||Machine learning has rapidly become a practical tool spanning all areas of science including materials sciences. The past few years have witnessed rapid progress in using machine learning for atomistic simulations, materials design and discovery, literature information extraction, and quantum information systems. Several achievements have been made, such as rapidly predicting materials properties, building machine-learning potentials for simulating larger structures with longer time scale, or guiding experimental design. However, despite significant effort, one central question remains unsolved: it is known the number of theoretically stable materials structures will grow hyper-exponentially with the number of atoms in a unit cell, yet, there is only a very small fraction of materials has actually been found.
This symposium envisions to promote the machine learning driven science advancement made in the landscape of condensed matter physics and materials science to push the boundary on materials searching and discovery, that to identify the crux of why some hypothetically exist materials remain to be undiscovered and provide possible solutions with state-of-the-art machine learning architectures. There are gaps in identifying missed information from non-linear large datasets, uncertainty quantification of predictions by surrogate models, optimizing theoretical simulations and experimental findings. Given the vast new opportunities that machine learning offers for understanding materials behaviors ranging from atomistic physics understanding to building advanced technologies such as nuclear reactors, we feel obliged to organize a symposium to address the vast opportunities and challenges applying state-of-the-art machine learning architectures to address key challenges mentioned above with a focus on finding fundamentally new materials.
This symposium will cover recent progress in machine learning-driven materials design by theoretical simulations, automated high-throughput workflows, reduced physics-based surrogate models, and adaptive learning approaches for transferable models. We will cover various types of materials ranging from simple to complex quantum materials. We emphasize the recent progress in machine learning, such as new architectures, new algorithms and workflows that aim to anti-noise, address missing values and with dataset shift. Particular attention will be paid on the strategy on applying machine learning to augment experimental data for novel materials design, as well as the emerging new characterization and analysis tools for complex materials which are not available even a few years ago. The goal of this symposium is to provide an interactive forum to facilitate materials scientists in various fields to quickly digest the exciting recent progress of machine learning and quantum materials with reduced knowledge barrier. Specific sessions will be organized regarding the scientific theme topics rather than with the similarity of a category of materials to benefit cross-fertilization. A couple of sessions will focus on recent methodological advances of the machine learning capabilities to probe the atomistic physics with unprecedented detail.
Symposium topics include, but are not limited to, the following:
Generative models, including GAN, VAE, and diffusion models
Representation of materials
Materials Genome Initiative
Accelerated structure-property relationships