Comparison of sequential data analysis and functional data analysis for locomotor adaptation.

Area of Science:

• Cognitive Science
• Computational Neuroscience
• Machine Learning

Background:

• Human skill learning rates can fluctuate based on external factors and time.
• Incentives, such as rewards or punishments, significantly impact the speed and efficiency of skill acquisition.

Purpose of the Study:

• To model dynamically changing learning rates in human skills.
• To investigate the influence of incentivization on these learning rates.
• To address the computational challenges in estimating learning rate parameters from extensive data.

Main Methods:

• A state-space model was employed to represent dynamic learning rates.
• A dynamically weighted particle filter, an efficient sequential Monte Carlo method, was utilized for parameter estimation.
• Functional data analysis was explored as an alternative approach for analyzing learning rates and incentivization effects.

Main Results:

• Both the dynamically weighted particle filter and functional data analysis yielded reasonable estimations of learning rates.
• The study presents estimated learning rates and quantifies the impact of incentivization.
• Comparisons between the two analytical approaches were conducted.

Conclusions:

• Dynamically changing learning rates are a key aspect of human skill acquisition.
• Efficient computational methods like the dynamically weighted particle filter can overcome the burden of parameter estimation.
• The findings provide a framework for understanding how incentives modulate learning dynamics.