Open Hypothesis Learning Framework for Autonomous Microscopy
A new framework integrating symbolic regression with large language model (LLM) evaluation enables autonomous scanning probe microscopy to generate physical models from experimental data, moving beyond fixed objective spaces. The system produces candidate analytical relationships from sparse measurements, which are then ranked by the LLM for physical plausibility and consistency. Demonstrated on piezoresponse force microscopy, this approach opens the door to hypothesis learning in materials discovery.
Key facts
- Framework combines symbolic regression with LLM-based physical evaluation
- Designed for autonomous scanning probe microscopy
- Generates candidate analytical relationships from sparse measurements
- LLM ranks candidates by physical plausibility, scaling behavior, and consistency
- Demonstrated on autonomous piezoresponse force microscopy
- Overcomes limitations of fixed objective or hypothesis spaces
- Enables closed-loop optimization in imaging and spectroscopy tuning
- Published on arXiv (2605.06839)
Entities
Institutions
- arXiv