On Decentralizing Selection Algorithms
Kenneth Alan De Jong, Jayshree Sarma · 1995
The increasing availability of parallel computing architectures provides an opportunity to exploit this power as we scale up evolutionary algorithms (EAs) to solve more complex problems. To effectively exploit fine grained parallel architectures, the control structure of an EA must be decentralized. This is difficult to achieve without also changing the semantics of the selection algorithm used, which in turn generally produces changes in an EA's problem solving behavior. In this paper we analyze the implications of various decentralized selection algorithms by studying the changes they produce on the characteristics of the selection pressure they induce on the entire population. This approach has resulted in significant insight into the importance of selection variance and local elitism in designing effective distributed selection algorithms. 1 INTRODUCTION One of the frequently stated virtues of evolutionary algorithms (EAs) is their "natural" parallelism. The increasing availabili...