Deep anomaly detection using self-supervised learning: application to time series of cellular data

International audienceWe present a deep self-supervised method for anomaly detection on time series. We apply this methodology to detect anomalies from cellular times series, in particular cell dry mass, obtained in the context of lens-free microscopy.We propose an innovative, self-supervised, two-step method for anomaly detection on time series. As a first step, a representation of the time series is learned thanks to a 1D-convolutional neural network without any labels. Then, the learnedrepresentation is used to feed a threshold anomaly detector. This new self-supervised learning method is tested on an unlabeleddataset of 9100 time series of dry mass and succeeded in detecting abnormal time series with a precision of 96.6%