Graph Auto-Encoders For Adajcency matrix and Node feature reconstruction
In this project I used T Kipf's graph auto-encoder (VAE) version to do node reconstruction as well as adjacency matrix recovering. The encoder is consisted of 2 layers of graph convolutional layer, and the decoder first generates the features, then the adjacency matrix. The model achieves great accuracy for a simple two layers encoding-two layers decoding autoencoder.
For more information on the Graph Auto-Encoder, please see Thomas Kipf's original github page for graph auto-encoders. To train the model, simply run
python train.pyReference:
@article{kipf2016variational,
title={Variational Graph Auto-Encoders},
author={Kipf, Thomas N and Welling, Max},
journal={NIPS Workshop on Bayesian Deep Learning},
year={2016}
}