Reducing Random Fluctuations in Mutative Self-adaptation
A simple method of reducing random fluctuations experienced in step-size control under mutative self-adaptation is discussed. The approach taken does not require collective learning from the population, i.e. no recombination. It also does not require knowledge about the instantiations of the actual random mutation performed on the object variables. The method presented may be interpreted as an exponential recency-weighted average of trial strategy parameters sampled by a lineage.
KeywordsSphere Model Strategy Parameter Object Variable Collective Learning Canonical Approach
Unable to display preview. Download preview PDF.