AI BriefingLG AI Research: The Convergence of Basic Science and AI
Key point
LG AI Research dramatically shortened new material development timelines using a molecular pre-training model that leverages 3D structures.
Details
Recently, research on AI-driven structure prediction and molecular design has been actively progressing in the fields of chemistry, biology, and pharmaceuticals. LG AI Research is carrying out new material development projects in collaboration with LG affiliates, and in particular, through joint research with LG H&H, achieved the result of developing cosmetic efficacy substances with improved solubility and safety.
In this process, EXAONE Discovery, an AI model specialized in new material discovery, played a key role. Previously, selecting candidate substances based on researchers' experience and papers took about 2 years, but EXAONE Discovery completed this process in just one day.
The paper "3D Denoisers are Good 2D Teachers," which formed the basis of this research, drew attention after being selected as an Oral Paper at AAAI-25. This research focuses on combining AI's inductive approach with the deductive methodology of basic science.
The Materials Intelligence (MI) Lab presented a strategy that increases model accuracy through denoising distillation technology using 3D coordinates, and enables the implementation of models with new properties even in low-data environments.
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.