Figure 9. Gaussian mixture model Expectation Maximization Algorithm (GEMA) was run on the third simulation, which contained clusters with an average σ of ~3 nm. GEMA probabilistically assigned atoms to clusters (blue) and the matrix (black). The learned cluster centers are shown in red. On the left, the receiver operating characteristic plot for this simulation is shown. Again, GEMA is shown in black and the maximum separation method is shown in red.
Figure 10. Gaussian mixture model Expectation Maximization Algorithm (GEMA) results for the fourth simulation in which cluster size was varied. GEMA probabilistically assigns atoms to clusters (blue) and the matrix (black). The learned cluster centers are shown in red.
Figure 11. Gaussian mixture model Expectation Maximization Algorithm (GEMA) is run on raw atom probe data (left), resulting in the clustered output shown (right). GEMA probabilistically assigns atoms to clusters (blue) and the matrix (black). The learned cluster centers are shown in red. Although several of the clusters appear to be overlapping, they are, in fact, separate and distinct. This is simply an artifact of the two-dimensional (2D) projection of the 3D reconstruction.
