NOMATEN HYBRID-SEMINAR September 24: Accelerating materials research with machine learning

NOMATEN HYBRID-SEMINAR

online: https://meet.goto.com/NCBJmeetings/nomaten-seminar
In-person: NOMATEN seminar room (102)

Wednesday, SEPTEMBER 24^th  2025 1 PM (CET)

Accelerating materials research with machine learning

Dario Massa, Ph.D.
(1) IDEAS Research Institute, Warsaw, Poland; (2) NOMATEN, NCBJ

Abstract:

This talk presents a comprehensive approach to computational materials science, integrating machine learning with first-principles methods to address the challenges of predicting material properties and dynamic processes. We introduce a framework for alloy informatics that utilizes electron charge density profiles as bias-free descriptors to capture defect-crystal interactions and correlate with a wide range of atomic properties, as detailed in our work on hydrogen interstitials in face-centered cubic crystals [1]. Building on this foundation, we explore transfer learning and multimodal approaches, demonstrating how knowledge from existing pre-trained neural networks and large language models can be effectively applied to physics domain data for efficient structure-property predictions [2]. Finally, we extend this methodology to dynamic phenomena, showcasing how accelerated machine learning molecular dynamics simulations employing active learning and fine-tuning can accurately predict hydrogen diffusion coefficients and activation energies, aligning remarkably with experimental results while significantly reducing computational costs [3]. Together, these studies highlight the power of combining innovative, physics-informed descriptors with advanced machine learning techniques to enhance the effectiveness of materials property prediction and simulation.

References:

1. Massa, D., Kaxiras, E., & Papanikolaou, S. (2024). Alloy informatics through ab initio charge density profiles: Case study of hydrogen effects in face-centred cubic crystals. Acta Materialia, 268, 119773.
2. Massa, D., et al. (2024). Transfer Learning in Materials Informatics: structure-property relationships through minimal but highly informative multimodal input. arXiv preprint arXiv:2401.09301.
3. Angeletti, A., et al. (2025). Hydrogen diffusion in magnesium using machine learning potentials: a comparative study. npj Computational Materials, 11(1), 85.

Bio:

Dario Massa obtained his B.Sc. in Physics from University of Rome La Sapienza and his M.Sc. in Theoretical Condensed Matter Physics from the University of Padova, Italy. In Padova, he won a fully funded fellowship for the theoretical research on the modelling of van der Waals interactions. He recently obtained his Ph.D. with honors in Physics from the University of Warsaw, defending his thesis containing the results of three years of research performed at NOMATEN CoE, with title Quantum Informed Descriptors in Materials Informatics. His research work is the result of a tight collaboration with multiple renewed institutions, including Harvard University, MIT, University of Vienna, University of Bologna, University of Warsaw, Technical University of Poznan. Currently, he is a Senior Research and Technical Specialist in the new IDEAS Research Institute in Warsaw, working on AI4Science topics involving deep learning interatomic potentials.