We are working on an on-device learning algorithm for unsupervised anomaly detection. One of the biggest issues when applying AI to industries is to prepare training data sets beforehand. Our approach learns normal patterns in a deployed environment extemporarily to detect unusual ones, so no prior training is needed.