Researchers at the University of Pennsylvania have proposed a new, unified account of how the brain forms perceptions by integrating two influential ideas in sensory neuroscience.
For decades two complementary theories have tried to explain why our subjective view of the world often diverges from physical reality. One of these, the Bayesian decoding framework, emphasizes how prior experience shapes perception: the brain combines incoming sensory signals with expectations, or priors, derived from past encounters. For example, because school buses are usually yellow, the brain tends to interpret ambiguous visual input of a bus as more likely yellow than blue.
The other influential idea, efficient coding, focuses on how the sensory system allocates limited neural resources. According to this view, neurons and sensory tissue represent the statistics of the environment more accurately where it matters most. Frequently encountered stimuli—like a yellow school bus—are encoded with higher precision, while rare or unexpected patterns receive less representational bandwidth.
Alan Stocker, an assistant professor jointly affiliated with the psychology and electrical and systems engineering departments, and Xue‑Xin Wei, a graduate student in psychology, combined these approaches to develop a single model that links stimulus frequency to both expectation and encoding fidelity. Their proposal is that how often we experience a given stimulus shapes two things at once: the prior expectation we bring to future encounters, and the fidelity with which sensory systems encode similar stimuli.
When these two forces—prior expectations and efficient allocation of encoding resources—interact, they do not always produce intuitive results. Stocker and Wei found that under certain conditions the combined model predicts perceptual biases that run opposite to the prior: stimuli can be perceived as pushed away from expected values, a phenomenon sometimes called repulsive or “anti‑Bayesian” bias. Returning to the bus example, if the actual vehicle were blue, the interaction of efficient coding and prior expectations could make it appear even bluer rather than merely less yellow.
The researchers also emphasize that the source and nature of uncertainty matter. Uncertainty that arises from the external stimulus (blur, brief presentation, visual clutter) has different effects on perceptual bias than uncertainty that originates from internal neural noise. In practical terms, a briefly shown object produces a weaker, lower‑spike neural response, which lowers representational accuracy and interacts differently with prior expectations than noise introduced at later stages of processing.
Because the Bayesian framework has been so dominant, repulsive biases have sometimes been labeled “anti‑Bayesian” and met with skepticism. Stocker notes that embedding efficient coding constraints into the Bayesian observer framework reframes those data: many perceptual patterns previously viewed as exceptions now follow naturally from a model that respects both prior knowledge and resource allocation in sensory encoding.
This synthesis opens new avenues for interpreting psychophysical data. It predicts not only when perception will be attracted toward priors, as standard Bayesian models do, but also when perception will be repelled from priors, depending on stimulus statistics and noise sources. The combined account explains reported biases in visual orientation and spatial frequency tasks and resolves discrepancies that neither theory alone could fully address.
Source: Michele Berger – University of Pennsylvania
Image Source: The image is credited to the researchers and is adapted from the Research Square video
Video Source: Research Square Vimeo page (video available on the Research Square channel)
Original Research: Abstract for “A Bayesian observer model constrained by efficient coding can explain ‘anti‑Bayesian’ percepts” by Xue‑Xin Wei and Alan A. Stocker, published in Nature Neuroscience (online September 7, 2015). doi: 10.1038/nn.4105
Abstract
A Bayesian observer model constrained by efficient coding can explain ‘anti‑Bayesian’ percepts
Bayesian observer models provide a principled account of the fact that our perception of the world rarely matches physical reality. The standard explanation is that our percepts are biased toward our prior beliefs. However, reported psychophysical data suggest that this view may be simplistic. We propose a new model formulation based on efficient coding that is fully specified for any given natural stimulus distribution. The model makes two new and seemingly anti‑Bayesian predictions. First, it predicts that perception is often biased away from an observer’s prior beliefs. Second, it predicts that stimulus uncertainty differentially affects perceptual bias depending on whether the uncertainty is induced by internal or external noise. We found that both model predictions match reported perceptual biases in perceived visual orientation and spatial frequency, and were able to explain data that have not been explained before. The model is general and should prove applicable to other perceptual variables and tasks.
“A Bayesian observer model constrained by efficient coding can explain ‘anti‑Bayesian’ percepts” by Xue‑Xin Wei and Alan A. Stocker, Nature Neuroscience. Published online September 7, 2015. doi:10.1038/nn.4105