Asynchronous Batch Bayesian Optimisation with Improved Local Penalisation

Batch Bayesian optimisation (BO) has been successfully applied to tuning using parallel computing, but it is wasteful of: workers that complete jobs ahead of others are left idle. We address problem by developing an approach, Penalising Locally for Asynchronous Optimisation on k workers (PLAyBOOK), for asynchronous parallel BO. demonstrate empirically the efficacy of PLAyBOOK and its variants on tasks and a real-world problem. We undertake a comparison between and asynchronous BO, and show that asynchronous BO often synchronous batch BO in both wall-clock time and number of function.