Mutation Rate Modulation for Faster GAs

Expected Results

Faster convergence rates.

The GA is used on the population of rate-strings to provide selection pressures that require the rates to provide faster convergences for their dummy populations. From this it is expected that the use of a time-varying mutation rate will create a convergences that are at least as fast as the fixed mutation-rates.

There an exception to this is possible, though: if the mutation rates cease to matter at any point in the convergence of the strings, selection pressures will mean nothing since the rate strings won't be able to affect the speed of convergence of their buddy populations.

Increased mutation rates with increased dummy-population convergence.

Given that the election operator should afford a greater exploration rate when the dummy strings are more converged, it is to be expected that the optimal mutation rate will be greater for more converged populations. That is to say that the latter bytes of the string will have larger values than the beginning byte.

Increased mutation-rates with increasing fitness function discontinuity.

Since a greater exploration rate pays greater dividents in a situation where the optima are elusive and not necessarily found by inductive methods, it is expected that the more discontinuous fitness function will show a greater average mutation rate.