Abstract: Neural Architecture Search

by Yash Mehta and Arber Zela

In this hands-on tutorial, we will go through the basics of neural architecture search (NAS) with NASLib, which is an open-source library developed from the AutoML group in Freiburg to facilitate and unify NAS research. We will introduce the basic building blocks of NAS: search spaces, optimization strategies, performance predictors and evaluators, and provide examples how these can be implemented in NASLib and how to run the NAS procedure in a few lines of code.