Machine Learning with Adversaries: Byzantine Tolerant Gradient Descent