Fig. 2.

Idealization (red) of the observations in Fig. 1 by $\text{SMUCE}$ (Frick et al. 2014) displayed on three different temporal scales. In the lower panels, we also show the convolution of the idealization with the low-pass filter (blue). $\text{SMUCE}$ is a multiscale method similar to our model-free idealization approaches, but does not take into account filtering. Hence, local dependencies due to colored noise are misinterpreted as events. Moreover, abrupt conductance changes are split into multiple steps, since the convolution of the conductance with the low-pass filter is ignored. Consequently, $\text{SMUCE}$ overestimates massively the number of conductance changes