Creating Neuroevolution Framework for Tensorflow 2.0, preimplementing 'Neuroevolution of Augmenting Topologies' (NEAT)

Implementing a framework for Neuroevolution in Tensorflow 2.0, providing a variety of preimplemented Neuroevolution algorithms, genomes and environments to test them in. The framework will focus on extensibility and fast prototyping.

During the GSoC project duration will I be able to implement the most prominent Neuroevolution algorithm 'Neuroevolution of Augmenting Topologies' (NEAT) with the prerequisite direct-encoding genome. Though the architecture of the framework will provide such extensibility that the implementation of follow-up research such as 'HyperNEAT', 'EvoCNN' or 'Regularized Evolution' will be possible in a straightforward manner. Implementing this follow-up research will furthermore be pursuit by me in my master thesis following the completion of the GSoC project.

The framework will moreover provide a variety of environments to test and benchmark the Neuroevolution algorithms and genomes in. Among those environments do I plan to implement, or at least support, the XOR problem, the cartpole-balancing problem, various Tensorflow Image classifier datasets and the OpenAI Retro environment.