“…Research related to reduced Lagrangian representations that enable time-varying vector fields has advanced along multiple axes. These include in situ sampling techniques (Agranovsky et al, 2014;Rapp et al, 2019;Sane et al, 2019Sane et al, , 2021b, post hoc reconstruction strategies (Agranovsky et al, 2015;Bujack and Joy, 2015;Chandler et al, 2014;Hlawatsch et al, 2010), theoretical and empirical error analysis (Chandler et al, 2016;Hummel et al, 2016;Sane et al, 2018), feature extraction (Froyland and Junge, 2018;Froyland and Padberg-Gehle, 2015;Hadjighasem et al, 2017;Jakob et al, 2020;Schlueter-Kuck and Dabiri, 2017), and application to various domains (Nardini et al, 2017;Sane et al, 2021a;Siegfried et al, 2019). In this paper, we study the use of deep learning to perform post hoc reconstruction.…”