Graph Barlow Twins: A self-supervised representation learning framework for graphs

The self-supervised learning (SSL) paradigm is an essential exploration area, which tries to eliminate the need for expensive data labeling. Despite great success of SSL methods in computer vision and natural language processing, most them employ contrastive objectives that require negative samples, are hard define. This becomes even more challenging case graphs a bottleneck achieving robust representations. To overcome such limitations, we propose framework graph representation — Graph Barlow Twins, utilizes cross-correlation-based loss function instead samples. Moreover, it does not rely on non-symmetric neural network architectures contrast state-of-the-art method BGRL. We show our achieves as competitive results best fully supervised ones while requiring fewer hyperparameters substantially shorter computation time (ca. 30 times faster than BGRL).