Interpretable Fourier Neural Ordinary Differential Equations

Numerous interacting elements can form complex systems. However, describing complex systems is extremely challenging and lacks interpretability. The construction of models still heavily relies on scientists’ understanding of complex systems. This paper explores the use of neural ordinary differential equations in the field of complex system modeling. The proposed Fourier neural ordinary differential equations offer interpretability and a relatively simple structure. By building upon interpretable modeling, the scale is significantly reduced. It is possible to deduce the system’s expression from the model’s parameters. Finally, a real-world system is provided to illustrate the effectiveness of the proposed model, demonstrating its commendable robustness.