A robust estimator of mutual information for deep learning interpretability