Whom to Trust? Elective Learning for Distributed Gaussian Process Regression