Tracking the contribution of inductive bias to individualised internal models

Area of Science:

• Cognitive Science
• Computational Neuroscience
• Machine Learning

Background:

• Internal models are crucial for humans to adapt to environmental regularities.
• Existing research often assumes an ideal observer model, overlooking individual learning dynamics.
• Humans utilize approximate, adaptable internal models rather than perfect ones.

Purpose of the Study:

• To develop a method for inferring dynamic, individualized internal models from behavioral data.
• To quantify the discrepancy between human internal models and the ideal observer model.
• To track the adaptation of internal models during learning.

Main Methods:

• Combined non-parametric Bayesian methods and probabilistic programming.
• Inferred internal models from response time data.
• Applied the method to an implicit visuomotor sequence learning task.

Main Results:

• Successfully characterized the difference between individual and ideal observer models.
• Tracked the evolving contribution of the ideal observer model during training.
• Identified a consistent, yet variable, inductive bias across individuals.

Conclusions:

• The new approach provides a powerful tool for understanding individual learning processes.
• Discrepancies highlight inherent inductive biases in human learning.
• The method offers insights into the dynamics of model adaptation in visuomotor learning.