Representation learning using graph autoencoders with residual connections

Figure 1. End-to-end framework of ResVGAE: given two distant nodes X and Y, the model predicts a link between them. Original figure from the arXiv HTML version of the paper.

Abstract

Graph autoencoders are efficient at embedding graph-based data sets. Most graph autoencoder architectures have shallow depths which limits their ability to capture meaningful relations between nodes separated by multi-hops. In this paper, we propose Residual Variational Graph Autoencoder, ResVGAE, a deep variational graph autoencoder model with multiple residual modules. We show that our multiple residual modules, a convolutional layer with residual connection, improve the average precision of the graph autoencoders. Experimental results suggest that our proposed model with residual modules outperforms the models without residual modules and achieves similar results when compared with other state-of-the-art methods.

The arXiv record was revised in 2022 under the title “ResVGAE: Going Deeper with Residual Modules for Link Prediction”. This page uses the title and author order as listed by the author, with the current arXiv abstract above.

Framework

Residual modules

A convolutional layer with a residual connection is added to the graph autoencoder, letting it go deeper without losing accuracy.

Deep variational autoencoder

ResVGAE stacks multiple residual modules in a deep variational graph autoencoder.

Link prediction

The model is evaluated on link prediction, where it outperforms the same models without residual modules.

Resources

Preprint: arXiv:2105.00695
DOI: 10.48550/arXiv.2105.00695

BibTeX

@misc{nallbani2021representation,
  title = {Representation learning using graph autoencoders with residual connections},
  author = {Nallbani, Indrit and Ayanzadeh, Aydin and Keser, Reyhan Kevser and {\c{C}}al{\i}k, Nurullah and T{\"o}reyin, Beh{\c{c}}et U{\u{g}}ur},
  year = {2021},
  eprint = {2105.00695},
  archivePrefix = {arXiv},
  primaryClass = {cs.LG},
  doi = {10.48550/arXiv.2105.00695},
  url = {https://arxiv.org/abs/2105.00695}
}