Abstract
The typical human visual system is able to decipher information about the visual world with impressive efficiency and speed. But not all individuals are equally competent at recognising what is presented to their eyes. Critically, very little is known about the brain mechanisms behind variations in recognition abilities. Here, we ask if interindividual variation in face cognition can be accurately “read” from brain activity, and use computational models to characterise the underlying brain mechanisms. We recorded high-density electroencephalography (EEG) in typical (n=17) and “super-recogniser” participants (n=16; individuals in the top 2% of face-recognition ability spectrum) while they were presented with images of faces, objects, animals, and scenes. Relying on more than 100,000 trials, we trained linear classifiers to predict whether trial-by-trial brain activity belonged to an individual from the “super” or “typical” recogniser group. Significant decoding of group-membership was observed from ~85ms, peaking within the N170 window, and spreading well after stimulus offset (>500ms). Using fractional ridge regression, we extended these findings by predicting individual ability scores from EEG in similar time-windows. Both results held true when decoding from face or non-face stimuli. To better understand the brain mechanisms behind these variations, we used representational similarity analysis and computational models that characterise visual (convolutional neural networks trained on object recognition; CNNs) and semantic processing (deep averaging network trained on sentence embeddings). This computational approach uncovered two representational signatures of higher face-recognition ability: mid-level visual computations (representations within the N170 window and mid-layers of CNNs) and high-level semantic computations (representations within the P600 window and the semantic model). Altogether, our results indicate that an individual’s ability to identify faces is supported by domain-general brain mechanisms distributed across several information processing steps, from low-level feature integration to high-level semantic processing, in the brain of the beholder.