TODO: Extend neural network implementation to work also with e.g. input from images, sounds or binary