• Corpus ID: 231846636

A Convex Optimization Approach to Learning Koopman Operators

  title={A Convex Optimization Approach to Learning Koopman Operators},
  author={Mario Sznaier},
  • M. Sznaier
  • Published 7 February 2021
  • Computer Science
  • ArXiv
Koopman operators provide tractable means of learning linear approximations of non-linear dynamics. Many approaches have been proposed to find these operators, typically based upon approximations using an a-priori fixed class of models. However, choosing appropriate models and bounding the approximation error is far from trivial. Motivated by these difficulties, in this paper we propose an optimization based approach to learning Koopman operators from data. Our results show that the Koopman… 

Figures from this paper

Linear Matrix Inequality Approaches to Koopman Operator Approximation
Koopman operator theory [1–4] provides a means to globally represent a nonlinear system as a linear system by transforming its states into an infinite-dimensional space of lifted states. The Koopman
System Norm Regularization Methods for Koopman Operator Approximation
DMD and DMD with control are reformulated as convex optimization problems with linear matrix inequality constraints and hard asymptotic stability constraints and system norm regularizers are considered as methods to improve the numerical conditioning of the approximate Koopman operator.
Control Oriented Learning in the Era of Big Data
  • M. Sznaier
  • Computer Science
    IEEE Control Systems Letters
  • 2021
The main message is twofold (i) computational complexity in control oriented learning is driven both by system order and the presence of uncertainty, rather than the dimension of the data, and (ii) exploiting the underlying sparsity provides a way around the “curse of dimensionality”.


Linearly-Recurrent Autoencoder Networks for Learning Dynamics
A new neural network architecture combining an autoencoder with linear recurrent dynamics in the encoded state is used to learn a low-dimensional and highly informative Koopman-invariant subspace of observables.
Iterative reweighted algorithms for matrix rank minimization
This paper proposes a family of Iterative Reweighted Least Squares algorithms IRLS-p, and gives theoretical guarantees similar to those for nuclear norm minimization, that is, recovery of low-rank matrices under certain assumptions on the operator defining the constraints.
Deep learning for universal linear embeddings of nonlinear dynamics
It is often advantageous to transform a strongly nonlinear system into a linear one in order to simplify its analysis for prediction and control, so the authors combine dynamical systems with deep learning to identify these hard-to-find transformations.
Hankel based maximum margin classifiers: A connection between machine learning and Wiener systems identification
This paper considers the problem of nonparametric identification of Wiener systems in cases where there is no a-priori available information on the dimension of the output of the linear dynamics and shows that tractable convex relaxations can be obtained by exploiting a combination of recent developments in polynomial optimization and matrix rank minimization.
Chordal decomposition in operator-splitting methods for sparse semidefinite programs
Chordal decomposition is employed to reformulate a large and sparse semidefinite program (SDP), either in primal or dual standard form, into an equivalent SDP with smaller positive semideFinite (PSD) constraints, enabling the development of efficient and scalable algorithms.
Analysis of Fluid Flows via Spectral Properties of the Koopman Operator
This article reviews theory and applications of Koopman modes in fluid mechanics. Koopman mode decomposition is based on the surprising fact, discovered in Mezic (2005), that normal modes of linear
Low order dynamics embedding for high dimensional time series
This paper considers the problem of finding low order nonlinear embeddings of dynamic data, that is, data characterized by a temporal ordering, and proposes a semi-definite programming based manifold embedding algorithm that exploits the dynamical information encapsulated in the temporal ordering of the sequence to obtainembeddings with lower complexity that those produced by existing algorithms, at a comparable computational cost.
Nonlinear dimensionality reduction by locally linear embedding.
Locally linear embedding (LLE) is introduced, an unsupervised learning algorithm that computes low-dimensional, neighborhood-preserving embeddings of high-dimensional inputs that learns the global structure of nonlinear manifolds.
Chaos as an intermittently forced linear system
A universal, data-driven decomposition of chaos as an intermittently forced linear system is presented, combining delay embedding and Koopman theory to decompose chaotic dynamics into a linear model in the leading delay coordinates with forcing by low-energy delay coordinates.