Article Figures & Data
Figures
- Figure 1.
Binaural correlation and its effects on spatial tuning. The response of space-specific neurons changes with BC. Overlaid ITD curves for PSPs (a). Space-specific neurons were tuned to ILD despite changes in BC (b). These data were obtained while the other parameter (ITD or ILD) was kept at the most favorable value for the neuron. Dashed lines represent mean resting potential. For BC = 0, the membrane potential showed a smaller EPSP than for BC = 1 and favorable ITD (c). The amplitude of the depolarizing PSP decreased as BC became small (d). Error bars are omitted for clarity in a and b and represent the SD in d. Traces were median filtered to eliminate spikes and averaged over five repetitions.
- Figure 2.
Multiplication of ITD and ILD as a function of binaural correlation. The singular value decomposition of subthreshold receptive fields obtained for five different BCs in the same neuron (0, 0.25, 0.5, 0.75, and 1) yielded different singular vectors for ITD (a) but nearly constant singular vectors for ILD (b). Despite BC and in all the neurons, most of the weight was carried by the first singular value (c), indicating that multiplication of two independent vectors applies in each case.
- Figure 3.
Performance of multiplicative and additive models. ILD singular vectors for BC = 1 are plotted against ILD vectors for BC = 0 to show their similarity in shape (a). The solid line represents the linear regression and R2 is the coefficient of determination. There is a high correlation between the ILD singular vector for BC = 1 and all lower BCs (b). The correlation between the ITD vector for BC = 1 and BC = 0, where the ITD input is reduced to noise, is low. However, ITD vectors for nonzero BCs were highly correlated with that for BC = 1 (c). The multiplicative model better fits the data than the additive model for those BCs (d). The means of squared differences obtained by subtracting the additive (empty circles) and multiplicative (solid circles) models from the data matrices are plotted against BC. The number of neurons used in each case (N) is indicated above the plot. Error bars represent SD.
- Figure 4.
Convergence of multiplicative and additive models. We used model matrices to study the effects of reduced input on the differences between multiplication and addition (a). We systematically scaled the amplitude of the ITD vector by numbers ranging from 0 to 1 (as indicated next to each plot). We performed SVD on a set of matrices created by addition of the ILD input to each of the scaled ITD inputs. The difference between these matrices and those derived by multiplication of the first singular vectors computed by SVD decreased as the amplitude of the ITD input declined (b). The normalized differences between the data and the model matrices are the sum of the squared differences obtained by the element-by-element subtraction of the two matrices and divided by the maximum value.
