TU Wien CAIML

CAIML Seminar: “Expressive Graph Embeddings via Homomorphism Counts”

Pascal Welke discusses graph learning and homomorphism counting to understand and enhance the expressivity and capabilities of graph neural networks (GNNs).

Pascal Welke
Pascal Welke

On This Page

CAIML Seminar with Pascal Welke took place on November 25, 2024 in EI8 Pötzl Hörsaal.

Abstract

Graphs can naturally model complex systems such as chemical molecules or social networks. However, their complex and irregular structure makes learning from graphs a challenging and fascinating area of research. In this talk, I explore the intersection of graph representation learning and homomorphism counting—a technique to measure how frequently specific patterns occur in graphs. Homomorphism counting has emerged as a powerful tool in the study and development of graph neural networks (GNNs). I focus on two areas: (1) Understanding and quantifying what GNNs can do and where they fail and (2) improving the capabilities of GNNs. Special interest in both these areas lies on the expressivity of GNNs, i.e., their ability to learn different representations for nonisomorphic graphs. I present and discuss the basics, as well as recent results in this interesting and active area of research.

Slides

Here you can download the slides:
Download the slides.