Robustness analysis. Internal (LGN) and external (Retinal) noise, and parameter (LTD/LTP rate ratio) variation were introduced in the model. The internal (left column) and external (middle column) noises were modeled as additive white Gaussian processes with SNR levels between −3 and 10 dB. The ratio of the LTD (α⁻) and LTP (α⁺) rates (right column) was varied in logarithmic steps from 10⁻² to 10² (i.e., α⁻ = α⁺/100 to α⁻ = α⁺ × 100). This ratio is a crucial parameter that determines the relative contributions of synaptic potentiation and depression to the overall learning rate. A, CI (see Materials and Methods) for the three types of noise plotted against the number of iterations (training period). For a stable synaptic convergence, the CI should approach zero asymptotically. Internal and external noise are plotted in green and red colors, respectively, with the lightness of the curve coding for the SNR (the darker the curve, the higher the noise). The LTD/LTP rate ratio is plotted in purple colors for α⁻ ≤ α⁺ and orange/brown colors for α⁻ ≥ α⁺. The darker the color, the closer α⁻ is to α⁺, with black being α⁻ = α⁺. This color coding is used throughout the figure. B, Mean R_max². The left and middle plots show the mean R_max² plotted against the SNR of the internal and external noise respectively. The right plot shows the mean R_max² plotted against the α⁻/α⁺ ratio. An additional dot (outlined in pink) is added to this plot to indicate the value of α⁻/α⁺ (= 3/4) used for all the simulations presented in the article (except this figure). C, Population disparity tuning and BII. Each curve in this panel shows the probability density (PD) function estimated over the preferred disparities of a neural population (corresponding to a given condition). Each PD curve is accompanied by an estimate of the mean BII of the neural population (for details, see Materials and Methods), calculated using responses to RDS stimuli. Only neural populations that show a reasonable response to RDS stimuli are shown (otherwise, the BII and PD curves are not meaningful).