Efficient evolution of neural networks through complexification

Artificial neural networks can potentially control autonomous robots, vehicles, factories, or game players more robustly than traditional approaches. Neuroevolution, i.e. the artificial evolution of neural networks, is a method for finding the right topology and connection weights to specify the desired control behavior. The challenge for neuroevolution is that difficult tasks may require complex networks with many connections, all of which must be set to the right values. Even if a network exists that can solve the task, evolution may not be able to find it in such a high-dimensional search space. This dissertation presents the NeuroEvolution of Augmenting Topologies (NEAT) method, which makes search for complex solutions feasible. In a process called complexification, NEAT begins by searching in a space of simple networks, and gradually makes them more complex as the search progresses. By starting minimally, NEAT is more likely to find efficient and robust solutions than neuroevolution methods that begin with large fixed or randomized topologies; by elaborating on existing solutions, it can gradually construct even highly complex solutions. In this dissertation, NEAT is first shown faster than traditional approaches on a challenging reinforcement learning benchmark task. Second, by building on existing structure, it is shown to maintain an “arms race” even in open-ended coevolution. Third, NEAT is used to successfully discover complex behavior in three challenging domains: the game of Go, an automobile warning system, and a real-time interactive video game. Experimental results in these domains demonstrate that NEAT makes entirely new applications of machine learning possible.