Representation and Learning in Graph Neural Networks
- Stefanie Jegelka (Machine Learning Group at MIT, USA)
Graph Neural Networks (GNNs) have become a popular tool for learning representations of graph-structured inputs, with applications in computational chemistry, recommendation, pharmacy, reasoning, and many other areas. This talk will show recent results on representational power and learning in GNNs. First, we will address representational power and important limitations of popular message passing networks and of some recent extensions of these models. Second, we consider learning, and provide new generalization bounds for GNNs. Third, although many networks may be able to represent a task, some architectures learn it better than others. I will show results that connect the architectural structure and learning behavior, in and out of the training distribution.
This talk is based on joint work with Keyulu Xu, Behrooz Tahmasebi, Jingling Li, Mozhi Zhang, Simon S. Du, Ken-ichi Kawarabayashi, Vikas Garg and Tommi Jaakkola.