Efficient Hyperparameter Optimization for Deep Learning Algorithms Using Deterministic RBF Surrogates