Hypothesis Learning in Automated Experiment: Application to Combinatorial Materials Libraries.

Area of Science:

• Materials Science
• Physics
• Computer Science

Background:

• Machine learning accelerates experimental physical discovery through automation.
• Active learning methods are crucial for efficient exploration of complex parameter spaces.

Purpose of the Study:

• To develop an active learning approach for optimizing experimental exploration.
• To combine hypothesis and experimental spaces for efficient scientific discovery.

Main Methods:

• Introduced an active learning approach navigating hypothesis and experimental spaces.
• Combined structured Gaussian processes (probabilistic models) with reinforcement learning (policy refinement).
• Applied the method to study concentration-induced phase transitions in Sm-doped BiFeO3 using piezoresponse force microscopy.

Main Results:

• Demonstrated efficient exploration of parameter spaces for material discovery.
• Successfully identified concentration-induced phase transitions in Sm-doped BiFeO3.

Conclusions:

• The proposed active learning framework enhances the efficiency of physical discovery.
• The approach is extensible to higher-dimensional parameter spaces and complex physical systems.