Abstract:

Bayesian optimization (BO) has been widely adopted for optimizing closed-loop performance in data-limited settings, especially for systems with unmodeled dynamics or performance functions. The BO algorithm efficiently trades-off exploration and exploitation by leveraging uncertainty estimates using surrogate models learned using data from the target closed-loop system to be optimized. The convergence rate of BO can be greatly improved if the underlying surrogate model can accurately predict the target system performance despite having very limited data. In this paper, we propose the use of Bayesian meta-learning to generate an initial surrogate model based on data collected from closed-loop performance optimization tasks performed on a variety of systems that are different to the target system. To this end, we employ deep kernel networks (DKNs) which are simple to train and which comprise encoded Gaussian process models that integrate seamlessly with classical BO. The effectiveness of our proposed DKN-BO approach for speeding up closed-loop performance optimization is demonstrated using a well-studied uncertain nonlinear system with unknown dynamics and an unmodeled performance function.