The evolutionary-gradient-search procedure in theory and practice

Abstract

The pertinent literature controversially discusses in which respects evolutionary algorithms differ from classical gradient methods. This chapter presents a hybrid, called the evolutionary-gradient-search procedure, that uses evolutionary variations to estimate the gradient direction in which it then performs an optimization step. Both standard benchmarks and theoretical analyses suggest that this hybrid yields superior performance. In addition, this chapter presents inverse mutation, a new concept that proves particularly useful in the presence of noise, which is omnipresent in almost any real-world application. © 2009 Springer-Verlag Berlin Heidelberg.