Overview

Koopman operator learning represents a nonlinear system through observables whose evolution is linear, making forecasting and control easier to analyze and compute. Methods range from dynamic mode decomposition and extended DMD (EDMD) with fixed or learned dictionaries to kernel and reservoir methods, neural Koopman autoencoders, and structured continuous-time generators. Researchers test these models on chaotic and fluid systems, time-series benchmarks, and robots, often pairing them with model predictive control or reinforcement learning. A central challenge is balancing expressive learned representations with reliable spectra, stable long-horizon predictions, and guarantees under noise, limited data, or changing dynamics.

Latest papers

CardsList

92 papers11 papers in the last four weeks, down 8% on the four weeks before. 0.1% of all new papers.

  1. Adversarial dynamical systems characterize when data-driven learning succeeds or fails

    Jul 8, 2024Matthew J. Colbrook, Igor Mezić, Alexei StepanenkoKoopman Operator LearningSpectral Methods

  2. Nonparametric Control Koopman Operators

    May 12, 2024Petar Bevanda, Bas Driessen, Lucian Cristian Iacob +3Koopman Operator LearningSystem Identification