A computational framework for physics-informed symbolic regression with straightforward integration of domain...

Area of Science:

• Computational Physics
• Data Science
• Scientific Discovery

Background:

• Discovering symbolic expressions for experimental data is a key scientific challenge.
• Existing symbolic regression (SR) methods often lack domain knowledge integration.

Purpose of the Study:

• Introduce Scientist-Machine Equation Detector (SciMED), an open-source framework.
• Integrate scientific wisdom with state-of-the-art SR methods in a scientist-in-the-loop approach.

Main Methods:

• SciMED employs a genetic algorithm-based wrapper selection, automated machine learning, and dual-level SR.
• Tested on settling sphere dynamics with varying complexity and noise levels.

Main Results:

• SciMED successfully discovers physically meaningful symbolic expressions from noisy data.
• Demonstrates superior performance compared to state-of-the-art SR software, even without domain knowledge.
• Highlights the benefit of integrating domain knowledge for enhanced performance.

Conclusions:

• SciMED is a robust tool for scientific equation discovery.
• The framework effectively integrates domain knowledge to improve SR performance.
• SciMED offers an advantage by alerting users to potential missing features in the data.