Reproduce “Automated discovery of symbolic laws governing skill acquisition from naturally occurring data” on OpenResearch

About this paper

Sannyuya Liu, Qing Li, Xiaoxuan Shen, Jianwen Sun, Zongkai Yang

This research introduces a two-stage algorithm that combines deep learning with symbolic regression to automatically discover interpretable algebraic laws governing skill acquisition from large-scale, real-world training data. The method successfully identified both known and novel mathematical patterns across various cognitive skills, outperforming classical and contemporary learning models in fitting accuracy and model parsimony.

Key insights

• The integrated deep learning and symbolic regression approach effectively bridges the gap between high predictive accuracy and scientific interpretability for complex, real-world skill acquisition data.
• Analysis revealed that age, education level, and specific types of practice (e.g., attention selection practice) are key factors influencing skill mastery, sometimes more so than general practice frequency.
• The methodology can both reaffirm established learning phenomena, such as the varying fit of power versus exponential laws, and discover novel, non-conventional symbolic laws for specific skills.

Summary by alphaXiv.