Fig. 5: Dissecting feedback influences on V1 neural population code.
From: A central and unified role of corticocortical feedback in parsing visual scenes
a Decoding performance (population d′) based on FLD analysis of the original dataset (1st column); a surrogate dataset constructed using original key metrics (2nd column); and three simulated datasets assuming that V4 inactivation altered only the rsc structure, response variances, or mean firing rates (columns 3-5). Analyses were performed for cooling experiments (top row) and sham-cooling controls (middle row) to assess the true effects of V4 inactivation (bottom row). Results for each contour length and each contour position were analyzed separately. By averaging across different contour positions (n = 8), the mean (solid curve) and SEM (light shading) were calculated as a function of contour length (top and middle rows). By averaging across the 4 contour lengths and 8 contour positions (bottom row), the mean changes (Δd’) due to V4 cooling (ΔCool) and sham cooling (ΔSham) were computed (blue vs. green bars, n = 32 conditions, 8 contour positions × 4 contour lengths; from left to right, p = 0.023, p = 0.020, p = 6.3 × 10−3, p = 0.054, p = 2.3 × 10−3, two-tailed Wilcoxon signed-rank test). b Similar analysis to a, for monkey MB (upper and middle rows: n = 8; bottom row: n = 32, from left to right, p = 9.3 × 10−4, p = 4.7 × 10−4, p = 0.092, p = 0.024, p = 1.1 × 10−4). Error bars represent SEM; n.s., p ≥ 0.05; *, p < 0.05; **, p < 0.01; ***, p < 0.001. Source data are provided as a Source Data file.
