abojchevski/graph2gauss

Implements unsupervised inductive node embedding via a ranking-based objective that learns probabilistic Gaussian representations rather than point embeddings, enabling uncertainty quantification. Built in TensorFlow with support for attributed graphs or one-hot/adjacency-based variants when node features are unavailable. The approach leverages graph structure and node attributes through a deep architecture that optimizes pairwise ranking losses to capture both local and global graph properties.

179 stars. No commits in the last 6 months.