Automating the analysis of time series data

Many crucial real-world problems are characterized by the presence of time series. These are harder to work with from an analysis perspective for a variety of reasons such as their ordered nature, long-range dependencies, and interactions among multiple series that are collected simultaneously. Beyond these, they pose two other big challenges. First, most machine learning methods are not directly applicable to time series meaning that time series have to be manipulated into a form that is amenable for analysis. Second, evaluation is much trickier for these problems. Alas, these two challenges force a lot of (repetitive) work onto data scientists. The goal of this Ph.D. to explore how to partially automated these tasks to better support data scientists.