Efficient Coding and the Bayesian Origin of Perceptual Repulsion

You are reading version 1 of this direction, frozen 2026-10-04. The latest version is version 2.

Version 1 of 210 papers18 conceptsfrozen 2026-10-04

The "Bayesian Brain" hypothesis traditionally suggests that our perception is a compromise between noisy sensory evidence (the likelihood) and our past experiences (the prior). In this standard view, the prior acts like a magnet: if you are unsure about a stimulus, your estimate should "regress to the mean," shifting toward the most likely values. This is known as perceptual attraction.

However, decades of psychophysical experiments show that the human brain frequently does the exact opposite. After staring at a tilted line, a vertical line appears tilted away from the previous one (adaptation). After deciding a stimulus belongs to a certain category, our subsequent estimates are pushed away from the category boundary. This perceptual repulsion seems "anti-Bayesian" at first glance, but recent research reveals it is actually the signature of an efficient system optimizing limited resources.

Adaptation as Resource Reallocation

For a long time, researchers assumed that sensory adaptation—the change in perception following prolonged exposure—was the result of the brain updating its "prior" to reflect a world where the adaptor is more common. But Alan A. Stocker and Eero P. Simoncelli challenged this in Sensory Adaptation within a Bayesian Framework for Perception.

They argued that adaptation isn't about changing what we expect to see, but about changing how we allocate our sensors. When we adapt to a specific stimulus, the brain reallocates its "measurement resources" (neurons) to be more precise near that stimulus. This increases the discriminability of similar items but, counterintuitively, creates a repulsive bias. Because the likelihood function becomes sharper and more asymmetric, the decoded estimate is pushed away from the region of highest precision.

This relationship between how well we can tell things apart (discriminability) and how much our perception is skewed (bias) is not accidental. Jonathan W. Pillow and Michael J. Morais demonstrated in Power-law efficient neural codes provide general link between perceptual bias and discriminability that these biases emerge from a broad family of "efficient codes."

If the brain wants to maximize information transfer given a limited number of neurons, it must distribute those neurons according to the statistics of the environment. Xin Tong, Xue-Xin Wei, and colleagues applied this logic to probability judgment in The Bayesian Origin of the Probability Weighting Function in Human Representation of Probabilities. They found that the classic "S-shaped" curve—where we overweight rare events—emerges naturally when an optimal observer decodes noisy signals from a system that has prioritized its resources for extreme values (0 and 1).

Self-Consistency and Categorical Commitment

Repulsion doesn't just happen because of how we encode signals; it also happens because of how we decide on them. In A Bayesian Model of Conditioned Perception, Stocker and Simoncelli introduced the "conditional Bayesian observer."

Imagine you are asked two questions: "Is this dot moving left or right?" and then "What is its exact angle?" If you decide the dot is moving left, your brain effectively "locks in" that choice to remain self-consistent. It discards all sensory evidence that would have suggested the dot was moving right. This truncation of the posterior distribution—chopping off the "right-side" possibilities—mechanically shifts the remaining mean of the distribution further to the left, away from the decision boundary. This explains why categorical judgments "squeeze" our perceptions, a phenomenon that Arthur Prat-Carrabin and Michael Woodford show is a fundamental trade-off between prior-driven attraction and encoding-driven repulsion in Bias and variance of the Bayesian-mean decoder.

Causal Inference and Selection Bias

Finally, repulsion can emerge from the brain's attempt to make sense of multiple objects. In Causal Inference in Multisensory Perception, Ulrik Beierholm, Joshua B. Tenenbaum, and Ladan Shams showed that when the brain receives a sound and a flash, it must decide if they share a single cause.

If the signals are close together, the brain integrates them (attraction). But if they are far apart, the brain decides they are independent. This creates a selection bias: we only categorize signals as "separate" when the internal noise happens to push them further apart than they actually are. Consequently, whenever we perceive two things as distinct, we are likely to overestimate the distance between them.

Summary: A Feature, Not a Bug

While Tom Putzeys and Matthias Bethge note in A New Perceptual Bias Reveals Suboptimal Population Decoding of Sensory Responses that these processes can be "blurry" or suboptimal in specific tasks, the broader research direction suggests a deeper logic. Perceptual repulsion is the necessary byproduct of a brain that is:

  1. Efficient: Concentrating precision where it matters most.
  2. Consistent: Conditioning future estimates on past commitments.
  3. Causal: Actively deciding which signals in a noisy world belong together.

Papers in this version

History & citationVersion 2 · 2026-10-04

Current · Permanent link

  1. Version 2 (latest) · 2026-10-04 · 12 papers
  2. Version 1 · 2026-10-04 · 10 papers
Cite this direction

Latest version (a living page):

@misc{lacuna_perceptual_selection_bias_cf0be0ed,
  title = {perceptual selection bias},
  author = {{Lacuna}},
  howpublished = {Lacuna research direction},
  year = {2026},
  url = {https://lacuna.tiptreesystems.com/direction/perceptual-selection-bias/rdr_cf0be0ed2a80487a8433ad354c0ee2fa},
  note = {Latest version of a living research direction; cite a version for a fixed text. Accessed 2026-11-20},
}

This exact version (version 1, frozen text):

@misc{lacuna_efficient_coding_and_the_bayesian_origin_cf0be0ed_v1,
  title = {Efficient Coding and the Bayesian Origin of Perceptual Repulsion},
  author = {{Lacuna}},
  howpublished = {Lacuna research direction},
  year = {2026},
  url = {https://lacuna.tiptreesystems.com/direction/efficient-coding-and-the-bayesian-origin-of-perceptual-repulsion/rdr_cf0be0ed2a80487a8433ad354c0ee2fa/v/1},
  note = {Version 1, frozen 2026-10-04. Accessed 2026-11-20},
}