Open-Ended Reinforcement Learning with Neural Reward Functions Robert Meier

We propose a different approach that uses reward functions encoded by neural networks. These are trained iteratively to reward more complex behavior.