********************************************************************** * * * Einladung * * * * Informatik-Kolloquium * * * ********************************************************************** Zeit: Dienstag, 15. Juni 2021, 14:00 Uhr Ort: https://rwth.zoom.us/j/94452318075?pwd=WTd0eUVPNmlVaTUzUDlvWWVvNHFlQT09 Meeting ID: 944 5231 8075 Passcode: 402666 ********************************************************************** Vortragender: Prof. Stephan Günnemann, TUM Titel: Machine Learning for Graphs: Expressiveness, Robustness, and Scalability Abstract: Graphs are a universal language to model complex systems and are omnipresent in the natural sciences (protein interaction networks, molecular graphs), engineering sciences (IoT networks, smart grids), social sciences (social networks), and many more. In recent years, graph neural networks (GNNs) have started to unlock the potential of machine learning for the graph domain -- yet, our understanding of these principles is still limited. In my talk, I will address some of the core challenges which hinder the use of GNNs in scientific and industrial application domains. Specifically, I will highlight (i) the limited expressiveness of standard GNNs and ways to enhance them by taking domain knowledge into account, (ii) the GNNs’ robustness properties and principles to ensure their reliability, making them suitable for even sensitive and safety-critical domains, and (iii) their lack in scalability along with approaches enabling efficient learning.