Brian Tung
Speaking Sessions
Unlocking Magnetic Materials Discovery through Generative Modelling with Specialised Data
Magnetic materials play a critical role in technologies across energy, electronics, and data storage. However, identifying and optimising magnetic materials with targeted properties, such as precise Curie temperatures or tailored magnetic anisotropy remains challenging due to experimental and computational complexity. At MatNex, we have leveraged generative models with our specialised datasets to unlock their capabilities in the discovery of magnetic materials. In this talk, we’ll introduce our materials discovery platform and explain our methods for generating synthetic data, including validation against experimental measurements. We’ll also discuss examples of generative modelling, such as tailoring labels for property conditioning, integrating surrogate models, and hyperparameter selection. Finally, we’ll share our benchmarking results, demonstrating how generative models could accelerate the materials discovery process, reduce the need for extensive experimental testing, and enable quicker development of precisely targeted materials.
Biography
Dr Brian (Po-Yen) Tung specialises in generative models and active learning to accelerate materials discovery. With a PhD from the Max Planck Institute and prior research roles at the University of Cambridge, Brian brings deep expertise in combining physics-based simulations and experimental data to design advanced materials. His work has led to major breakthroughs, including the rapid discovery of high-entropy alloys published in Science, and an active optimisation method published in Nature Computational Science. Brian also develops cutting-edge ML techniques for 3D electron microscopy and spectral analysis, and contributes strong research and engineering capability to MatNex’s ML efforts.