AI BriefingLG AI Research 378
Key point
Google DeepMind expanded the inorganic materials database tenfold through active learning and discovered more than 380,000 new stable materials.
Details
At the recent MRS (Materials Research Society) conference, AI-driven materials development and experimental automation emerged as key topics. In particular, AI-based Interatomic Potential (IAP) research and the use of LLMs are drawing attention.
Google DeepMind garnered significant attention at the conference by unveiling its GNoME (Graph Networks of Material Exploration) research. This technology uses an active learning-based material generation approach to expand the scale of the existing inorganic materials database by roughly 10x.
Key achievements include:
- Expanding structural data from the previous level of 100,000–200,000 to 2.2 million
- Identifying more than 380,000 new stable materials (including 543 battery-related materials)
- Confirming improvements in the model's stability and energy prediction performance (MAE) as a result of the expanded data
The GNoME potential, trained on the expanded data, is an E(3)-equivariant model that demonstrates excellent predictive performance, accelerating the process of new materials discovery.
This summary was generated automatically by AI. Check the original for the author's claims and context. Copyright belongs to the original author.
Our guide explains how the AI works. Report summary errors, attribution issues, or removal requests via Contact.