Learning Discrete Structures for Graph Neural Networks

Time: Monday 27-May-2019 22:30 (This is a past event.)

Discussion Facilitators:


Motivation / Abstract
Graph networks rely on the graph structure existing before hand which is not always the case. Here the authors present a novel method for learning the graph structure (using bilevel programming) simultaneously with training a GCN on the resulting graph. 
Stream Categories:
 Main Stream