Fig. 3.

Performance of PisCES in a scenario with outlier at time $t=6$ (main text). Simulations were generated from a dynamic DCBM with $n=500$, $K=10$, $p_{\text{in}}=(0.3,0.3)$, $p_{\text{out}}=0.1$, $r=1$ for $A_6$, and $r=0.1$ outside the change point. (A) ARI performance of PisCES applied to $A_1,\dots ,A_{11}$ (blue line), static spectral clustering (static, red dashed line), and PisCES applied separately to $A_1,\dots ,A_5$ and to $A_7,\dots ,A_{11}$ (“separated,” green lines). (B) ${\parallel {\overline{U}}_t-{\overline{U}}_t^{\text{static}}\parallel }_1$, where ${\overline{U}}_t$ is the output of PisCES (green) or “separate” (blue), and ${\overline{U}}_t^{\text{static}}$ is the output of static.