Learning Graph Augmentations to Learn Graph Representations
This addresses the problem of learning generalizable graph representations for machine learning practitioners, offering a novel method for a known bottleneck in graph contrastive learning.
The paper tackles the challenge of devising augmentations for graph contrastive learning by introducing LG2AR, an end-to-end automatic graph augmentation framework, which achieves state-of-the-art results on 18 out of 20 graph-level and node-level benchmarks compared to previous unsupervised models.
Devising augmentations for graph contrastive learning is challenging due to their irregular structure, drastic distribution shifts, and nonequivalent feature spaces across datasets. We introduce 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. We show that LG2AR achieves state-of-the-art results on 18 out of 20 graph-level and node-level benchmarks compared to previous unsupervised models under both linear and semi-supervised evaluation protocols. The source code will be released here: https://github.com/kavehhassani/lg2ar