[1Fa01-05] Data-driven research and development focused on data generation/storage/analysis: Data-Driven Surface Science Division's Session

[1Fa05] Multimodal Deep Learning for Materials Discovery and Optimization

*Shun Muroga¹ (1. National Institute of Advanced Industrial Science and Technology)

With rising needs for advancements in materials and manufacturing, computational materials science, notably data-centric methods, have expanded. Traditional methods focus on molecular descriptors, aiding molecules, inorganics, and crystal discovery but falter with materials like plastics and alloys due to structural complexities. This research proposes a novel multimodal deep learning method for materials design. It combines diverse data of physical or chemical structures to predict multiple mechanical, thermal, electrical properties for materials design and optimization. In addition, generative models were introduced to represent physical or chemical structure information, such as images or spectra. Based on our proposed method, over 114,210 material compositions were assessed, revealing optimal mixtures and trade-offs. This method presents transformative prospects for materials research, promising revolutionary insights in future materials design.