Master's thesis presentation. Oliver is advised by Prof. Dr. Felix Dietrich.
Previous talks at the SCCS Colloquium
Oliver Beck: Sampling weights of Graph Neural Networks
SCCS Colloquium |
In this thesis, we extend the SWIM framework to Graph Neural Networks. We evaluate the method on standard benchmark datasets for both graph and node classification tasks, and compare its performance to conventional gradient based training. Our experiments investigate the trade offs between accuracy, runtime, and efficiency, providing insights into the applicability of data-dependent weight sampling methods in GNNs.