Abstract
We propose to adopt statistical regression as the projection operator toenable data-driven learning of the operators in the Mori--Zwanzig formalism. Wepresent a principled method to extract the Markov and memory operators for anyregression models. We show that the choice of linear regression results in arecently proposed data-driven learning algorithm based on Mori's projectionoperator, which is a higher-order approximate Koopman learning method. We showthat more expressive nonlinear regression models naturally fill in the gapbetween the highly idealized and computationally efficient Mori's projectionoperator and the most optimal yet computationally infeasible Zwanzig'sprojection operator. We performed numerical experiments and extracted theoperators for an array of regression-based projections, including linear,polynomial, spline, and neural-network-based regressions, showing a progressiveimprovement as the complexity of the regression model increased. Ourproposition provides a general framework to extract memory-dependentcorrections and can be readily applied to an array of data-driven learningmethods for stationary dynamical systems in the literature.