In Pursuit of the Exceptional: Research Directions for Machine Learning in Chemical and Materials Science

Area of Science:

• Materials Science
• Computational Chemistry
• Machine Learning

Background:

• Exceptional molecules and materials offer technological value and fundamental insights, often stemming from novel phenomena or compositions.
• Historically, discovery relied on serendipity, but machine learning (ML) and automation are now proposed to accelerate identification and synthesis.
• Current data-driven ML methods excel at optimization but struggle with discovering truly novel, exceptional materials.

Purpose of the Study:

• To argue that conventional ML approaches are insufficient for discovering new exceptional materials.
• To propose a shift in ML methodology towards identifying outliers rather than optimizing existing properties.
• To provide actionable recommendations for developing ML methods capable of exceptional materials discovery.

Main Methods:

• Analysis of case studies involving high-temperature (high-Tc) oxide superconductors and superhard materials.
• Critique of traditional ML approaches in the context of materials discovery.
• Formulation of six recommendations for advancing ML-driven discovery of exceptional materials.

Main Results:

• Demonstration of challenges in ML-guided discovery using oxide superconductors and superhard materials.
• Identification of limitations in current automation for discovering exceptional materials.
• Proposal of six key principles for developing effective ML methods for outlier discovery.

Conclusions:

• Conventional ML methods are better suited for optimizing known material properties than for discovering novel exceptional materials.
• A paradigm shift in ML is required, focusing on exploring extrema and embracing the unexpected.
• Integrating proposed recommendations into automated workflows can enable the discovery of groundbreaking molecules and materials.