Abstract: We consider the problem of constructing confidence intervals for the locations of change points in a high-dimensional mean shift model. We develop a locally refitted least squares estimator and obtain component-wise and simultaneous rates of estimation of change points. The simultaneous rate is the sharpest available by at least a factor of log p, while the component-wise one is optimal. These results enable existence of limiting distributions for the locations of the change points. Subsequently, component-wise distributions are characterized under both vanishing and non-vanishing jump size regimes, while joint distributions of change point estimates are characterized under the latter regime, which also yields asymptotic independence of these estimates. We provide the relationship between these distributions, which allows construction of regime adaptive confidence intervals. All results are established under a high dimensional scaling, in the presence of diverging number of change points. They are illustrated on synthetic data and on sensor measurements from smartphones for activity recognition.

Key words and phrases: High dimensional, inference, limiting distributions, multiple change points, optimal estimation, regime adaptation.