Catalyst deep neural networks (Cat-DNNs) in singlet fission property prediction

• Published in: Physical chemistry chemical physics : PCCP 2021-09, Vol.23 (37), p.2835-284
• Main Authors: Ye, Shuqian, Liang, Jiechun, Zhu, Xi
• Format: Article
• Language: English
• Subjects: Artificial intelligence; Artificial neural networks; Catalysis; Catalysts; Fission; Machine learning; Model accuracy; Neural networks; Optimization; Physical chemistry
• ISSN: 1463-9076; 1463-9084
• DOI: 10.1039/d1cp03594k

Many current deep neural network (DNN) models only focus on straightforward optimization over the given database. However, most numerical fitting procedures depart from physical laws. By introducing the concept of "catalysis" from physical chemistry, we propose that the physical correlations among molecular properties could spontaneously act as a catalyst in the DNNs, which increases the accuracy, and more importantly, guides the DNNs in the right way. These Catalysis-DNNs (Cat-DNNs) could precisely predict both the ground and excited-state properties, especially the molecules' screening with singlet fission character. We show that traditional machine learning metrics are not suitable for evaluating model accuracy in physical-chemical tasks and issue new physical errors. We believe that the agile transfer of fundamental physics or chemistry domain knowledge, like the catalyst, could significantly benefit both the architecture and application of artificial intelligence technology in the future. Our Cat-DNNs scheme adopts physical correlations as "catalysts" for any multi-task deep neural network model to achieve higher physical accuracy. It can keep all output physically-reliable and precisely predict singlet fission properties.