Abstract
The rapid progress in hardware technology has made it possible for many sensors to capture multiple measurements simultaneously, leading ultimately to spatio-temporal tensor streams. This work develops a framework for detecting anomalies in data with tensor structures which make traditional matrix-based spectral methods for anomaly detection inadequate for such data. An anomaly is defined as an observation that deviates significantly from the local or global distribution of a given system. Identification of such anomalous spatial locations using all the information obtained from the multiple measurements while preserving the underlying correlation structure of the measurements is the main goal of the applications relate to the topic. This work makes two fundamental contributions. First, it proposes a novel framework that detects anomalies in spatio-temporal tensor streams. Second, it provides effective data visualisation methods for spatio-temporal tensor data. The wide applicability and usefulness of this proposed framework will be demonstrated using various synthetic and real world datasets. This framework is implemented in the open source R package mask.
Priyanga Dilini Talagala
PhD in Statistics
My research interests include Computational Statistics, Anomaly Detection, Time Series Analysis and Machine Learning.