A survey on Bayesian nonparametric learning for time series analysis

Abstract

Time series analysis aims to understand underlying patterns and relationships in data to inform decision-making. As time series data are becoming more widely available across a variety of academic disciplines, time series analysis has become a rapidly growing field. In particular, Bayesian nonparametric (BNP) methods are gaining traction for their power and flexibility in modeling, predicting, and extracting meaningful information from time series data. The utility of BNP methods lies in their ability to encode prior information and represent complex patterns in the data without imposing strong assumptions about the underlying distribution or functional form. BNP methods for time series analysis can be applied to a breadth of problems, including anomaly detection, noise density estimation, and time series clustering. This work presents a comprehensive survey of the existing literature on BNP methods for time series analysis. Various temporal BNP models are discussed along with notable applications and possible approaches for inference. This work also highlights current research trends in the field and potential avenues for further development and exploration.