Duet model unifies diverse neuroscience experimental findings on predictive coding.

Area of Science:

• Neuroscience
• Computational Neuroscience
• Cognitive Science

Background:

• The brain constantly predicts upcoming sensory information.
• Deviant stimuli elicit amplified responses, indicating surprise or prediction errors.
• Existing models often focus on single prediction error types.

Purpose of the Study:

• To introduce a minimal, biologically plausible computational model of predictive coding.
• To explain how the brain detects unexpected stimuli using dual prediction errors.
• To unify findings across different predictive coding paradigms, especially omission and oddball tasks.

Main Methods:

• Development of the duet predictive coding model, incorporating positive and negative prediction errors.
• Simulation of neural responses within the model.
• Comparison of model predictions with existing experimental data from vision and audition.

Main Results:

• The duet predictive coding model accurately reproduces neural responses in various paradigms, including omission.
• The model predicts the existence of neurons encoding negative prediction errors, supported by mouse studies.
• Response magnitude correlates with stimulus dissimilarity and decreases with interstimulus interval.

Conclusions:

• The brain employs dual-error computation for effective deviance detection.
• This framework offers a unified explanation for responses to unexpected stimuli.
• The model provides a testable circuit mechanism for predictive coding in the brain.