Fig. 3: Preferred axes of units.

a–d Data for example unit 258 in the MSB region of monkey T. b A scatter plot depicting the relationship between the unit’s distance on its preferred axis (model response) and its actual response. c Stimuli that elicited a high response, as shown in (b), are ordered in descending response strength (coded in shades of green; darker: stronger response). a Stimuli that elicited a low response, as shown in (b), are ordered in ascending response strength (blue). d Poses, estimated using the 2D keypoint model, corresponding to a range of standard deviations [σ = −3, σ = 3] along the preferred axis. The color of a keypoint indicates its max-normalized weight, representing its contribution to the unit’s selectivity. We randomly permuted the stimulus labels (N = 1000) to compute a null distribution of weights of each keypoint and computed a p-value for the observed weight. It was assigned a value of 0 when the observed weight of the keypoint was not significant (False Discovery Rate corrected). e–h Data for example unit 625 in the MSB region of monkey G, using the same conventions as (a–d). i–l Data for unit 669 in the ASB region of monkey T. The same conventions as in (a–h), except that the preferred axis was estimated using the 3D_VD keypoint model and is plotted in 3D.