Hands-On Neuroevolution with Python

Speciation

In the evolutionary process, organisms can create diverse topologies through generations, but they fail to produce and maintain topological innovations of their own. The smaller network structures optimize faster than larger ones, which artificially reduces the chances of survival of a descendant genome after adding a new node or connection to it. Thus, the freshly augmented topologies experience negative evolutionary pressure due to the temporary decrease of fitness of the organisms within the population. At the same time, novel topologies can introduce innovations that lead to a winning solution in the long run. To address the temporal drop of fitness, the concept of speciation was introduced in the NEAT algorithm. The speciation ...