Deep Learning for Sustainable Materials, Energy Systems, and Climate Adaptation: AI Applications across Scales

Recent advances in machine learning, coupled with the rapid growth of computing power and scientific data infrastructures, led to increasing interest in applying deep learning-based methods for scientific discovery. This presentation will briefly introduce our recent efforts in developing deep learning models and relevant AI tools in the quest for sustainability. Specifically, we will first present the results of several research projects, in collaboration with experimental materials scientists, to illustrate the applications of deep learning on: (a) Integrated design of metal-organic frameworks (MOFs) and adsorption processes via rapid, in-silico screening of MOFs for economically-efficient post-combustion CO2 capture and separation; (b) Understanding hierarchical protein assembly and developing soft sensors based on liquid crystal (LC) systems by analyzing complex spatiotemporal LC optical responses indicating the phase states of multi-protein assemblies formed at interfaces through high-throughput experiments; and (c) Inverse molecular design by developing expressive representations and effective data-driven models for quantitative structure-property relationships and quantum computing-assisted automated molecular structure optimization. The second half of the presentation will focus on the multi-scale applications of deep learning in materials discovery, polymer and protein structure optimization, plastic waste management, natural resource conservation, carbon-neutral energy transition, sustainable food production, reliable power supply, and climate adaptation. The presentation will conclude with our perspectives on integrating research and educational activities and catalyzing interdisciplinary collaborations to unlock unique opportunities toward a paradigm shift in AI for Science.