Deficits in Prediction Ability Trigger Asymmetries in Behavior and Internal Representation

Area of Science:

• Cognitive Neuroscience
• Computational Psychiatry
• Developmental Neuroscience

Background:

• Predictive coding explains perception and behavior by integrating sensory input with prior predictions.
• Deviations in predictive coding precision are linked to developmental disorders.
• Computational models offer insights into how aberrant prediction affects development.

Purpose of the Study:

• To computationally test if modifying prediction ability generates a spectrum of network representations mirroring typical to disordered development.
• To investigate how altered reliance on sensory versus prior signals impacts learning and internal representations.

Main Methods:

• Recurrent neural networks were trained to draw figure trajectories.
• Parametric modifications were applied to the networks' reliance on sensory and prior signals during learning.

Main Results:

• Altering the balance between sensory and prior signals affected network performance and internal representations.
• Both excessive and insufficient reliance on predictions impaired network representations.
• Behavioral performance (drawing) did not consistently reflect the impairment in internal representations.

Conclusions:

• Aberrant predictive coding leads to asymmetries between behavioral output and internal representations.
• These findings support the hypothesis that altered prediction reliance may underlie the diverse symptoms of autism spectrum disorder.