Abstract
Prior knowledge powerfully facilitates object recognition. In a dramatic example of one-shot perceptual learning, a previously unrecognizable, degraded image of a real-world object becomes instantly recognizable after exposure to the corresponding original image. We previously showed that neural activity changes driven by one-shot perceptual learning are widespread across the ventral visual stream, extending into frontoparietal (FPN) and default-mode (DMN) networks. However, it remained unclear where the image-specific prior knowledge is encoded in the brain and what type of information is stored. To address these questions, we modified the original images based on receptive field sizes and orientation tuning properties of object representation across the ventral visual hierarchy. We then tested for potential generalization of one-shot perceptual learning to systematically map out the invariance properties of prior knowledge encoded in the brain. First, changing the image size from the original 12 DVA to 6 or 24 DVA, thereby altering perceptual information available to the small receptive fields of early visual cortex, did not change the perceptual learning effect. However, shifting the image position by 6 DVA to the left or right, and therefore strongly influencing but not eliminating inferotemporal (IT) neural coding, significantly diminished the effect without abolishing it. Similarly, rotations and inversions (targeting orientation invariance properties emerging within IT cortex) significantly diminished without abolishing the learning effect. These results suggest that priors are likely encoded in the inferotemporal cortex, wherein rotation invariance emerges, although they do not rule out the possible involvement of higher order regions. Interestingly, we found no change in the learning effect when we biased image properties to selectively activate parvo- or magnocellular pathways, suggesting that either pathway alone can encode the perceptual prior. Together, we show that encoding of priors in one-shot perceptual learning depends on regions involved in whole-object representations.