How Do Expectations Shape Perception?

Abstract

Perception and perceptual decision-making are strongly facilitated by prior knowledge about the probabilistic structure of the world. While the computational benefits of using prior expectation in perception are clear, there are myriad ways in which this computation can be realized. We review here recent advances in our understanding of the neural sources and targets of expectations in perception. Furthermore, we discuss Bayesian theories of perception that prescribe how an agent should integrate prior knowledge and sensory information, and investigate how current and future empirical data can inform and constrain computational frameworks that implement such probabilistic integration in perception.

Evaluating the neurophysiological evidence for predictive processing as a model of perception

“…According to PP, the range of interpretations of sensory stimulation entertained by the perceptual system is constrained by prior experience. Some priors can be bestowed through phylogenetic development and hardwired into neural structure, 138 while others will emerge from persistent regularities in natural sensory experience and will likely be instantiated in the tuning properties of sensory cortex 4,19,139 . For example, the prevalence of cardinal orientations in the sensory environment is reflected in the more narrow tuning and overrepresentation of neurons selective for cardinal orientations in early visual areas 140 .…”

“…Our immersion in a seamless and coherent perceptual experience veils the reality that it must be assembled from a sea of noisy and ambiguous sensory information 1–3 . It was Helmholtz who first proposed that perception does not obediently reflect the sensory inputs that undergird it, as exposed in myriad perceptual illusions, 4–6 but arises from an inferential process in which stimuli are interpreted in light of our past experiences. Quite how this is achieved in the brain is still hotly debated 7–11 .…”

“…The success of feedforward models in accommodating neurophysiological and anatomical observations has established them as the orthodox perspective on sensory processing. In recent years, however, a growing contingent of scholars has entertained a radical new framework that casts perception as an entirely inferential process in which predictions about the outside world shape information processing at all levels of the cortical hierarchy 1,4,9,17–19 …”

“…According to PP, expectation units are expected to primarily occupy deep cortical layers, while error units should be found in superficial layers (adapted from Ref. 214). (B) Kok et al 169 .…”

“…In recent years, however, a growing contingent of scholars has entertained a radical new framework that casts perception as an entirely inferential process in which predictions about the outside world shape information processing at all levels of the cortical hierarchy. 1,4,9,[17][18][19] Predictive processing (PP) claims that the brain confronts the inherent ambiguity in sensory input by assembling "generative models" of the causes underlying sensory events. Generative models yield 18 Here, each level is composed of a population of reciprocally connected expectation units (Ex) and error units (Er).…”