Super-recognisers sample visual information of superior computational value for facial recognition

Expertise involves superior perceptual-cognitive skills, exemplified by super-recognisers with their exceptional face recognition abilities. The role of information sampling in their face recognition ability remains unclear. This study investigates this by assessing the computational value of face information that is actively sampled by typical human observers and super-recognisers. Using eye-tracking data, we reconstructed the face information sampled by super-recognisers and typical viewers as they learned new faces. We then evaluated the computational value of this information using pre-trained Deep Neural Networks (DNNs) optimised for face identity matching. Our results showed that identity matching accuracy improved in 9 DNNs when using human-guided visual sampling compared to random fixation patterns on the face stimuli. Notably, the DNNs demonstrated higher accuracy when using visual information sampled by super-recognisers compared to typical viewers. Thus, super-recognisers do not merely sample more face information, but also sample information that is more diagnostic of identity. This shows that super-recognisers' expertise is founded on superior information encoding, and implies that DNNs could benefit from incorporating human attention patterns.