Towards Explainable Generative Adversarial Networks