Invited talk on GNNs and Bayesian DL in computational science

Abstract

I presented my approach to UQ and ML when dealing with large-volumes of high-dimensional data that contain some structure. I briefly presented the supervised learning work on crack localization with arbitrarily positioned strain-sensors and parts of the work on building supervised/semi-supervised graph-structured latent variable models with the relational VAE. Moreover, I presented some results that appear in my PhD work using trained RVAEs for selecting optimal sub-sets of turbines for monitoring.

Charilaos Mylonas

Data/Computational Scientist, Consultant

Data Scientist