Fig. 6: Orientation and its variance can be decoded from resilient neurons.
From: Cortical recurrence supports resilience to sensory variance in the primary visual cortex
a Time course of the accuracy for decoding θ × Bθ of Motion Clouds. Lines are the mean accuracy and contour the SD. Significantly better decoding from resilient neurons is shown as a gray overlay (Wilcoxon signed-rank test, threshold p < 0.01). Decoding at chance level (1/96) is represented by the gray dashed line. b Population tuning curves for the likelihood of decoding each θ × Bθ in four time windows, centered around the correct θ × Bθ. c Correlation between classification accuracy and population circular variance for correct Bθ population tuning curves (left) and averaged across other Bθ tuning curves (right). Linear regression is shown as solid lines with slope m indicated (all significant p < 0.001, Wald Test with t-distribution). d Time course of the θ × Bθ decoder, marginalized over Bθ to produce θ-only outputs. e Mean decoding coefficients of the two groups yielded from the whole population θ × Bθ decoder. f Score-based decoding for θ (first and second columns), Bθ (third) and θ × Bθ (fourth). Raw scores (points) are fitted with a linear regression (dashed curve), with Spearman R shown in the case of a significant correlation (p < 0.05).
