Pareto-Optimal Experimentation: Human-Guided Multi-Objective Bayesian Optimization in Scanning Probe Microscopy

Area of Science:

• Artificial Intelligence
• Chemistry
• Materials Science
• Robotics
• Scientific Methodology

Background:

• Automated experimentation accelerates discovery but struggles with optimizing conflicting objectives under uncertainty.
• Systematic methods are crucial for managing complex, multi-objective experimental landscapes.

Purpose of the Study:

• Introduce multi-objective Bayesian optimization (MOBO) as a general framework for autonomous experimentation.
• Enable balancing competing rewards and integrating human guidance in scientific optimization.
• Provide a principled approach to explore parameter space and make quantifiable decisions.

Main Methods:

• Developed a multi-objective Bayesian optimization (MOBO) framework.
• Constructed the Pareto front to represent all trade-off solutions.
• Integrated human-in-the-loop control for adjusting objectives and reference points.

Main Results:

• MOBO identifies the Pareto front, detailing trade-offs between partially known reward functions.
• Revealed interdependencies between objectives, enabling systematic exploration.
• Demonstrated MOBO's capability to incorporate expert judgment without halting automation.

Conclusions:

• MOBO transforms experimental optimization from trial-and-error to a reproducible, interpretable process.
• Human guidance enhances MOBO, allowing researchers to steer experiments toward desired outcomes.
• MOBO offers a scalable methodology for efficient, trustworthy self-driving laboratories.