A0897
Title: Fast online change point detection for high dimensional regression models
Authors: Xianru Wang - Southwestern University of Finance and Economics (China) [presenting]
Abstract: The aim is to present a fast online algorithm for detecting change points in high-dimensional regression models, a challenging task due to the lack of direct observations of the parameter of interest. A novel test statistic-based algorithm is introduced designed to operate in an online fashion, ensuring that storage requirements and computational complexity per new observation remain independent of the number of previous observations. A key innovation of the approach is the departure from traditional moving window-based methods, which are limited by a fixed window size that can negatively affect detection performance. To address this, an online change point detection method is proposed that evaluates all possible window sizes when constructing the test statistics. Although this increases computational complexity, an efficient algorithm is developed that can compute the optimal window size. Theoretical results are provided, showing that the average run length has a lower bound and that the detection delay is bounded from above, ensuring the reliability and effectiveness of the proposed method.
