Figure 2.

Segmentation of a time series of length T into M+ 1 contiguous, non-overlapping segments with (inclusive) upper segment boundaries q_m ∈ {q_m}. The observation model in each segment m is given by P(D_m|q_{m − 1}, q_m), where D_m is that part of the data which is in segment m. Importantly, the model assumes that the data are independent across segments given the {q_m} and M.