Systems | Information | Learning | Optimization

The World of Graph Neural Networks: From the Mystery of Generalization to Foundational Limitations

Speaker: Gitta Kutyniok

The tremendous importance of graph structured data due torecommender systems or social networks led to the introduction ofgraph neural networks (GNNs). After a general introduction to GNNs,we will discuss results about their amazing generalizationcapabilities. We will study the more specialized question to whichextent GNNs are able to generalize to graphs, which describe asimilar phenomenon as present in the training data set, as well asthe fully general problem. We will present results for both messagepassing and spectral GNNs. We will finish with a word of cautionwhen training GNNs on classical digital hardware, and presentfundamental limitations.

May 11 @ 12:30
12:30 pm (1h)

Orchard View Room, Virtual

Gitta Kutyniok