Data quantity governance for machine learning in materials science.

Area of Science:

• Materials Science
• Data Science
• Machine Learning

Background:

• Machine learning (ML) is crucial for materials science, aiding in structure-activity relationship analysis, performance optimization, and materials design.
• A significant challenge in applying ML to materials science is the scarcity of data, leading to a mismatch between feature space dimensionality and sample size, or between model parameters and sample size.
• This data limitation often results in poor ML model performance.

Purpose of the Study:

• To review existing strategies for addressing data limitations in ML for materials science, including feature reduction, sample augmentation, and specialized ML approaches.
• To highlight the importance of balancing sample size with feature dimensionality or model parameters in data quantity governance.
• To propose a novel synergistic data quantity governance framework incorporating materials domain knowledge.

Main Methods:

• Literature review of techniques for tackling data scarcity in materials ML.
• Analysis of the interplay between sample size, feature space, and model parameters.
• Development of a data quantity governance flow integrating materials domain knowledge.

Main Results:

• Existing methods like feature reduction and sample augmentation are discussed as ways to mitigate data limitations.
• The critical need for careful consideration of the balance between data quantity and model complexity is emphasized.
• Incorporating materials domain knowledge into ML data governance schemes demonstrates significant advantages.

Conclusions:

• Effective data quantity governance is essential for successful ML applications in materials science.
• A synergistic approach combining data governance with materials domain knowledge can significantly enhance ML model performance.
• This work provides a pathway to generating high-quality data, accelerating ML-driven materials design and discovery.