NEAT with Hindsight Experience Replay

After implementing NEAT in Unity a few weeks ago, I tried to think of ways to improve its performance.

I remember reading about a novel technique used in reinforcement learning algorithms known as Hindsight Experience Replay. With this technique, failed states that an agent reached are tracked and learned from as if they were successes.

At first, I didn’t think much of it. Due to the speciation method NEAT employs, one population is capable of searching a domain in …

genetic-algorithm machine learning