DOS-GAN: A Distributed Over-Sampling Method Based on Generative Adversarial Networks for Distributed Class-Imbalance Learning.