Graph Neural Networks through the Lens of Measure Theory
Despite their growing popularity, graph neural networks (GNNs) still suffer from multiple unsolved problems, including lack of embedding expressiveness, propagation of information to distant nodes, and training on large-scale graphs. Understanding the roots of and providing solutions for such problems require developing analytic tools and techniques. In this talk we provide a measure theoretic point of view for the above-mentioned problems, and derive a notion of “recoverability” which will serve us as a tool for GNN embedding analysis, unsupervised graph representation learning and regularization. At the end of the talk, we will show a tight relationship between recoverability loss minimization and mutual information maximization.
M.Sc. student under supervision of Prof. Avi Mendelson and Dr. Chaim Baskin.