Learning Graph Augmentations to Learn Graph Representations

Kaveh Hassani, Amir Hosein Khasahmadi

Abstract: Devising augmentations for graph contrastive learning is challenging due to their irregular structure, and drastic distribution shifts and nonequivalent feature spaces across datasets. To address this, we propose LG2AR, Learning Graph Augmentations to Learn Graph Representations, which is an end-to-end automatic graph augmentation framework that helps encoders learn generalizable representations on both node and graph levels. LG2AR consists of a probabilistic policy that learns a distribution over augmentations and a set of probabilistic augmentation heads that learn distributions over augmentation parameters. Under linear evaluation protocol, LG2AR achieves state-of-the-art results on 8 out of 8 graph classification tasks and 6 out of 7 node classification benchmarks.

One-sentence Summary: We propose LG2AR, the first fully-automated end-to-end graph contrastive learning framework

Community Implementations: [ 1 code implementation](https://www.catalyzex.com/paper/arxiv:2201.09830/code)