A unifying theory explains seemingly contradictory biases in perceptual estimation

Area of Science:

• Neuroscience
• Cognitive Science
• Computational Vision

Background:

• Perceptual biases offer insights into neural computations.
• Existing theories for perceptual biases are diverse and sometimes contradictory, including Bayesian prior attraction, efficient coding repulsion, and central tendency effects.
• A unified understanding of perceptual biases is lacking.

Purpose of the Study:

• To develop a unifying Bayesian theory for perceptual estimation biases.
• To derive a general principle for predicting the direction and magnitude of perceptual biases.
• To reconcile existing, seemingly contradictory, explanations for perceptual biases.

Main Methods:

• Developed a unifying Bayesian theory for perceptual biases from first principles.
• Theoretically demonstrated an additive decomposition of perceptual biases.
• Analyzed biases in the perception of various stimulus attributes (orientation, color, magnitude).

Main Results:

• Introduced a universal rule for predicting perceptual biases.
• Decomposed perceptual biases into three components: attraction to a prior, repulsion from high-precision encoding, and regression from a boundary.
• The theory successfully accounts for existing findings and provides new insights.

Conclusions:

• The proposed Bayesian theory offers a unified framework for understanding perceptual biases.
• The additive decomposition provides a simple, universal rule for predicting perceptual biases.
• These findings offer crucial constraints for understanding the neural basis of Bayesian computations in perception.