Skip to main content
. Author manuscript; available in PMC: 2016 Nov 18.
Published in final edited form as: Neuron. 2015 Nov 5;88(4):819–831. doi: 10.1016/j.neuron.2015.10.009

Figure 4. Comparison of Restricted Models.

Figure 4

(A) Comparison of fit quality of the full model against various restricted models, over populations of V1 (dark gray) and V2 (light gray) cells. Fit quality is captured with the correlation of model-predicted and measured responses for a held-out set of data and is plotted on Fisher transformed axes. For restricted models, only one model component is free to vary across neurons to account for variable orientation selectivity.

(B) Fraction of tuning diversity in V1 (dark gray) and V2 (light gray) populations that is explained by full and restricted models. Explained diversity expresses the fraction of across-neuron variance in OSI that is accounted for by the model. It is bounded below 100% because neural responses are variable (see Experimental Procedures). Confidence intervals, where visible, illustrate the interquartile-range of the fraction of explained diversity. n.s., not significant. *p < 0.05; **p < 0.01, ***p < 0.001 (Wilcoxon signed-rank test).