Hierarchical Bayesian Augmented Hebbian Reweighting Model of Perceptual Learning

Area of Science:

• Cognitive Science
• Computational Neuroscience
• Machine Learning

Background:

• The Augmented Hebbian Reweighting Model (AHRM) is established for modeling collective perceptual learning.
• Existing methods often analyze individual or population data separately.

Purpose of the Study:

• Introduce a novel hierarchical Bayesian Augmented Hebbian Reweighting Model (HB-AHRM).
• Simultaneously model individual participant and population learning curves within a unified framework.
• Compare HB-AHRM performance against a Bayesian Inference Procedure (BIP).

Main Methods:

• Developed a hierarchical Bayesian framework (HB-AHRM).
• Implemented a likelihood function approximation using feature engineering and linear regression.
• Achieved a 20,000x speed increase in estimation procedures.
• Computed joint posterior distributions at population, observer, and test levels.

Main Results:

• HB-AHRM successfully models both individual and population learning curves.
• The likelihood approximation drastically reduces computational demands.
• Enables robust statistical inference across hierarchical levels.
• HB-AHRM outperforms BIP in integrated modeling.

Conclusions:

• HB-AHRM provides a powerful, unified framework for perceptual learning analysis.
• The likelihood approximation technique has broad applicability for stochastic models.
• This methodology facilitates accurate prediction of human performance at multiple levels.