Yiwen Wang¹, Ling Lee Chong¹, Diane M. Beck¹; ¹University of Illinois at Urbana-Champaign

Real-world statistical regularities, learned through a lifetime of exposure to structured environments, allow the visual system to form predictions about incoming stimuli. Under predictive coding models, statistically regular images generate smaller prediction errors and should therefore be processed more efficiently than statistically irregular images. One important form of such regularity is scene representativeness, where highly representative “good” exemplars are detected more efficiently than “bad” exemplars (Caddigan et al., 2017). The present study tested whether the influence of representativeness arises during unconscious or conscious stages of visual processing. We combined the standard Breaking Continuous Flash Suppression (bCFS) task with its reverse variant (rev-bCFS; Ciorli, Pia, & Stein, 2025). In bCFS, the image is initially suppressed, and the time it takes to break into awareness reflects early, unconscious processing efficiency. In contrast, rev-bCFS presents the image fully visible at trial onset and then introduces suppression, measuring how long a consciously perceived image resists suppression, a marker of conscious-level properties. Our results showed that good exemplars broke suppression faster than bad exemplars in bCFS, demonstrating a representativeness advantage during initial, unconscious processing. However, in rev-bCFS, good and bad exemplars exhibited comparable suppression-resistance times, showing no effect once images were consciously available. These findings align with the predictive coding framework, indicating that good exemplars gain an early processing advantage because they produce smaller prediction errors, but this benefit does not extend into conscious perception, where prediction errors no longer drive performance.