Learning Neural Koopman Operators with Dissipativity Guarantees

Yuezhu Xu, S. Sivaranjani, Vijay Gupta · 2025

Read Learning Neural Koopman Operators with Dissipativity Guarantees by Yuezhu Xu, S. Sivaranjani, Vijay Gupta (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2509-07294.txt.

Description

Yuezhu Xu and 2 co-authors learn a neural Koopman operator model with a safety property built in, in a 2025 paper. A Koopman operator rewrites a nonlinear system, one whose response is not proportional to its input, as a linear model in a lifted space. Dissipativity means the system steadily loses energy, a trait many real systems have and a learned model should respect.

The method works in two stages: first fit an unconstrained neural Koopman model closely to data, then nudge its parameters just enough to enforce strict dissipativity. The authors prove the learned model's dissipativity carries back to the true system, accounting for noisy data, truncation, and unseen cases. A Duffing oscillator, a classic vibrating system with a nonlinear spring, serves as the test case. Guaranteed energy behavior makes learned models safer to use inside controllers.

Catalog record

CREATORYuezhu Xu, S. Sivaranjani, Vijay Gupta
YEAR2025
COLLECTIONPAPERS
FORMATTXT
FILE SIZE37 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2509-07294

Browse the full library · About the Tower of Babel