Piecewise power laws in individual learning curves

Area of Science:

• Cognitive Psychology
• Learning Sciences
• Computational Neuroscience

Background:

• The traditional power law (PL) model assumes smooth, continuous learning gains.
• Averaging individual learning curves can obscure complex, non-linear dynamics.
• Understanding individual learning trajectories is crucial for skill acquisition theories.

Purpose of the Study:

• To investigate if a piecewise power law (PPL) model better describes individual learning curves than a single PL.
• To analyze the dynamics and characteristics of transitions between learning phases.
• To explore factors influencing these learning transitions, such as age.

Main Methods:

• Analysis of 25,280 individual learning curves from four cognitive tasks, each with 500 performance measurements.
• Comparison of model fit between a single PL and a PPL model, controlling for complexity.
• Examination of performance changes and transition rates between PPL components.

Main Results:

• The PPL model provided a significantly better fit to individual learning curves than the single PL model.
• Transitions between learning pieces showed an initial performance drop followed by higher performance.
• The rate of transition between learning pieces was negatively correlated with age.

Conclusions:

• Individual learning is characterized by both smooth, within-strategy improvement (PL) and discrete, between-strategy shifts (PPL).
• The PPL model offers a more nuanced understanding of human learning dynamics.
• These findings have implications for optimizing skill acquisition and refining learning theories.