Figure 5. Box plots summarizing the inference errors without and with averaging for.

(a) Sontag et al.’s³³ method ; (b) -min with noise magnitude given , (c) -min with noise magnitude underestimated as 10% the actual , and (d) -min with noise magnitude overestimated as 10 times the actual . The inference error was reduced by 10⁵ times when using the -min approach (21, 22), compared to Sontag et al.’s³³ method. Averaging further reduced inference error by at least 30% in all cases (p-values of the paired t-tests consistently were below 0.0282).