Using multi-objective genetic programming to evolve stochastic logic gate circuits

Abstract

A new stochastic logic gate language is presented. Blossey et al.'s stochastic gene gate language is extended with a complete set of stochastic Boolean gates. Although the gates have behavioural similarities to conventional logic gates, a major difference is that they operate on quantities of products or substances that dynamically vary over time. A gene gate circuit's behaviour is characterized by a time-course plot of the substance quantities. The paper studies the Boolean gate language by using multi-objective genetic programming to evolve logic gate circuits that conform to a number of different target systems. Circuit behaviour is characterized by sets of up to 15 time course statistics, and sum of ranks is used as a many-objective scoring strategy. Results show that the language is highly compositional, just like conventional logic expressions, and that multiple circuits can exhibit similar behaviours. The new gate language uses Blossey et al.'s gates as a rudimentary basis within evolved circuits, with the advantage of using higher-level Boolean gates when necessary. The identification of candidate solutions can be challenging, however, and must account for noise inherent in the time course behaviours. Circuit behaviour is also highly dependent on channel rates, and future work applying the language to real-world data will need to address this sensitivity.