Genetic algorithms

Another popular class of black-box methods is genetic algorithms (GAs). It is a large family of optimization methods with more than two decades of history behind it and a simple core idea of generating a population of N individuals (concrete model parameters), each of which is evaluated with the fitness function. Then, some subset of top performers is used to produce the next generation of the population (this process is called mutation). This process is repeated until we’re satisfied with the performance of our population.

There are a lot of different methods in the GA family, for example, how to perform the mutation of the individuals for the next generation or how to rank the performers. Here, we will consider the simple GA method with some extensions, published in the paper by Such et al., called Deep neuroevolution: Genetic algorithms are a competitive alternative for training deep neural networks for reinforcement learning [Suc+17].

In...