Explore Graph Convolutional Networks (GCNs) in this comprehensive two-hour lecture by Xavier Bresson. Begin with traditional convolutional neural network architecture and convolution before extending to the graph domain. Delve into graph characteristics, define graph convolution, and introduce spectral graph convolutional neural networks. Learn about spectral convolution implementation, spatial networks, and various GCN architectures. Examine the pros and cons of different approaches, experiments, benchmarks, and applications. Cover topics including spectral GCNs, template matching, isotropic and anisotropic GCNs, and conclude with insights on the field's current state and future directions.
Overview
Syllabus
– Week 13 – Lecture
– Architecture of Traditional ConvNets
– Convolution of Traditional ConvNets
– Spectral Convolution
– Spectral GCNs
– Template Matching, Isotropic GCNs and Benchmarking GNNs
– Anisotropic GCNs and Conclusion
Taught by
Alfredo Canziani
