局限与展望:当前未支持原子结构生成(需可逆晶体表示法),未来可结合神经样条流(Neural Spline Flows)提升复杂分布建模能力。 引用:
Victor Fung, Jiaxin Zhang, Guoxiang Hu, P. Ganesh and Bobby G. Sumpter, Inverse design of two-dimensional materials with invertible neural networks, npj Computational Materials (2021)7:200; https://doi.org/10.1038/s41524-021-00670-x