Xinyan Liu
Institute of Fundamental and Frontier Sciences, University of Electronic Science and Technology of China, Chengdu 611731, China
EXTENDED ABSTRACT: Machine learning (ML) technique has gained much attention in the recent years due to its great potential to uncover hidden correlations from massive data. However, most ML models adopted are black box models, which require data with high quality or large quantity and cannot be interpreted easily. Given the unique characteristics of data in the field of energy chemistry, this talk focuses on establishing non-black box models, which are designed to digest cheap theoretical and experimental data to generate effective and accurate predictions on materials and devices. In addition, by incorporating the insights extracted from the model, the mechanism decoupling and material inverse design can be realized. On the material level, this talk will introduce a ML model that utilized non-ab initio data to predict the catalytic performances of various stepped alloys. Leveraging the information theory, a few design principles can be outlined and utilized to accelerate the catalyst design. On the device level, this talk will introduce the ML model devised for the next-generation conversionbased batteries. This hybrid interpretable ML framework can not only predict battery lifespan accurately, but also disentangle the degradation mechanism effectively, providing new physical understandings for battery optimization. This talk will also introduce a time-series-based online ML model designed to predict the degradation trajectory of lithium metal batteries, using only its historical data.
Keywords: machine learning; interpretability; high-throughput screening; inverse design
REFERENCES
[l] Liu, X. Y.; Peng, H.J.*; Li, B. Q.; Chen, X.; Li, Z.; Huang, J. Q.; Zhang, Q ., Angewandte Chemie International Edition 2022,61,e202214037
[2] Liu, X. Y.; Zhang, X. Q.; Chen, X.; Zhu, G. L.; Yan, C.; Huang, J. Q.; Peng, H.J.*, Journal of Energy Chemistry 2022, 68, 548-555
[3] Liu, X. Y.; Cai, C.; Zhao, W. H.; Peng, H. J. *; Wang, T. *, ACS Catalysis 2022, 12, 4252-4260
Xinyan Liu is a research professor at the university of electronic science and technology of China. She obtained her bachelor and doctoral degree from Tsinghua university and Stanford university in 2013 and 2018, respectively. Her main research focus is the interdisciplinary research of machine learning and energy chemistry, including electrocatalysis, microkinetics, catalyst screening and battery prognosis. She has published more than 40 papers on esteemed journals such as Nat. Commun.、JACS、 Angew. Chem. Int. Ed.