Active Learning Improves Performance on Symbolic Regression Tasks in StackGP

This paper introduces an active learning method for symbolic regression using StackGP. The approach begins with a small number of data points for StackGP to model. To improve the model the system incrementally adds the data point characterized by maximizing prediction uncertainty as measured by the model ensemble. Symbolic regression is re-run with the larger data set. This cycle continues until the system satisfies a termination criterion. The Feynman AI benchmark set of equations is used to examine the ability of the method to find appropriate models using as few data points as possible. The approach successfully rediscovered 72 of the 100 Feynman equations without the use of domain expertise or data translation.