Enabling identification of component processes in perceptual learning with nonparametric hierarchical Bayesian...

Area of Science:

• Cognitive Psychology
• Neuroscience
• Machine Learning

Background:

• Perceptual learning involves complex processes like consolidation and relearning, often obscured by traditional learning curve analyses.
• High temporal resolution is crucial for dissecting these component processes in perceptual learning.

Purpose of the Study:

• To develop and compare novel nonparametric inference procedures for analyzing perceptual learning.
• To assess the ability of these methods to resolve component processes within the learning curve.

Main Methods:

• Developed a Bayesian inference procedure (BIP) and a hierarchical Bayesian model (HBM) for estimating contrast thresholds.
• Applied both methods to Gabor orientation identification data across six sessions with varying block sizes (L=10 to 320).
• Utilized nonparametric inference to analyze learning curves with high temporal resolution.

Main Results:

• The HBM provided significantly better fits, smaller standard deviations, and more precise estimates than the BIP across all block sizes.
• HBM yielded unbiased estimates, while BIP showed bias with smaller block sizes.
• Small block sizes (L=10, 20, 40) with HBM successfully identified general learning, between-session forgetting, and within-session relearning.

Conclusions:

• The nonparametric HBM offers a robust framework for detailed assessment of perceptual learning.
• This method enables fine-grained identification of component processes within perceptual learning.
• HBM advances the analysis of learning curves, providing deeper insights into cognitive processes.