From Simple Reinforcement Learning with Tensorflow Part 0: Q-Learning with Tables and Neural Networks by Arthur Juliani

…to solve the FrozenLake problem, it turns out it doesn’t do so quite as efficiently as the Q-Table. While neural networks allow for greater flexibility, they do so at the cost of stability when it comes to Q-Learning. There are a number of possible extensions to our simple Q-Network which allow for greater performan…