Section: New Results

Computational and statistical trade-offs in change-point detection

Participants : Christophe Biernacki, Maxime Brunin, Alain Celisse.

The change-point detection problem aims to detect changes in the distribution of observations collected over the time between the instants 1,...,T in the offline context. These changes occur at some instants called change-points. Our method provides consistent estimates of the change-points obtained by the Kernel Binary Segmentation algorithm with stopping rule (KBS). Moreover, the proposed method has a lower complexity in time and in space than the Kernel Dynamic Programming (KDP). This work has been presented to a national conference with an international audience [34].