A0267
Title: Continuous optimization for offline change point detection and estimation
Authors: Hans Reimann - University of Potsdam (Germany) [presenting]
Sarat Moka - The University of New South Wales (Australia)
Georgy Sofronov - Macquarie University (Australia)
Abstract: The application of novel advances in best subset selection for regression modelling is explored via continuous optimization for offline change point detection and estimation in univariate Gaussian data sequences. The main idea hereby lies in reformulating the normal mean multiple change-point model into a regularized statistical inverse problem enforcing the sparsity of the parameter vector. After introducing the problem statement, criteria and recalling previous investigation via Lasso-regularization for sparsity, the novel and enabling framework of continuous optimization for best subset selection (COMBSS) is briefly introduced and connected to the problem at hand. Both supervised and unsupervised perspectives are explored, with the latter testing different approaches for the choice of regularization penalty parameters via the discrepancy principle and a confidence bound. The main result is an adaptation and evaluation of the COMBSS approach for offline normal mean multiple change-point detection via experimental results on simulated data for different choices of regularisation parameters. Results, as well as further directions for investigations, are then critically discussed.
