Improving Accuracy and Temporal Resolution of Learning Curve Estimation for within- and across-Session Analysis

Area of Science:

• Cognitive Science
• Neuroscience
• Animal Behavior

Background:

• Learning curve estimation commonly uses moving windows of correct response proportions.
• This method assumes stable performance within windows, which is often violated.
• Violations lead to systematic errors in analyzing learning dynamics.

Purpose of the Study:

• To demonstrate errors caused by violated assumptions in conventional learning curve analysis.
• To explore how window size, statistical models, and learning phase influence these errors.
• To propose improved statistical methods for accurate learning curve estimation.

Main Methods:

• Analysis of errors in conventional moving window methods for learning curves.
• Development and application of new trial-by-trial statistical methods.
• Testing methods on rodent avoidance learning data.

Main Results:

• Violations of the constant performance assumption introduce systematic errors.
• Proposed methods accurately estimate learning curves and confidence intervals trial-by-trial.
• Revealed multi-scale performance changes in rodent learning data, obscured by conventional analysis.

Conclusions:

• Accurate, high-resolution assessment of learning dynamics refines existing learning concepts.
• Improved methods disambiguate the relationship between neurophysiological signals and learning.
• Offers a more precise understanding of behavioral and neural processes during learning.