Adversarial Domain Adaptation for Stable Brain-Machine Interfaces