Hierarchical Temporal Memory Method for Time-Series-Based Anomaly Detection

Abstract

Time-series-based anomaly detection is a quite important field that has been researched over years. Many techniques have been developed and applied successfully for certain application domains. However, there are still some challenges, such as continuously learning, tolerance to noise and generalization. This paper present Hierarchical Temporal Memory, a novel biological neural network, to time-series-based anomaly detection. HTM is able to learn the changing pattern of the data and incorporate contextual information from the past to make accurate prediction. We have evaluated HTM on real and artificial datasets. The experiment results show that HTM can successfully discover anomalies in time-series data.