Title: A simple guide from Machine Learning outputs to statistical criteria

Abstract: In this paper we propose ways to incorporate Machine Learning training outputs into a study of statistical significance. We describe these methods in supervised classification tasks using a CNN and a DNN output, and unsupervised learning based on a VAE. As use cases, we consider two physical situations where Machine Learning are often used: high-$p_T$ hadronic activity, and boosted Higgs in association with a massive vector boson.