Dropout as a Bayesian approximation: Representing model uncertainty in deep learning diman zad tootaghaj