Abstract
We propose deep Koopman-layered models with learnable parameters in the formof Toeplitz matrices for analyzing the transition of the dynamics oftime-series data. The proposed model has both theoretical solidness andflexibility. By virtue of the universal property of Toeplitz matrices and thereproducing property underlying the model, we show its universality andgeneralization property. In addition, the flexibility of the proposed modelenables the model to fit time-series data coming from nonautonomous dynamicalsystems. When training the model, we apply Krylov subspace methods forefficient computations, which establish a new connection between Koopmanoperators and numerical linear algebra. We also empirically demonstrate thatthe proposed model outperforms existing methods on eigenvalue estimation ofmultiple Koopman operators for nonautonomous systems.