Feeding genetic heterogeneity via a smart mutation operator in the Memetic Phase Retrieval approach

A memetic algorithm is a stochastic optimization method obtained by hybridizing an evolutionary approach with common deterministic optimization procedures. The recently introduced Memetic Phase Retrieval (MPR) approach exploits this synergy to face the so-called phase retrieval problem in Coherent Diffraction Imaging (CDI). Here we focus on the development of a smart mutation genetic operator; our aim is the improvement of MPR performance by continually feeding with relevant information the genetic heritage of the population of candidate solutions. Remarkably, statistical tests on synthetic CDI data performed using MPR enhanced via a smart mutation operator reveal a smaller reconstruction error with respect to an MPR implementation supplied with a blind random mutation only.