An Asynchronous Parallel Evolution Strategy

This work introduces a new asynchronous parallel self-adaptive evolution strategy. An asynchronous scheme is chosen because the load on the processors is usually unbalanced. The proposed algorithm is nonblocking leaving no processor idle at any given time. The motivation and theory for the algorithm is based on the author’s previous analysis of continuous evolution strategies [9] and the reduction of random fluctuation in mutative self-adaptation [7]. The algorithm is implemented in C++ using Posix threads and tested on a number of unimodal and multimodal benchmark problems. The result is an efficient and effective numerical optimization strategy for multi-processor machines.