Materials Intelligence Lab (MI Lab) uses the most advanced AI technology to challenge itself to the most traditional area of science - new materials development.
As can be seen from many researches on protein structure and biotech firms’ vaccine researches, utilizing data and AI technology can help overcome the limits of existing methods of research, reduce research times, and enhance the overall research activities of scientists. We utilize material information from various open DBs, textual information from past papers or patents, data accumulated from direct experiments, etc. to ascertain the relationship between the structure, composition, and performance of materials and use this information to suggest new materials with optimized performance. In addition, we utilize the acquired data in a way that transmits and analyzes the unique information relating to materials, interpreting the ultimately analyzed results and converting them to scientific knowledge. Many LG subsidiaries including LG Electronics, LG Chem, and LG Display utilize functional materials to produce various products. MI Lab accelerates material development using the AI technology, allowing more LG products to change the lives of our customers.
PredictionAI models that have been taught various material information using experiments, calculations, simulations, etc. predict the chemical, mechanical, and electrical properties of materials and even the movements of electrons and combinations of proteins with high accuracy. In addition, it expands the utilization of molecule generation models by predicting the synthesizability using only the structural information of the material.
OptimizationA genetics algorithm combined with material performance prediction models can be used to quickly find global optimums, thereby quickly scanning materials close to target performance. In addition, active learning based on Bayesian Optimization is used, and an optimal test design that accelerates the new material screening process is suggested.
GenerationData regarding various materials is utilized to produce distribution based on structure-performance relationships. New, unlearned structures are produced by generating new samples within the same distribution. VAE, GAN, and other generative models are used this way and combined with RL to design materials with the desired performance.
RepresentationThe actual information of given material is preserved and compressed as best as possible, finding optimized expressions for conversion for the execution of prediction, classification, generation, and other various AI tasks. Various researches are being carried on low-molecule organic material, high-molecule proteins, and atomic-level expression of inorganic material. As of late, graph-based representation research is pursued actively.
Knowledge ExtractionOpen DB or textual standard (SQL databases) and non-standard (e.g. text, document, image) material data is used to extract meaningful knowledge, information, and relationships and identify patterns.
Meet our leader!
Sehui Han, Head of Materials Intelligence Lab
"Materials Intelligence Lab designs materials using AI. We always use the latest AI technology and experience various real industry materials. We are always waiting for those who are willing to participate in the research on materials AI that will change human life though AI and AI materials that will create a comfortable, prosperous, and pleasant tomorrow."