Combining multi-fidelity modelling and asynchronous batch Bayesian Optimization

Bayesian Optimization is a useful tool for experiment design. Unfortunately, the classical, sequential setting of does not translate well into laboratory experiments, instance battery design, where measurements may come from different sources and their evaluations require significant waiting times. Multi-fidelity addresses with sources. Asynchronous batch provides framework to select new experiments before results prior are revealed. This paper proposes an algorithm combining multi-fidelity asynchronous methods. We empirically study behaviour, show it can outperform single-fidelity methods As application, we consider designing electrode materials optimal performance in pouch cells using coin approximate performance.