About this Abstract |
Meeting |
TMS Specialty Congress 2025
|
Symposium
|
8th World Congress on Integrated Computational Materials Engineering (ICME 2025)
|
Presentation Title |
NIMS's Data-Driven Materials Research Platform: Enhancing MLOps With Literature-Based Data Integration |
Author(s) |
Toshihisa Anazawa, Hayato Sonokawa, Hideki Yoshikawa, Takuya Kadohira, Satoshi Minamoto |
On-Site Speaker (Planned) |
Toshihisa Anazawa |
Abstract Scope |
NIMS Materials Data Platform (MDPF) supports data-driven materials research through "DICE", a system that comprehensively handles the creation, accumulation, and utilization of materials data. MDPF is currently developing a materials development platform using MLOps by connecting systems named "RDE" (data accumulation), "pinax" (modeling), and "MInt" (execution). Simultaneously, we're training coordinators to promote data-driven materials development.
While pinax can use developers' own data as well as RDE and other public databases, materials are diverse in their applications. Even when properties are reported in academic papers, they are often not databased in a reusable form.
Regarding this, we propose a tool for MLOps that searches related papers, collects desired property data and process conditions from relevant papers, and aggregates them as machine learning data. This system will not only incorporate the aggregated data into pinax for MLOps but also utilize metadata extracted from papers as reference for materials developers and the coordinators. |
Proceedings Inclusion? |
Undecided |