An Advanced Evolution Should Not Repeat its Past Errors

A safe control of genetic evolution consists in preventing past errors of evolution from being repeated. This could be done by keeping track of the history of evolution, but maintaining and exploiting the complete history is intractable. This paper investigates the use of machine learning (ML), in order to extract manageable information from this history. More precisely , induction from examples of past trials and errors provides rules discriminating errors from successful trials. Such rules allow to a priori estimate the desirability of future trials; this knowledge can support powerful control strategies. Several strategies of ML-based control are applied to a genetic algorithm, and tested on the Royal Road, a GA-deceptive, and a com-binatorial optimization problem. Comparing mutation control with crossover control yields unexpected results.