Figure 3.

Comparison of the estimation results from the modified Kalman filter with the unscented Kalman filter from the Monte-Carlo simulation (50 realizations). (A) The bias for parameter estimation as a percentage of the true value for the connectivity gain using the UKF. (B) The bias for parameter estimation as a percentage of the true value for the connectivity gain using the analytic mean. (C) RMS error for state tracking of the post synaptic potentials using the UKF. (D) RMS error for state tracking of the post synaptic potentials using the analytic mean. The center line of the box plots shows the median error and the box covers are the 25th to 75th percentiles. The whiskers cover the entire range of errors that are not considered outliers, which are shown by the dots. The outliers are determined to be outside q₁ − 1.5(q₃ − q₁) to q₃ + 1.5(q₃ − q₁) where q₁ and q₃ denote the 25th and 75th percentiles.