EAs are inspired by biological evolution and implement techniques and mechanisms that simulate biological evolution. This means that EAs go through many trials to create a population of new candidate solutions. These solutions are also called individuals (in RL problems, a candidate solution is a policy) that are better than the previous generation, in a similar way to the process within nature wherein only the strongest survive and have the possibility to procreate.
One of the advantages of EAs is that they are derivative-free methods, meaning that they don't use the derivative to find the solution. This allows EAs to work very well with all sorts of differentiable and non-differentiable functions, including deep neural ...
