Properties of the feedforward neural network model.

<p>(A) Histogram of the number of effective hidden units (HU) of the NRF model fits for each neuron. (B) Distribution of the number of effective excitatory and inhibitory HUs for each neuron. (C) Frequency tuning width of the NRF model versus that of the LN model. The red dot indicates the average frequency tuning width of both models. (D) Temporal tuning width of the power STRF of the NRF model versus that of the LN model. The red dot indicates the average temporal tuning width of both models. (E) A plot of the expansive/compressive (EC) score (which measures how the nonlinear activation function is used) against the excitatory/inhibitory (IE) score for all 246 effective HUs from the 76 neurons. Contour plot shows the density. (F) Distribution of the EC score for excitatory (red) and inhibitory (blue) HUs. (G) Distribution of the EC score for the output unit of the NRF model (red) and for the LN model (cyan). (H) Distribution of the EC score for excitatory (red) and inhibitory (blue) HUs for the bi-feature neurons. (I) Distribution of the EC score for excitatory (red) and inhibitory (blue) HUs for the multi-feature neurons.