28 novembre 2022, 14:00-15:00.
ENS, S16.
Detection of Anomalies and Identification of their Precursors in Large Data Series Collections
Extensive collections of data series are becoming a reality in many scientific and social domains, such as environmental sciences, astrophysics, neurosciences, and engineering. Informally, a data series is an ordered sequence of points or values. Once these series are collected and available, users often need to query them. These queries can be simple, such as selecting time intervals, but also complex, such as detecting anomalies, often synonymous with malfunctioning systems or failures. This last type of analysis represents a crucial problem for applications in a wide range of domains, all sharing the same objective: to detect anomalies as soon as possible to avoid critical events. Therefore, in this talk, we discuss the following two objectives: (i) retrospective unsupervised subsequence anomaly detection in data series. (ii) classification explanation of known anomalies in data series to identify possible precursors. We will first introduce the fundamental definitions and taxonomy of data series. We then present two novel methods along the two axes mentioned above. Finally, we illustrate the applicability of our developments through a use case about identifying undesirable vibration precursors occurring in water supply pumps in french electrical power plants.
