Using time efficiently: genetic-evolutionary algorithms and the continuation problem
David E. Goldberg · 1999
This paper develops a macro-level theory of efficient time utilization for genetic and evolutionary algorithms. Building on population sizing results that estimate the critical relationship between solution quality and time, the paper considers the tradeoff between large populations that converge in a single convergence epoch and smaller populations with multiple epochs. Two models suggest a link between the salience structure of a problem and the appropriate population-time configuration for best efficiency. 1 INTRODUCTION Great strides have been made in our understanding and design of a variety of diversity prolonging or rejuvenating operators. Such diverse mechanisms as adaptive and self-adaptive mutation operators (Back & Schwefel, 1995), dominance and diploidy (Goldberg & Smith, 1987; Smith, 1987), and even linkage learning (Harik & Goldberg, 1997; Harik, 1997) have been used to prolong or rejuvenate diversity in the face of continued selection and the vagaries of genetic drift. ...