Machine Learning: Auto-encoders for Monitoring

The Weizmann Institute is a major collaborator in building the Muon detectors in ATLAS. IN particular, Weizmann developed the technique of the Thin Gap Chambers (TGC) which is now used for the upgraded detector. Our group is developing autoencoders to detect anomalies in detector monitoring patterns. We hope to achieve an automatic detection of detectors malfunctions, so the detectors can be fixed even before installation. Autoencoders is a standard Machine learning tool to detect anomalies. The network is trying to duplicate the detector input (trained with healthy detectors). The real DATA output is compared with the expected one, and if the detector has an anomalous output, the autoencoder will spot it immediately.