and Machine Translation
• Inductive Representation Learning on Large Graphs
• Geometric deep learning: going beyond Euclidean data
• Spectral Networks and Locally Connected Networks on Graphs
• Deep Convolutional Networks on Graph-Structured Data
• Fast Graph Representation Learning with PyTorch Geometric
• Simplifying Graph Convolutional Networks
• Representation Learning on Graphs with Jumping Knowledge Networks
• Hierarchical Graph Representation Learning with Differentiable Pooling
• Learning Convolutional Neural Networks for Graphs
• Modern temporal network theory: A colloquium
• Predict then Propagate: Graph Neural Networks meet Personalized PageRank
• An overview and comparative analysis of Recurrent Neural Networks for Short Term Load Forecasting
• Autoregressive Moving Average Graph Filtering
• Revisiting Graph Neural Networks: All We Have is Low-Pass Filters
• LanczosNet: Multi-Scale Deep Graph Convolutional Networks
• Chebyshev Polynomial Approximation for Distributed Signal Processing
• Distributed Autoregressive Moving Average Graph Filters
• Graph Convolutional Neural Networks via Scattering
• Graph Neural Networks in TensorFlow and Keras with Spektral
• Accelerated filtering on graphs using Lanczos method
• Change Detection in Graph Streams by Learning Graph Embeddings on Constant-Curvature Manifolds
• Stability of Graph Scattering Transforms
• Dynamical Systems as Temporal Feature Spaces

Cited by in corpus (71)

• Fast Graph Representation Learning with PyTorch Geometric
• Physics-informed graph neural Galerkin networks: A unified framework for solving PDE-governed forward and inverse problems
• SIGN: Scalable Inception Graph Neural Networks
• Joint Detection and Localization of Stealth False Data Injection Attacks in Smart Grids using Graph Neural Networks
• Differentiable Graph Module (DGM) for Graph Convolutional Networks
• Adaptive Propagation Graph Convolutional Network
• Measuring and Relieving the Over-smoothing Problem for Graph Neural Networks from the Topological View
• Spectral Clustering with Graph Neural Networks for Graph Pooling
• Power Flow Balancing with Decentralized Graph Neural Networks
• RoSGAS: Adaptive Social Bot Detection with Reinforced Self-Supervised GNN Architecture Search
• Hierarchical Representation Learning in Graph Neural Networks with Node Decimation Pooling
• Regular Splitting Graph Network for 3D Human Pose Estimation
• High-Level Synthesis Performance Prediction using GNNs: Benchmarking, Modeling, and Advancing
• Graph Convolutional Neural Networks with Node Transition Probability-based Message Passing and DropNode Regularization
• Predicting Basin Stability of Power Grids using Graph Neural Networks
• Drug-Target Interaction/Affinity Prediction: Deep Learning Models and Advances Review
• Bridging the Gap between Spatial and Spectral Domains: A Survey on Graph Neural Networks
• Incorporating Symbolic Domain Knowledge into Graph Neural Networks
• BernNet: Learning Arbitrary Graph Spectral Filters via Bernstein Approximation
• Deeper Insights into Deep Graph Convolutional Networks: Stability and Generalization
• Toward Dynamic Stability Assessment of Power Grid Topologies using Graph Neural Networks
• Beyond Low-Pass Filters: Adaptive Feature Propagation on Graphs
• GraphWorld: Fake Graphs Bring Real Insights for GNNs
• Graph Convolutional Networks based on Manifold Learning for Semi-Supervised Image Classification
• Graph-adaptive Rectified Linear Unit for Graph Neural Networks
• Graph Neural Networks with Diverse Spectral Filtering
• Inclusion of Domain-Knowledge into GNNs using Mode-Directed Inverse Entailment
• ES-GNN: Generalizing Graph Neural Networks Beyond Homophily with Edge Splitting
• Collaborate to Adapt: Source-Free Graph Domain Adaptation via Bi-directional Adaptation
• Geometric Scattering Attention Networks
• Discriminative structural graph classification
• Rank Flow Embedding for Unsupervised and Semi-Supervised Manifold Learning
• VQ-GNN: A Universal Framework to Scale up Graph Neural Networks using Vector Quantization
• EasyDGL: Encode, Train and Interpret for Continuous-time Dynamic Graph Learning
• Towards Lightweight Graph Neural Network Search with Curriculum Graph Sparsification
• Physics-Informed Graph-Mesh Networks for PDEs: A hybrid approach for complex problems
• DFNets: Spectral CNNs for Graphs with Feedback-Looped Filters
• Impact-Oriented Contextual Scholar Profiling using Self-Citation Graphs
• A pipeline for fair comparison of graph neural networks in node classification tasks
• Understanding Concepts in Graph Signal Processing for Neurophysiological Signal Analysis
• Adversarial Graph Disentanglement
• Blind Graph Matching Using Graph Signals
• A Comprehensive Benchmark on Spectral GNNs: The Impact on Efficiency, Memory, and Effectiveness
• Knowledge Distillation and Enhanced Subdomain Adaptation Using Graph Convolutional Network for Resource-Constrained Bearing Fault Diagnosis
• Spectral Graph Attention Network with Fast Eigen-approximation
• Human Action Recognition with Multi-Laplacian Graph Convolutional Networks
• Search For Deep Graph Neural Networks
• Graph Neural Networks Including Sparse Interpretability
• TactileSGNet: A Spiking Graph Neural Network for Event-based Tactile Object Recognition
• HighwayGraph: Modelling Long-distance Node Relations for Improving General Graph Neural Network
• Stacked Graph Filter
• ERGNN: Spectral Graph Neural Network With Explicitly-Optimized Rational Graph Filters
• On the Stability of Graph Convolutional Neural Networks under Edge Rewiring
• SoGCN: Second-Order Graph Convolutional Networks
• Learning Optimal Graph Filters for Clustering of Attributed Graphs
• Message Passing in Graph Convolution Networks via Adaptive Filter Banks
• BiGCN: A Bi-directional Low-Pass Filtering Graph Neural Network
• Fea2Fea: Exploring Structural Feature Correlations via Graph Neural Networks
• Program-to-Circuit: Exploiting GNNs for Program Representation and Circuit Translation
• Learning Conjoint Attentions for Graph Neural Nets
• Totally Deep Support Vector Machines
• Learning nuclear cross sections across the chart of nuclides with graph neural networks
• Understanding and Tackling Over-Dilution in Graph Neural Networks
• A Comparative Study of Graph Neural Network Layer Selection for Interaction Modelling in Driving Trajectory Prediction
• Action Recognition with Kernel-based Graph Convolutional Networks
• DURENDAL: Graph deep learning framework for temporal heterogeneous networks
• Graph Feature Gating Networks
• Pseudoinverse Graph Convolutional Networks: Fast Filters Tailored for Large Eigengaps of Dense Graphs and Hypergraphs
• Simple Graph Convolutional Networks
• LFGCN: Levitating over Graphs with Levy Flights
• Brain dynamics via Cumulative Auto-Regressive Self-Attention