Explainability and human intervention in autonomous scanning probe microscopy.

Area of Science:

• Materials Science
• Artificial Intelligence
• Experimental Design

Background:

• Widespread adoption of machine learning (ML)-based autonomous experiments (AEs) necessitates robust strategies for workflow analysis and intervention.
• Current methods lack real-time indicators for understanding the progression of ML-driven experimental processes.

Purpose of the Study:

• To introduce and demonstrate a post-experimental analysis strategy for deep kernel learning-based autonomous scanning probe microscopy.
• To provide real-time and post-experimental indicators for active learning processes in AEs.
• To illustrate the applicability of this strategy to human-in-the-loop AEs.

Main Methods:

• Developed a post-experimental analysis strategy tailored for deep kernel learning in autonomous scanning probe microscopy.
• Implemented real-time and post-experimental indicators to monitor active learning progression.
• Demonstrated the strategy's integration with human-in-the-loop AEs for high-level policy setting and low-level decision-making.

Main Results:

• The proposed strategy yields effective real-time and post-experimental indicators for active learning in AEs.
• Successfully applied the approach to human-in-the-loop autonomous experiments, enabling efficient human-AI collaboration.
• Validated the universality of the approach for diverse material characterization and synthesis applications.

Conclusions:

• The developed post-experimental analysis strategy enhances understanding and control of ML-based AEs.
• This approach facilitates seamless integration of human expertise into autonomous experimental workflows.
• The strategy is adaptable to various experimental techniques, including combinatorial library analysis.