Comparing the performance of our trained LN model to the performance of an LN model that uses the STA as the filter (LN0) or the first principal component of the STC (LN1).

For both LN0 and LN1, the nonlinear function is computed as the histogram of filter outputs versus spike counts. For the comparison here, either LN0 or LN1 was chosen, depending on which model had better performance.

(TIF)