Physics Intermediate Representation (PIR): A Structured Framework for Automated Physical Law Discovery from Data

Key Points

• PIR aims to automate the discovery of physical laws from data by providing a structured, interpretable framework.
• Employs dimensional filtering to refine data analysis
• Uses residual refinement to improve model accuracy
• Incorporates sparse model selection for optimal interpretations
• Applies a hybrid loss function combining MSE and Optimal Transport
• Achieves 100% success on core benchmark tasks
• Recovers interpretable physical laws accurately from data

Abstract