2 July, 20192 July, 2019 NitrakUncategorized Applying Procedurally Generated Zelda Levels from Generative Adversarial Networks to improve Deep Reinforcement Learning through Generalisation