monte carlo dropout

A method that allows uncertainty predictions in deep learning models by using dropout at test time and running the model several times.