Abstract
This paper proposes a data-driven framework to learn a finite-dimensional Koopman approximation under noisy observations. The method only assumes bounded measurement noise and modifies deep Koopman formulations to explicitly characterize and mitigate noise effects by updating observable-function parameters. Experiments on benchmark systems show improved robustness compared with related Koopman-learning methods.
Demo
Demo coming soon.
Citation
Hao, Wenjian, Devesh Upadhyay, and Shaoshuai Mou. 2025. "Deep Koopman Learning using Noisy Data." Transactions on Machine Learning Research.
@article{
hao2025deep,
title={Deep Koopman Learning using Noisy Data},
author={Wenjian Hao and Devesh Upadhyay and Shaoshuai Mou},
journal={Transactions on Machine Learning Research},
issn={2835-8856},
year={2025},
url={https://openreview.net/forum?id=j6Rm6T2lFU},
note={}
}