The paper “Tracking and forecasting oscillatory data streams using Koopman autoencoders and Kalman filtering“, co-authored by Stephen Falconer, David Lloyd, and Naratip Santitissadeekorn, has been accepted for publication in Physica D. A final-form version is available on the arXiv (link here). This paper formed part of Stephen Falconer‘s PhD thesis at Surrey. He now works at iProov (headquartered in London). In the paper, they apply the Koopman autoencoder (KAE) to high-dimensional oscillatory data to generate a low-dimensional latent space and model, where the system’s dynamics appear linear. The screenshot below shows Figure 8 from the paper (click on image to enlarge).