Title: Efficient offline nonparametric changepoint detection for univariate data Authors:  Dean Bodenham - Imperial College London (United Kingdom) [presenting]
Abstract: A novel offline changepoint detection method is introduced for identifying multiple changepoints in a univariate time series. This approach uses a popular nonparametric two-sample test combined with data structures from computer science to achieve $O(n\log n)$ computational complexity for identifying a single change. Current approaches to this problem tend to be either parametric, requiring distributional assumptions about the data, or nonparametric with a higher computational cost. Empirical results are presented comparing our approach to well-known changepoint detection methods in a variety of scenarios.