Optimal Learning from Verified Training Data