Tracker DQM Machine Learning studies for data certification

At CMS, the procedure of data certification ensures the good functioning of the detector and hence the validity of the data for physics analysis. Currently, the certification is performed by checking manually a large amount of distributions. In this note, we present ongoing developments with machine learning to automate anomaly detection and assist human beings in the certification process. In particular, the uses of non-negative matrix factorization, of auto-encoders, and of residual networks to analyse signals in the pixel tracker are investigated, such as cluster charge, cluster size, and occupancy in the different layers of the barrel pixel.