Supplementary Figure 2: Fraction of predictable variance accounted for by our model.
From: Spatiotemporal receptive fields of barrel cortex revealed by reverse correlation of synaptic input
Unbiased estimators of noise-related variance in our neural responses and the fraction of predictable variance accounted for by our model were calculated as in Sahani & Linden 2003. Black, training data. Gray, testing data. Extrapolating the performance of the model to conditions of zero noise gave an upper-bound of predictable variance of 0.75 ± 0.06 and a lower bound of 0.58 ± 0.03. b, The relationship between neural variability and firing rate. The firing rate is plotted by neuron and depth alongside a similar plot for variability by neuron and depth. c, There is no significant relationship between the firing rate and neural variability (right, p-value = 0.22). d, The firing rate (left) and signal-to-noise ratio, SNR (right), calculated as in Sahani & Linden 2003, are plotted by cell type. Neurons with high firing rates can have both low SNR (L5TT) or high SNR (L4) and neurons with low firing rates can have both low SNR (L2) as well as high SNR (L3 and L6).
