Instance Discrimination for Link Prediction
Researchers have adapted instance discrimination models, typically used for self-supervised learning in computer vision, for link prediction tasks in graph domains. Their evaluation showed that augmentation strategies significantly impact performance, similar to image-based methods. The study introduces two novel models, L-GRACE and L-BGRL, which focus on link representations rather than node representations, achieving state-of-the-art results, particularly on unattributed graphs. AI
IMPACT Introduces novel methods for link prediction in graphs, potentially improving performance in areas like recommendation systems and network analysis.