We gratefully acknowledge support from
the Simons Foundation and member institutions.
Full-text links:

Download:

Current browse context:

math.DS

Change to browse by:

References & Citations

Bookmark

(what is this?)
CiteULike logo BibSonomy logo Mendeley logo del.icio.us logo Digg logo Reddit logo ScienceWISE logo

Mathematics > Dynamical Systems

Title: Deep learning for universal linear embeddings of nonlinear dynamics

Abstract: Identifying coordinate transformations that make strongly nonlinear dynamics approximately linear is a central challenge in modern dynamical systems. These transformations have the potential to enable prediction, estimation, and control of nonlinear systems using standard linear theory. The Koopman operator has emerged as a leading data-driven embedding, as eigenfunctions of this operator provide intrinsic coordinates that globally linearize the dynamics. However, identifying and representing these eigenfunctions has proven to be mathematically and computationally challenging. This work leverages the power of deep learning to discover representations of Koopman eigenfunctions from trajectory data of dynamical systems. Our network is parsimonious and interpretable by construction, embedding the dynamics on a low-dimensional manifold that is of the intrinsic rank of the dynamics and parameterized by the Koopman eigenfunctions. In particular, we identify nonlinear coordinates on which the dynamics are globally linear using a modified auto-encoder. We also generalize Koopman representations to include a ubiquitous class of systems that exhibit continuous spectra, ranging from the simple pendulum to nonlinear optics and broadband turbulence. Our framework parametrizes the continuous frequency using an auxiliary network, enabling a compact and efficient embedding at the intrinsic rank, while connecting our models to half a century of asymptotics. In this way, we benefit from the power and generality of deep learning, while retaining the physical interpretability of Koopman embeddings.
Comments: v2: added another example and further details (increase from 9 pages to 14 pages and increase from 4 figures to 16 figures)
Subjects: Dynamical Systems (math.DS); Machine Learning (cs.LG); Machine Learning (stat.ML)
DOI: 10.1038/s41467-018-07210-0
Cite as: arXiv:1712.09707 [math.DS]
  (or arXiv:1712.09707v2 [math.DS] for this version)

Submission history

From: Bethany Lusch [view email]
[v1] Wed, 27 Dec 2017 23:10:35 GMT (1941kb,D)
[v2] Fri, 13 Apr 2018 06:17:30 GMT (6368kb,D)

Link back to: arXiv, form interface, contact.