Exponentially Converging Distributed Gradient Descent with Intermittent Communication via Hybrid Methods