Graph neural networks and covering spacesGraduate Colloquium
by
Endenicher Allee 60/1-016 - Lipschitzsaal
Mathezentrum
I would like to give a brief overview of some deep learning techniques, their applications, and their limitations. We will focus particularly on how covering spaces are relevant to understanding limitations of conventional neural networks in determining isomorphism classes of graphs (or in particular graph neural networks). A number of works presented in this talk are based on joint collaborations with Yun Young Choi, U Jin Choi, Dosang Joe, Minho Lee, Seunghwan Lee, Joohwan Ko, and Youngho Woo. My hope is to make the talk as accessible as possible, even for those who do not have prior knowledge in deep learning techniques.
Website of the Hausdorff Colloquium
Michel Alexis, Regula Krapf, Fred Lin, Christoph Thiele