The approach we present is based on Artificial Neural Networks (ANNs) which previously have been successfully applied in many contexts such as Image recognition and Natural Language Processing.
The first article in this series contained a high-level introduction to ANNs and this second article builds on that to describe how such networks can be used as a substitute to the more established methods when exposed to a problem that requires nested simulations. We specifically address what to consider when calibrating the model and how to approach the training process.