Fig. 9: Comparison of feature maps between primates and alternative models.

a Cosine angles between average features of three monkeys and the animacy feature, plotted on the left temporal lobe of the monkey template. b, c IC1’s feature map (b) and autocorrelation (c). d, e Each IC’s autocorrelation map was fitted with sinusoids of different frequencies and orientations. Explained variances, shown as a polar plot (d), were averaged across ICs (e). f Network architectures (see text). g Seven functions were used to constrain how response correlation between units vary with cortical distance. h Left, Relationship between response correlation and cortical distance for units in two trained networks (bin = 1.5 mm; n = 222828, 607829, 936397, 1220800, 1508523, 1632486, 1835337, 1938684, 2099738, 2109536, 2080679, 2052299, 2002141, 1932663, 1830959, 1676084, 1564828, 1389092, 1201783, 1040034, 841442, and 659991 unit pairs in each bin for network 6; n = 49004, 156206, 231122, 309040, 374379, 427758, 449747, 503386, 506777, 527946, 509935, 520644, 487729, 494298, 444406, 437258, 377210, 359787, 303565, 253240, 212334, and 166229 unit pairs for network 36). Dots: mean; error bars: SD. Curved lines represent corresponding constraining functions. Right, Distribution of the correlation between response correlations and values of the constraining function of cortical distances. i Face- and object-selective units in two networks. j Feature maps of two networks for IC1 (see Methods). k Same as (e), but for two networks. l Following (e) and (k), maximum explained variance across orientations is plotted against frequency for the monkey map and two networks. m Explained variance is plotted against orientation for a specific frequency. n Orientation selectivity (see Methods) is plotted against frequency. o Definition of orthogonality index (see Methods). p Frequency tuning width, orientation selectivity, and orthogonality index are plotted against preferred frequency for networks and monkeys. Red and gray dots represent monkeys and networks. q Same as (p), but for the average map of the same network trained three times or three monkeys. r Same as (p), but for human subjects and corresponding networks. Open and solid red dots indicate males and females, respectively. s Comparing monkeys and human subjects using four indices in (l–o). Each dot represents one hemisphere. Open and solid red dots indicate males and females. Student’s t-test (two-tailed) was performed for each index: p = 0.012, 0.001, 0.001, and 10⁻⁴; t(12) = 2.96, 4.18, 4.37, and 5.82; n = 6 and 8 hemipheres for monkey and human subjects, respectively. Source data are provided as a Source Data file.