Section: New Results

Evolving CRNs from data time series

Participants : Elisabeth Degrand, François Fages, Jérémy Grignard [former&future Member] , Mathieu Hemery, Sylvain Soliman.

Another approach to CRN design is by evolutionary algorithms. Given a function given by its graph with a finite set of points, and using the same framework based on PIVPs as above, we have designed a genetic algorithm which interleaves the evolution of a population of PIVPs with parameter optimization using CMA-ES for fitting the input curve. On the cosine function, this algorithm recovers PIVPs equivalent to the standard PIVP for cosine, while on Heavyside functions, the algorithm finds (mathematically mysterious) CRNs that are much simpler than Hill functions of high order for instance.