Why does seeing a small thing make the next one look 'large'?
The tendency to classify an item as the opposite of the previous one seems to come from the brain shifting its internal small/large boundary, not from the eye's sensory adaptation.
Source
Neural Evidence for Boundary Updating as the Source of the Repulsive Bias in Classification
Study at a glance
- Design
- Human experiment — Within-subject fMRI classification task (small vs large ring, three threshold-level sizes in m-sequence order, no trial feedback); Experiment 1 imaged V1 at high resolution, Experiment 2 imaged the whole brain and decoded latent variables of a fitted Bayesian boundary-updating model with searchlight regression.
- N
- Experiment 1: 19 adults; Experiment 2: 18 adults; 17 people took part in both, so no single pooled N is analysed.
- Population
- Healthy adults aged 20 to 30 at Seoul National University, trained on the task beforehand
- Outcome
- Repulsive bias in choices; adaptation of the V1 size-encoding signal; decoded brain signals of class boundary, inferred stimulus and decision variable, and whether their previous-stimulus-related variability predicts current choice
Structured fields used in claim comparison tables when every cited study has a complete layer.
What they did
Participants judged whether briefly flashed rings were small or large, using three sizes that differed by just their discrimination threshold and no trial-by-trial feedback. In Experiment 1 (19 people) the researchers imaged primary visual cortex at high resolution and read out a ring-size signal to see whether it adapted to the previous ring and whether that adaptation fed into choices. In Experiment 2 (18 people) they imaged the whole brain, fitted a Bayesian model in which the class boundary drifts toward recent rings, and searched for brain patterns tracking the model's boundary, perceived size and decision variable, testing each candidate against a long list of regressions implied by the model.
What they found
Both groups showed strong repulsive bias: after a small ring, the next ring was more often called large. The V1 size signal did adapt to the previous ring, but its adaptation-related variability did not predict current choices, which instead tracked only the current ring. The boundary-updating model reproduced the human bias closely (R-squared of 0.89 for the previous-stimulus effect), and signals of its moving boundary were found in the left inferior parietal lobe and posterior superior temporal gyrus. Unlike V1, these boundary signals' link to choice weakened when the previous stimulus was controlled, as the boundary-updating account predicts.
The limits
What it doesn't show
The samples were small and largely the same people across experiments, and the boundary signals rely on latent variables estimated from a model built with several simplifying assumptions (for example exponential memory decay). The analyses are correlational (average marginal effects and model-derived regression tests), so they cannot prove the boundary signal causes the bias; stimulation studies would be needed. The model ignores effects of previous decisions, and because choices and hand responses covaried the authors could not separate choice history from motor history. fMRI's coarse resolution may also explain why no decision-variable signal appeared in prefrontal cortex.
Key terms
- Repulsive bias
- A history effect in which an item tends to be classified as the category opposite to the preceding items.
- Sensory adaptation
- Reduced responsiveness of sensory neurons tuned to a recently seen stimulus, which can distort how the next stimulus is represented.
- Class boundary
- The internal criterion that divides a continuous property into two categories, such as the typical size separating small from large.
- Bayesian boundary-updating model
- A model in which the observer infers the class boundary from remembered recent stimuli, weighting recent ones more because older memories are noisier.
- Searchlight decoding
- An fMRI method that slides a small sphere across the brain and tests whether the local multivoxel pattern can predict a variable of interest.
- Average marginal effect (AME)
- The average change in the probability of a choice associated with a change in a predictor, used here to compare predictors across logistic models.
Flashcards
0 of 11 answers reviewed
Research intelligence for this paper
See its role on concept claims, tensions it is part of, placement history, and related discoveries.
Quiz yourself
According to the boundary-updating hypothesis, why is a medium ring called large after a small ring?
Common questions
Why can't behaviour alone tell the two hypotheses apart?
Classification depends on the relation between the perceived stimulus and the boundary; the same bias arises whether perception is pushed away from the last item or the boundary moves toward it, so neural signals are needed to see which route carries the bias.
If V1 adapted, why doesn't that explain the bias?
Adaptation changed V1's signal, but the part of V1 variability caused by the previous ring did not predict the current choice; the choice-relevant V1 variability came from the current ring.
Why test so many regressions for each brain signal?
A region correlating with the boundary could really be tracking a related variable such as the decision variable. Requiring each signal to satisfy every relationship implied by the model's causal structure guards against that confusion.
More on Perceptual decision