PIR-Bench: A Reproducible Benchmark Suite for Automated Physical Law Discovery

Key Points

• The aim is to provide a reproducible framework for assessing various physical law discovery methods.
• Developed a benchmark suite with over 20 physics discovery tasks.
• Included five difficulty levels for comprehensive testing.
• Utilized parameterized YAML configuration for task sweeps.
• Implemented SQLite for experiment tracking and management.
• Provided automated reporting features to generate paper-ready results.
• Facilitated systematic evaluation of physical law discovery methods.
• Enabled users to compare performances across multiple tasks.
• Introduced a formal Discovery Confidence Score for assessment.