Ensemble Fractional Sensitivity: A Quantitative Approach to Neuron Selection for Decoding Motor Tasks
Figure 1
An ensemble of M unique nonlinear models was trained using Monte Carlo simulations; each of these models related the input neuronal activity to the output behavior with high accuracy. In order to assess the contribution of each model input, , to encoding the final output, Y, a sensitivity analysis (SA) was performed. For each of the M models, the localized sensitivity coefficients, , were calculated by taking the partial derivative of the output with respect to the input activity. The sensitivity coefficients, , were then computed by taking the average of the localized sensitivity coefficients across all instances of the testing data. The sensitivity coefficients were normalized and expressed as a fraction of the cumulative sensitivity values across all inputs in order to yield the fractional sensitivity coefficients, . After this sensitivity analysis, neurons were rank-ordered by the mean fractional sensitivities from the ensemble, This rank-ordered list was passed to filter to decode the motor task.
