Uncertainty estimation with prediction-error circuits.

Area of Science:

• Neuroscience
• Computational Neuroscience
• Perception

Background:

• Neural circuits integrate sensory data with predictions, often facing conflicting inputs.
• Accurate integration requires estimating the uncertainty of both sensory stimuli and internal predictions.
• The mechanisms by which the brain tracks these uncertainties remain largely unknown.

Purpose of the Study:

• To elucidate how neural circuits estimate sensory and prediction uncertainty.
• To investigate the role of prediction-error neurons in uncertainty processing.
• To link uncertainty estimation to perceptual biases.

Main Methods:

• Development of a hierarchical prediction-error network model.
• Simulation of neural responses to noisy sensory stimuli and predictions.
• Perturbation of inhibitory interneurons within the model to assess their function.
• Analysis of model output to identify contributions to perceptual biases.

Main Results:

• A hierarchical prediction-error network can successfully estimate both sensory and prediction uncertainty.
• Positive and negative prediction-error neurons play distinct roles in this estimation.
• The model demonstrates that circuits rely more on predictions during noisy sensory input and stable environments.
• Inhibitory interneuron perturbation reveals their critical role in uncertainty processing and input weighting.
• Model simulations link stimulus and prediction uncertainty to the observed contraction bias in perception.

Conclusions:

• Hierarchical prediction-error networks provide a viable mechanism for estimating neural uncertainty.
• Uncertainty estimation is crucial for dynamically weighting sensory and predictive information.
• The findings offer insights into the neural basis of perceptual biases, specifically contraction bias.