Fig. 5: Top-down influences shaping the representation in Z1.
From: Top-down perceptual inference shaping the activity of early visual cortex
a Mean performance of texture family decoder from mean responses of Z1 with intact inference (0.1943) is higher than without stimulus-specific information at Z2 (0.0986) (n = 5 fits from different random seeds; one-sided independent two-sample t-test: t(df = 8) = 70, p = 9.6 × 10−13, \({d}_{e}^{{\prime} }=52.4\), 95% confidence interval = [0.093, ∞]). Bar heights: means; s.d. < 0.01 everywhere. b Qualitative time course of illusory contour responses in the same restricted Z1 population in TDVAE as in Fig. 4g (n = 9 Z1 units everywhere; two-sided one-sample t-test against 0 everywhere except late Illusory; pre Illusory: mean: 0.0011, t(df = 8) = 0.72, p = 0.49, 95% confidence interval = [−0.0025, 0.0047]; pre Rotated: mean: 0.00077, t(df = 8) = 0.47, p = 0.65, 95% confidence interval = [−0.0030, 0.0045]; early Illusory: mean: −0.068, t(df = 8) = −0.71, p = 0.50, 95% confidence interval = [−0.29, 0.15]; early Rotated: mean: −0.091, t(df = 8) = −0.85, p = 0.42, 95% confidence interval = [−0.34, 0.16]; late Illusory: mean: 0.076, one-sided one-sample t-test against 0: t(df = 8) = 4.7, p = 0.00076, 95% confidence interval = [0.046, ∞]; late Rotated: mean: 0.0065, t(df = 8) = 0.92, p = 0.39, 95% confidence interval = [−0.0098, 0.023]). Center line, median; box limits, upper and lower quartiles; whiskers, 1.5 × interquartile range. Inset: example stimuli. c Qualitative time course of texture family decoding performance from intact and phase-scrambled versions of textures (n = 5 fits from different random seeds; mean accuracies ± stds: pre: 0.0680 ± 0.0007, early: 0.0986 ± 0.0025, early scrambled: 0.1007 ± 0.0012, late: 0.1943 ± 0.0012, late scrambled: 0.1709 ± 0.0020; two-sided independent two-sample t-test between early and early scrambled: t(df = 8) = −1.6, p = 0.15, \({d}_{e}^{{\prime} }=1.19\), 95% confidence interval = [−0.0053, 0.00098]; one-sided independent two-sample t-test between late and late scrambled: t(df = 8) = 20.3, p = 1.8 × 10−8, \({d}_{e}^{{\prime} }=14.8\), 95% confidence interval = [0.021, ∞]). Bar heights: means; s.d. < 0.01 everywhere. Inset: example stimuli. d Qualitative time course of contour completion experiment. Top: Illustration of stimuli. Optimally oriented, identical line segment is shown in the receptive field of a Z1 neuron (left) under the condition that the neighboring visual field is filled with randomly oriented segments (center) or the neighboring segments continue the segment that covers the receptive field (right). Bottom: Qualitative time course of the response intensity difference between the contour completion and random conditions in the Z1 population (n = 57 central, localized, medium wavelength Z1 filters in TDVAE in both cases; early: mean effect size: −0.080, two-sided one-sample t-test against 0: t(df = 56) = −0.65, p = 0.52, 95% confidence interval = [−0.33, 0.17]; late: mean effect size: 0.36, one-sided one-sample t-test against 0: t(df = 56) = 3.9, p = 0.00014, 95% confidence interval = [0.21, ∞]). Center line, median; box limits, upper and lower quartiles; whiskers, 1.5 × interquartile range. Source data are provided as a Source Data file.
