Continuous Latent Spaces Sampling for Graph Autoencoder

This paper proposes colaGAE, a self-supervised learning framework for graph-structured data. While graph autoencoders (GAEs) commonly use reconstruction as pretext task, this simple approach often yields poor model performance. To address issue, colaGAE employs mutual isomorphism task continuous latent space sampling GAE (colaGAE). The central idea of is to sample from multiple views in the and reconstruct structure, with significant improvements terms model’s training difficulty. investigate whether can enhance GAEs’ representations, we provide both theoretical empirical evidence benefits task. Theoretically, prove that offer respect difficulty training, leading better Empirically, conduct extensive experiments on eight benchmark datasets achieve four state-of-the-art (SOTA) results; average accuracy rate experiences notable enhancement 0.3%, demonstrating superiority node classification tasks.