Zobrist hash-based duplicate detection in symbolic regression.

Area of Science:

• Computational Science
• Machine Learning
• Physical Sciences

Background:

• Symbolic regression aims to find optimal functions for data without predefined models.
• Genetic programming (GP) is a dominant evolutionary search technique for symbolic regression.
• Standard GP implementations suffer from redundant computations, wasting computational resources.

Purpose of the Study:

• To analyze the efficiency of evolutionary search in genetic programming.
• To introduce a novel caching mechanism to mitigate redundant computations in GP.
• To improve the runtime performance of symbolic regression without compromising search quality.

Main Methods:

• Introduced a caching mechanism based on Zobrist hashing.
• Implemented the caching approach within the open-source Operon framework.
• Evaluated performance on various real-world regression problems.

Main Results:

• Observed significant speedups, up to 34%, in computational time.
• Demonstrated that the caching mechanism does not negatively impact search quality.
• The Zobrist hashing approach efficiently avoids re-evaluation of previously computed states.

Conclusions:

• Zobrist hashing provides a straightforward method to enhance GP runtime performance.
• This caching strategy offers potential for adaptive search adjustments based on historical data.
• The technique is particularly relevant for symbolic regression applications in the physical sciences.