Fig. 3: An early shift in response subspace allows decoding of familiarity.
From: Temporal multiplexing of perception and memory codes in IT cortex
a, Responses of cells to stimuli from 36 familiar and 1,000 unfamiliar monkey faces, averaged over 50–300 ms following stimulus onset. b, Response time course across AM, PR and TP populations, averaged across cells and all familiar or unfamiliar faces from the 1,036 monkey face stimulus set. Shaded areas, s.e.m. Red arrowheads indicate the time when responses to faces became significantly higher than baseline (AM, 95 ms; PR, 105 ms; TP. 90 ms; Supplementary Methods). Green arrowheads indicate the time when responses to familiar versus unfamiliar faces became significantly different (AM, 125 ms; PR, 135 ms; TP, 105 ms). c, Black line, time course of accuracy for decoding familiarity; shaded area, s.e.m.; grey line, chance level. Arrowheads indicate the time at which decoding accuracy rose above chance (AM, 95 ms; PR, 105 ms; TP, 135 ms). d, Time course of neural distance between centroids of 36 familiar and 1,000 − 36 unfamiliar face responses (blue) and between centroids of responses to a subset of 36 unfamiliar faces and responses to the remaining 1,000 − 36 unfamiliar faces (orange). Arrowheads indicate the time when d′ along the two centroids became significant (AM, 95 ms; PR, 105 ms; TP, 135 ms). e, Distribution of differences between mean firing rates to familiar and unfamiliar faces at three different time intervals. Grey bars indicate cells showing a significant difference (Supplementary Methods). f, Distribution of cosine similarities between familiarity decoding and face feature decoding axes at short (50–150 ms) and long (150–300 ms) latency for the first 20 features (ten shape, ten appearance). g, Schematic illustration of neural representation of familiar (blue) and unfamiliar (orange) faces at short and long latency for AM and PR.
