Encoding 10,000 pictures
Friday, September 18, 2015 — Poster Session V
- DA Leopold
Using chronically implanted microwires capable of tracking neurons across days, we recently showed that neurons in inferotemporal (IT) cortex maintain consistent visual properties across many months (McMahon et al. PNAS 2014). Together with the empirical finding that IT response patterns are stationary, the ability to record from neurons longitudinally makes it possible to conduct large-scale screenings of neuronal feature space using massive stimulus sets. Here we realized this possibility by screening a library of 10,000 stimuli comprised of 10 different object categories. In 30 neurons in the AF (anterior fundus) face patch within the superior temporal sulcus, visual responses were significantly correlated when the same images were presented on different days. Robust discriminative information was present in the neurons’ responses to face stimuli, and to a lesser extent in the neurons’ responses to non-preferred stimulus categories. Scale tolerance was observed in the selectivity patterns for both face and non-face stimuli. In the same neurons, the magnitude of evoked responses varied monotonically with images scale. The majority of neurons showed a dissociation between the gain factor of scale tuning and stimulus category, in that response magnitude was positively coupled with face image size but negatively coupled with all other stimulus categories. These results illustrate how by spending the extra trials accrued over days broadly across many stimuli, we can obtain a picture of neuronal selectivity analogous to surveying a landscape via satellite photography rather than at ground level.