Naujienos ir renginiai
- Titulinis
- Naujienos ir renginiai
- Renginiai ir skelbimai
Renginiai ir skelbimai
„Universal machine learning based interatomic interaction potentials for crystal structure validation and generative models“. Justinas Šlepavičius
Elektrocheminės medžiagotyros seminaras:
Pristato: Justinas Šlepavičius
FTMC Energijos Elektrocheminės Konversijos Laboratorija
VU Ekscelencijos Centras „Duomenų centras mašininiam mokymui ir kvantiniams skaičiavimams gamtos ir biomedicinos mokslų srityse“
2024 m. gegužės 31 d., 14:00 val. E302 auditorijoje, Saulėtekio al. 3, Vilnius
Apie seminarą:
Crystallography Open Database (COD) is an open database containing experimental, theoretical and hybrid crystal structures of many different types materials: metals, organics, metalorganics, inorganics. Prior to the deposition in the COD, these structures need to be validated. One solution is to use physical based modelling such as Density Functional Theory (DFT) to ensure that the proposed crystal is physically reasonable. However, yet powerful, DFT is very computationally expensive. Recently, various Machine Learning (ML) methods that can learn interatomic interactions emerged. One of them, M3G-Net is a graph neural network that incorporates 3 body interactions and is able to describe the behaviour of elements up to actinium (Z=89). M3G-Net is pre-trained on DFT results from many databases is able to effectively model previously unseen systems. This model is computationally low cost and, therefore, much faster compared to DFT. The recent results on applying this methodology for validating problematic cases in the COD will be reviewed.
