No Free Lunch Theorems for Search

摘要: We show that all algorithms search for an extremum of a cost function perform exactly the same, when averaged over possible functions. In particular, if algorithm A outperforms B on some functions, then loosely speaking there must exist as many other functions where A. Starting from this we analyze number priori characteristics problem, like its geometry and information-theoretic aspects. This analysis allows us to derive mathematical benchmarks assessing particular algorithm's performance. also investigate minimax aspects validity using partial predict future behavior function, time-varying conclude with discussion justifiablility biologically-inspired methods.