Expectation Effects Based on Newly Learnt Object-Scene Associations Are Modulated by Spatial Frequency.

Objects typically appear within rich visual scenes. By capitalizing on learnt statistical pairings between objects and scenes, the visual system can use scene information to generate predictions about objects likely to occur within a given environment. Some models of visual system function propose that scene information is extracted from low-spatial frequency components and rapidly propagates through the visual processing hierarchy. This contextual information may help bias perceptual inferences toward objects that are likely to appear within a scene, enacted via top-down feedback carrying predictions. We tested this hypothesized influence of low spatial frequency information through newly learnt predictive object-scene associations. We recorded electroencephalographic (EEG) data from 40 participants who viewed high-spatial frequency objects either in isolation or embedded within low- or high-spatial frequency scenes. Object-scene pairings were probabilistically manipulated such that certain objects more frequently appeared in certain scenes. We trained classifiers on EEG data from object-only trials and tested them on object plus scene trials. We did not observe differences in classification accuracy across expected and unexpected objects for both low- and high-spatial frequency scenes, and any interaction between spatial frequency and expectation. However, we observed expectation effects on event-related potentials for both spatial frequency conditions. These effects arose at similar latencies for both low- and high-spatial frequency scenes but interacted with the expectation effects. Together, we report evidence that expectations induced by object-scene pairings influence visually evoked responses but do not modulate object representations.