Berlin 2024 – wissenschaftliches Programm

Bereiche | Tage | Auswahl | Suche | Aktualisierungen | Downloads | Hilfe

MM: Fachverband Metall- und Materialphysik

MM 11: Data Driven Material Science: Big Data and Workflows II

MM 11.1: Vortrag

Montag, 18. März 2024, 15:45–16:00, C 243

Leveraging Multi-Fidelity Data In AI-Driven Sequential Learning of Materials Properties: Identifying Stable Water-Splitting Catalysts — •Akhil S. Nair, Lucas Foppa, and Matthias Scheffler — The NOMAD Laboratory at the FHI of the Max-Planck-Gesellschaft and IRIS-Adlershof of the Humboldt-Universität zu Berlin, Germany

The sequential learning of materials properties can enable a cost-effective materials discovery by iteratively extending the training data guided by an AI model [1]. Such an approach balances the exploitation of the model and the exploration of unvisited regions of the materials space. However, the efficiency of sequential learning relies on the performance of the AI model and on the quality of the data used to train the models. In material science, high-quality data is typically scarce. To address this challenge, we develop a sequential learning framework which utilizes low-fidelity data to improve the performance of the AI models for high-fidelity materials properties. In particular, we employ the symbolic regression based sure-independence screening and sparsifying operator (SISSO) method, which is suitable for small data sets and can better capture the behaviour of unseen materials compared to widely used AI methods [2]. Our approach is demonstrated for the discovery of stable oxide catalysts for water splitting, a process of significant importance in sustainable hydrogen production. For this, low and high-fidelity data are obtained from DFT-PBE and DFT-HSE calculations, respectively.

Keywords: Sequential learning; Multi-fidelity data; Symbolic regression; Water splitting; DFT