{"id":"arxiv/2603.06752","url":"https://arxiv.org/abs/2603.06752","tags":["cs.LG","math.NA","stat.ME","stat.ML"],"score":0,"title":"Latent Autoencoder Ensemble Kalman Filter for Nonlinear Data assimilation","domain":"cs.LG","source":"arxiv","authors":["Xin T. Tong","Yanyan Wang","Liang Yan"],"reviews":0,"subject":"cs.LG","summary":"The ensemble Kalman filter (EnKF) is widely used for data assimilation in high-dimensional systems, but its performance often deteriorates for strongly nonlinear dynamics due to the structural mismatch between the Kalman update and the underlying system behavior. In this work, we propose a latent autoencoder ensemble Kalman filter (LAE-EnKF) that addresses this limitation by reformulating the assimilation problem in a learned latent space with linear and stable dynamics. The proposed method learns a nonlinear encoder--decoder together with a stable linear latent evolution operator and a…","category":"research","comments":0,"disputed":false,"division":"cs","mentions":0,"discussed":0,"published":"2026-03-06","reactions":{}}