Serial reproduction reveals the geometry of visuospatial representations

Area of Science:

• Cognitive Neuroscience
• Computational Vision
• Psychology

Background:

• The human visual system navigates by integrating sensory data with internal beliefs about spatial locations.
• Limited resources lead to systematic distortions and biases in spatial perception and memory.
• Bayesian models represent these internal beliefs as prior probability distributions over scene locations.

Purpose of the Study:

• To introduce a novel paradigm for measuring spatial priors in the human visual system.
• To investigate the detailed structure of these priors and their implications for understanding spatial memory biases.
• To test the hypothesis that spatial encoding is characterized by efficient allocation of resources with variable precision.

Main Methods:

• A memory task was iterated, where each participant's response served as the stimulus for the next participant (information transmission paradigm).
• This crowdsourced approach allowed for nonparametric, data-driven measurement of prior probability distributions over locations.
• Perceptual biases and discrimination accuracy were measured and compared with model simulations.

Main Results:

• The paradigm revealed rich and detailed spatial prior distributions, offering new insights into spatial memory biases.
• Findings support an "efficient encoding" model where coding resources are selectively allocated, leading to variable encoding precision across locations.
• Perceptual biases were found to covary with variations in discrimination accuracy, aligning with efficient encoding predictions.

Conclusions:

• The study introduces a powerful method for uncovering the structure of shared visual representations using crowdsourcing and controlled information transmission.
• Results challenge traditional views of fixed encoding precision, highlighting the role of efficient, location-dependent resource allocation in spatial cognition.
• This work opens new avenues for understanding biases in spatial memory and the underlying neural mechanisms.