Imaging prior information in the visual system

Abstract

Prior information and experience with visual stimuli enhance our ability to recognize images, but where and how does this facilitation occur in the brain? Using stimuli which temporally evolve out of noise and then degrade again, we show that recognition of degraded stimuli persists past the level of noise required for initial recognition, a phenomenon known as hysteresis. Furthermore, using machine learning algorithms, we can quantify the amount of information a given brain region or neuron contains about the stimulus as the subject learns the coherent image. Here we show how distinct brain regions from prefrontal cortex to V1 contain more information about degraded stimuli with prior knowledge, and that local information in the brain persists in line with behavioral hysteresis. Interestingly, behavioral and neural hysteresis depends critically on the complexity of the stimuli, so that prior information seems to be encoded over complex, real-world features, but not simple stimuli such as oriented gratings.