Abstract:The benefit of localized features within the regular domain has given rise to the use of Convolutional Neural Networks (CNNs) in machine learning, with great proficiency in the image classification. The use of CNNs becomes problematic within the irregular spatial domain due to design and convolution of a kernel filter being non-trivial. One solution to this problem is to utilize graph signal processing techniques and the convolution theorem to perform convolutions on the graph of the irregular domain to obtain feature map responses to learnt filters. We propose graph convolution and pooling operators analogous to those in the regular domain. We also provide gradient calculations on the input data and spectral filters, which allow for the deep learning of an irregular spatial domain problem. Signal filters take the form of spectral multipliers, applying convolution in the graph spectral domain. Applying smooth multipliers results in localized convolutions in the spatial domain, with smoother multipliers providing sharper feature maps. Algebraic Multigrid is presented as a graph pooling method, reducing the resolution of the graph through agglomeration of nodes between layers of the network. Evaluation of performance on the MNIST digit classification problem in both the regular and irregular domain is presented, with comparison drawn to standard CNN. The proposed graph CNN provides a deep learning method for the irregular domains present in the machine learning community, obtaining 94.23% on the regular grid, and 94.96% on a spatially irregular subsampled MNIST.

What problem does this paper attempt to address?

The main problem this paper attempts to address is the issue of applying Convolutional Neural Networks (CNNs) for feature extraction and classification in irregular spatial domains. Specifically, standard CNNs perform excellently when dealing with data that has a regular structure (such as images), but face difficulties when handling data with irregular structures because the design and convolutional kernel filters become very complex. To solve this problem, the authors propose graph convolution and pooling methods based on graph signal processing techniques to perform convolution operations in irregular spatial domains. ### Main Contributions: 1. **Graph Convolution and Pooling Operators**: The authors propose convolution and pooling operators similar to those in regular domains, for handling data in irregular spatial domains. 2. **Gradient Computation**: They provide methods for computing gradients of input data and spectral filters, enabling deep learning in irregular spatial domains. 3. **Graph Signal Filters**: They use spectral multipliers to perform convolution in the graph spectral domain, where smooth multipliers produce local convolutions in the spatial domain, and smoother multipliers provide sharper feature maps. 4. **Graph Pooling Method**: They introduce Algebraic Multigrid as a graph pooling method, which reduces the resolution of the graph through node aggregation. ### Experimental Results: - In the MNIST digit classification task, the proposed graph CNN achieved an accuracy of 94.23% on a regular grid and 94.96% on a spatially irregular subsampled MNIST. - Compared to standard CNNs, graph CNNs performed better in irregular spatial domains. ### Conclusion: This study proposes a new method for deep convolutional learning on irregular graphs, achieving feature extraction and classification in irregular spatial domains by combining standard graph signal processing techniques with backpropagation-based neural network design. Although its performance on regular grids is not as good as standard CNNs, its ability to handle problems in irregular spatial domains is a significant contribution.

Graph Based Convolutional Neural Network

Robust Spatial Filtering with Graph Convolutional Neural Networks

Graph Learning in Low Dimensional Space for Graph Convolutional Networks

Learning Graph Structure Via Graph Convolutional Networks

A graph convolutional neural network for classification of building patterns using spatial vector data

Analysis of Irregular Spatial Data with Machine Learning: Classification of Building Patterns with a Graph Convolutional Neural Network

Large-Scale Learnable Graph Convolutional Networks

DiffGCN: Graph Convolutional Networks via Differential Operators and Algebraic Multigrid Pooling

Graph classification based on structural features of significant nodes and spatial convolutional neural networks

Dynamic Spatio-Temporal Feature Learning via Graph Convolution in 3D Convolutional Networks

Learning Depthwise Separable Graph Convolution from Data Manifold

A Robust Alternative for Graph Convolutional Neural Networks via Graph Neighborhood Filters

Lookup subnet based Spatial Graph Convolutional neural Network

Revisiting 2D Convolutional Neural Networks for Graph-Based Applications

DGCNN: Disordered Graph Convolutional Neural Network Based on the Gaussian Mixture Model

Graph Analysis and Graph Pooling in the Spatial Domain

CNN-G: Convolutional Neural Network Combined With Graph for Image Segmentation With Theoretical Analysis

Convolutional neural networks on irregular domains based on approximate vertex-domain translations

Dynamic Edge-Conditioned Filters in Convolutional Neural Networks on Graphs

Topology and Prediction Focused Research on Graph Convolutional Neural Networks