Change-point detection in hierarchical circadian models

This paper addresses the problem of change-point detection in sequences high-dimensional and heterogeneous observations, which also possess a periodic temporal structure. Due to dimensionality problem, when time between change points is order dimension model parameters, drifts underlying distribution can be misidentified as changes. To overcome this limitation, we assume that observations lie lower-dimensional manifold admits latent variable representation. In particular, propose hierarchical computationally feasible, widely applicable data robust missing instances. Additionally, observations’ dependencies are captured by non-stationary covariance functions. The proposed technique particularly well suited (and motivated by) detecting changes human behavior using smartphones its application relapse psychiatric patients. Finally, validate on synthetic examples demonstrate utility behavioral real acquired smartphones.