Comment by michaelbarton

This reminds me of an an approach in mcmc where you run mutiple chains at different temperatures and then share the results between them (replica exchange MCMC sampling) the goal being not to get stuck in one “solution”