Temporal convolutional network-based time-series segmentation

Abstract

Time-series segmentation is a useful tool for identifying the underlying characteristics of time series and summarizing time series as a sequence of states. Partitioning time series into the states makes the complex time series easily understandable and interpretable. However, as labels for time points are not normally available, it is challenging to figure out the accurate segments and their states. Therefore, we propose an unsupervised time-series segmentation using properties inherent in times series. The states can be characterized by diverse length patterns inherent in time series, and thus identifying and capturing diverse patterns are crucial in an unsupervised time-series segmentation. We adopt a temporal convolutional network (TCN) as our key component to learn diverse length patterns since the intermediate layers in TCN contain both short and long patterns hierarchically from the first to the last layer. In this thesis, we propose a novel unsupervised time-series segmentation TCTS, which is featured with the joint optimization of two modules, TCN-based pattern learning and clustering-based classification. TCN-based pattern learning targets to grasp diverse length patterns that are characterized differently by the states, while the clustering-based classification improves the separability of the representations between the states. We conduct experiments by comparing several baselines with multiple datasets and demonstrate the superiority of TCTS.