Abstract
A minimum-error method was developed in our previous study to estimate model parameters from spike train data experimentally recorded from inferior olive neurons. Our method characterized the neuronal firing dynamics using 67 spatiotemporal features. The closest match between experimental and simulated data in the feature space determined two parameters that control the neural network dynamics of the simulation model. This approach, however, has not been fully verified, because the true parameter values were unknown for the experimental data. In the present study, we attempt to validate the minimum-error method using simulated spike data, for which the true parameter values are known, as the test data. Our performance evaluation on the test data confirmed that the minimum-error method is effective for resolving the inverse problem even when the simulation model is an imperfect representation of the experimental data.
