Multiple Source Domain Adaptation with Adversarial Learning