Table 1.

Quality of receptive-field fits for four cells illustrated in Figures 3, 4, 5, 6, 7

Correlation coefficient Kernel parameters Sum-of-squares deviation
1 2 3 4 5 6 7 8 9 10 11 12 13
Cell ID Rsps V/M Actual Bound Gabor Or.R. Or.G. SF.R. SF.G Reg. Spat. Spec.
f31205 4276 0.8 0.89 0.96 0.82 160 161 0.1 0.1 76% 63% 69%
f32610 1531 3.58 0.81 0.89 0.7 176 171 0.15 0.13 65% 51% 45%
f31008 359 0.81 0.96 0.7 165 163 0.13 0.09 66% 44% 55%
f32203 231 1.56 0.62 0.75 0.45 172 172 0.07 0.06 39% 16% 20%
  • The total number of action potentials generated in each experiment is shown (column 2), along with the ratio of the response variance to mean response for the 50 pictures giving the greatest mean response to the 10-12 repetitions (column 3). For cell f31008, each picture was presented only once, so no variance measure is available. The actual correlation coefficient between the measured and predicted responses to pictures (column 4) is compared with the upper bound (column 5) if the field had been a perfect reconstruction corrupted only by Poison response variance, in which response variance equals the response mean (but see column 3). Column 6 shows the correlation coefficient between the actual responses and those predicted from the best Gabor-fitted receptive-field map. Columns 7-8 show the peak orientation derived from the Reginv and Gabor kernel estimates, whereas columns 9-10 show the peak spatial frequency from the Reginv and Gabor kernels. Columns 11-13 show the sum-of-squares deviation between the measured responses and those predicted by three different models (Reginv kernel, spatial window, spectral window) as percentages of the total sum-of-squares of the measured response values. The models are described in Results. Rsps, Responses; V/M, variance to mean ratio; Or.R., Reginv peak orientation; Or.G., Gabor peak orientation; SF.R., Reginv peak spatial frequency; SF.G., Gabor peak spatial frequency; Reg., Reginv; Spat., spatial; Spec., spectral.