Incremental acquisition of multiple nonlinear forward models based on differentiation process of schema model

Area of Science:

• Computational Neuroscience
• Machine Learning
• Cognitive Science

Background:

• The human brain is thought to utilize multiple internal forward-inverse models.
• Existing computational models often rely on Bayesian frameworks for learning multiple models.

Purpose of the Study:

• To introduce the schema model as an alternative computational framework for representing multiple internal models.
• To enable agents to acquire multiple nonlinear forward models incrementally.

Main Methods:

• Developed a schema model based on hypothesis testing theory.
• Implemented a specific example: a schema model with a normalized Gaussian network (NGSM).

Main Results:

• The NGSM demonstrated incremental acquisition of multiple models.
• NGSM showed independence from initial forward model parameters, unlike MOSAIC's method.

Conclusions:

• The schema model provides an effective alternative for incremental learning of multiple internal models.
• NGSM offers significant advantages in model acquisition and parameter independence compared to previous methods.