EXTENDED ABSTRACT: The application of artiffcial intelligence (AI) in polymers has made signiffcant progress in recent years. Deep neural networks may help obtain complex features from large-scale, high-dimensional, and sparse datasets to elucidate the relationship between the chemical components, multi-scale structure, and properties of polymer materials. In this talk, I will discuss some cases of the application of AI in the ffeld of polymer materials, addressing the key role of expert knowledge in this process. To realize the design of polyimides with high glass transition temperature, we developed a graph neural network model, which achieves a good prediction efffcacy. A reinforced learning method is further developed to expand the chemical space used for machine learning, and the prediction efficacy is then improved. Another case is for the rubber composite system. To incorporate the condensed matter structure information induced by processing, we developed a transfer learning method, and obtained an improved prediction effect, implying the key role of microstructure information in the prediction of polymer property. I will also discuss the polymer language model and generation model, relative to some of the latest advances, challenges, and future research topics in the ffeld of AI-assisted polymer science.

Keywords: polymer; artiffcial intelligence; small data; language model

REFERENCES:

[1] Qiu Haoke; Liu Lunyang; Qiu Xuepeng; Dai Xuemin; Ji Xiangling; Sun Zhao-Yan; Chem. Sci., 15, (2024) 534;

[2] Zilong Wan, Siyuan Chen, Xueyang Feng, Zhao-Yan Sun, Composites Communications, 51, (2024) 102072;