Active oversight and quality control in standard Bayesian optimization for autonomous experiments

TL;DR

This paper introduces a Dual-GP method with human oversight to improve Bayesian optimization in autonomous experiments, addressing data quality issues and enhancing efficiency in materials research.

Contribution

The paper presents a novel Dual-GP approach that adds a secondary surrogate model for real-time data assessment and oversight in Bayesian optimization workflows.

Findings

Dual-GP improves optimization efficiency by focusing on promising experimental regions.

The approach effectively filters low-quality data, preventing misleading results.

Human-in-the-loop intervention enhances adaptability to unanticipated outcomes.

Abstract

The fusion of experimental automation and machine learning has catalyzed a new era in materials research, prominently featuring Gaussian Process Bayesian Optimization (GPBO) driven autonomous experiments navigating complex experimental conditions for accelerated scientific discovery. In traditional GPBO-driven experiments, a predefined scalarizer function is often required to preprocess the experimental data, transforming non-scalar raw data into scalar descriptors for GP training. However, such predefined scalarizer functions have limitations, which likely fail to accommodate the diversity and complexity of real-world experimental data, potentially skewing experimental outcomes. Thus, oversight and quality control are necessitated over the process to avoid GPBO from being misled by low quality scalarizers. To address the limitation, we introduce a Dual-GP approach that enhances traditional GPBO by adding a secondary surrogate model to dynamically constrain the experimental space based on real-time assessments of the raw experimental data. This Dual-GP approach enhances the optimization efficiency of traditional GPBO by isolating more promising space for BO sampling and more valuable experimental data for primary GP training. We also incorporate a flexible, human-in-the-loop intervention method in the Dual-GP workflow to adjust for unanticipated results. We demonstrate the effectiveness of the Dual-GP model with synthetic model data and implement this approach in autonomous pulsed laser deposition experimental data. This Dual-GP approach has broad applicability in diverse GPBO-driven experimental settings, providing a more adaptable and precise framework for refining autonomous experimentation for more efficient optimization.