Research Seminar on AI: Graph Learning with 1D Convolutions on Random Walks

 

Dienstag, 22.02.2022, 16.00 Uhr

Graphs are a very general form of data that naturally occurs across a wide range of applications, from cheminformatics to social network analysis. Due to this ubiquity, Machine Learning on graph-structured data is both crucial and challenging. Jan Tönshoff and his team propose CRaWl (CNNs for Random Walks), a novel neural network architecture for graph learning. It is based on processing sequences of small subgraphs induced by random walks with standard 1D CNNs. Thus, CRaWl is fundamentally different from typical message passing graph neural network architectures. It is inspired by techniques counting small subgraphs, such as the graphlet kernel and motif counting, and combines them with random walk based techniques in a highly efficient and scalable neural architecture. CRaWl is not constraint by the well known limitations of the standard message passing framework, which currently dominates graph learning. In particular, it can detect arbitrary substructures up to a chosen window size. Jan Tönshoff and his team demonstrate empirically that CRaWl matches or outperforms state-of-the-art GNN architectures across a multitude of benchmark datasets for graph learning.

Jan Tönshoff is a PhD student at the Chair for Computer Science 7 of the RWTH Aachen, where he works for Prof. Martin Grohe. His research focuses on applying machine learning to relational data, such as graphs and databases.
Applications of these techniques span from physics and chemistry to social data science and combinatorial optimization.