Evolving Neural Controllers

A Tutorial Presented By Nicolas Oros

Neural networks can be pretty hard to design or tune. This is why many researchers use optimization techniques like evolutionary computation to help them design such systems.

In this tutorial, I will explain how a genetic algorithm works, and introduce the different levels of artificial evolution (evolving connectivity, weights, learning rules ...). I will also describe the different ways to encode the evolvable parameters by showing examples from different models created in the field of evolutionary robotics.