Hierarchical approximate Bayesian computation.

Area of Science:

• Computational Neuroscience
• Cognitive Psychology
• Statistical Modeling

Background:

• Approximate Bayesian computation (ABC) is vital for models lacking explicit likelihood functions, common in computational neuroscience.
• Conventional ABC struggles with high-dimensional hierarchical models due to computational complexity.
• Extending ABC to hierarchical Bayesian models remains a significant challenge.

Purpose of the Study:

• To summarize existing methods for hierarchical ABC.
• To introduce a novel algorithm, Gibbs ABC, for improved hierarchical model estimation.
• To assess the performance of Gibbs ABC on signal detection models.

Main Methods:

• Review of current hierarchical Approximate Bayesian computation (ABC) approaches.
• Development and implementation of the Gibbs ABC algorithm.
• Application of Gibbs ABC to two signal detection models, one with and one without a tractable likelihood.

Main Results:

• The Gibbs ABC algorithm demonstrates improved accuracy and efficiency for hierarchical model estimation.
• Successful application of Gibbs ABC to complex signal detection models.
• Validation of Gibbs ABC in scenarios with and without explicit likelihood functions.

Conclusions:

• Gibbs ABC offers a robust solution for parameter estimation in high-dimensional hierarchical models.
• This method enhances the applicability of ABC to complex computational models in cognitive science.
• The algorithm provides a more efficient and accurate approach for Bayesian inference in challenging modeling scenarios.