Application of gravity vectors and moving vectors for the acceleration of both differential evolution and interactive differential evolution

We propose and evaluate two methods for accelerating differential evolution and interactive differential evolution (IDE). The first acceleration method, which we call DE/gravity, aims to realize performance similar to that of paired-comparison-based IDE/best while removing the requirement that the IDE user must choose the best individual among all displayed individuals. The second acceleration method generates not only a conventional trial vector but also a second and third trial vector. It calculates a moving average vector, X _moving, for the population between generations, and compares a given target vector with the three trial vectors of a conventional trial vector, a target vector + X_moving, and a trial vector + X_moving, and uses the best one among the four vectors as offspring in the next generation. We evaluate these acceleration methods and a conventional method by applying them to Gaussian mixture models and demonstrate the effectiveness of our proposed methods.