Yi Yu (University of Warwick)
Fri 18 Oct 2019, 15:00 - 16:00
JCMB 5323

If you have a question about this talk, please contact: Tim Cannings (tcannin2)

In this paper we study the change point detection problems in high-dimensional vector autoregressive models.  We assume that the models possess piecewise-constant coefficient matrices and are piecewise stable.  We demonstrate the change point localisation consistency of a dynamic programming approach.  The model parameters, including the dimensionality, the entry-wise sparsity in the coefficient matrices, the minimal spacing between two consecutive change points and the minimal jump size in terms of the Frobenius norm of the difference of two consecutive coefficient matrices, are allowed to vary with the sample size.  We also provide an optional second step in the algorithm to further refining the localisation error rate.  

Beyond the high-dimensional vector autoregressive models, we also provide a general framework, with the high-dimensional regression problem as a vehicle, unveiling the key ingredients in the consistency of dynamic programming approaches in change point detection in high-dimensional regression and autoregression problems.