Massively Parallel Genetic Optimization Through Asynchronous Propagation of Populations

We present Propulate, an evolutionary optimization algorithm and software package for global in particular hyperparameter search. For efficient use of HPC resources, Propulate omits the synchronization after each generation as done conventional genetic algorithms. Instead, it steers search with complete population at time breeding new individuals. provide MPI-based implementation our algorithm, which features variants selection, mutation, crossover, migration is easy to extend custom functionality. compare established tool Optuna. find that up three orders magnitude faster without sacrificing solution accuracy, demonstrating efficiency efficacy lazy approach. Code documentation are available https://github.com/Helmholtz-AI-Energy/propulate