MI-6 Ltd., in collaboration with a research group from National Yang Ming Chiao Tung University, has announced the publication of a research paper proposing new methodologies for enhancing data quality and optimization processes to advance AI utilization in materials development. The paper was published in the international academic journal Materials Today Electronics.
While the application of machine learning for materials discovery (Materials Informatics, or MI) is expanding rapidly, many projects face significant hurdles regarding "data infrastructure," such as data quality, labeling, and accessibility. Consequently, there has been a pressing need to establish standardized methods for measuring objective data quality and increasing model reusability from limited datasets.
In response to these challenges, this study proposes an eight-step MI process. This comprehensive workflow ranges from restructuring data using encoder-decoder technology to introducing a proprietary data quality metric known as the "R:M Ratio" (the ratio of the coefficient of determination R2 to the mean squared error MSE). Furthermore, the process incorporates Bayesian optimization and a specialized "masking process."
Experimental results confirmed that high prediction accuracy and confidence levels are achieved in datasets where the R:M Ratio is 5.0 or higher, or where the Regression Deviation Density Threshold (RDDT) falls between 3% and 8%. Conversely, the study also identified certain constraints; since the method depends on the accuracy of regression analysis, its application may remain a complementary alternative to mainstream discovery methods in cases where the likelihood is not sufficiently high.
These findings provide a clear guideline for data quality management based on objective metrics, leading to more efficient materials design. By integrating this methodology into a common MI platform, the research group expects to facilitate the reuse of knowledge among researchers and accelerate the discovery of innovative new materials.
Paper Information
- Title: The advancement of materials discovery through the applied artificial intelligence(Link)
- Journal: Materials Today Electronics
- Volume/Issue/Year: Volume 15, 2026, 100186
- DOI:https://doi.org/10.1016/j.mtelec.2025.100186
- Authors *Affiliations are based on the information at the time of publication.
- Raymond Wu, Susumu Otsuki (MI-6 Ltd.)
- Haishang Wu (National Yang Ming Chiao Tung University)
