Extended Data Fig. 3: Quantification of familiarity-related behavior, face selectivity, and axis tuning.

a, Schematic illustration of face identification task, a sample face with different Gaussian blur level was presented for 1 s followed by a test period with two faces presented side by side. The subject had to choose the one matching the sample to get reward (see Supplementary Methods). b, Rate of correct performance on the face identification task across different difficulty levels (accomplished by varying Gaussian blur of the sample face, see Supplementary Methods); n = 30 faces. Error bar, SEM. c, Histograms of face selectivity indices computed using screening stimuli (see Supplementary Methods). d, Preferential looking test. Comparing looking time to personally familiar faces versus novel unfamiliar faces, unfamiliar faces (from 1000 face set), personally familiar faces (two distinct personally familiar faces were presented on each trial), pictorially familiar faces, and cinematically familiar faces. Error bar, SEM. e, Distribution of explained variance by the linear axis model for responses to 1000 unfamiliar faces; shaded bars indicate the subset of cells for which the explained variance was significantly higher than for stimulus-shuffled data (1000 repeats). f, Distributions of mean cosine similarity of preferred axes across repeated split halves (100 repeats) of responses to 1000 unfamiliar faces for AM and PR. Same conventions as in e. g, h, Same as e and f but for 36 familiar and unfamiliar faces.