Multiscale breakpoint detection in piecewise stationary AR models

In this paper, we are interested in the problem of detecting breakpoints in piecewise stationary autoregressive (AR) processes. For short time series, stationarity assumption is common. However, for longer time series, this assumption is often unrealistic. Besides, many naturally occurring phenomena cannot be modelled as stationary processes. Therefore modelling stochastic time series under nonstationary assumption is often appealing, and finnds its application in many areas, such as speech processing, biomedical signal processing, and seismology to name a few.