Figure 5.

High-level outline of our algorithm for decoding the profile. The algorithm we use for decoding the probabilistic profile into a set of regions consists of a series of recursive stages. At each level, we choose to partition the portion of the profile shown in black into two different region classes that differ either in $\overline{P_{\text{present}}}$ or $\overline{P_{\text{id}}}$. We then recursively split each of the two classes until we have partitioned the profile into enough different region classes.