Regensburg 2025 – scientific programme

Parts | Days | Selection | Search | Updates | Downloads | Help

SYMD: Symposium AI-driven Materials Design: Recent Developments, Challenges and Perspectives

SYMD 1: AI-driven Materials Design: Recent Developments, Challenges and Perspectives

SYMD 1.4: Invited Talk

Monday, March 17, 2025, 16:45–17:15, H1

Inverse Design of Materials — •Hongbin Zhang — Institute of Materials Science, TU Darmstadt, 64287 Darmstadt, Germany

Machine learning has been widely applied to obtain statistical understanding and rational design of advanced materials to map out the processing-(micro-)structure-property-performance relationships, mostly in the forward manner. In this work, focusing on the structure-property relationships, I am going to introduce the concept of inverse design and to showcase how it can be carried out based on Bayesian optimization and generative deep learning. To explore a well-defined and possibly vast design space efficiently, Bayesian optimization can be applied for reliable recommendations, either based on ranking schemas balancing exploration and exploitation or by using proper sampling strategies. This leads to a closed loop adaptive design strategy, which can be integrated with theoretical scale-bridging simulations and experimental synthesis and characterization, resulting in a domain expertise- and physics-informed active learning paradigm. Furthermore, to go beyond the known design space, generative deep learning (such as GAN, VAE, and diffusion models) can be applied. I will demonstrate such a strategy for the polycrystalline microstructure-property mapping, with the physical properties constrained based on an integrated ControlNet in stable diffusion models.

Keywords: inverse design; diffusion models; Bayesian optimization; generative deep learning