Dynamical Systems Learning and Optimization

The field of dynamical systems is moving towards more accurate and efficient learning methods, with a focus on addressing numerical artifacts and optimizing function approximation. Researchers are exploring new approaches to learn dynamical systems from data, including data-driven optimal approximation and sparse identification of nonlinear dynamics. Notable advancements include the development of structure-preserving deflation strategies for critical eigenvalues and the proposal of library optimization mechanisms for improved model interpretability and reliability. Noteworthy papers include:

  • Numerical Artifacts in Learning Dynamical Systems, which highlights the potential effects of numerical schemes on learning outcomes.
  • Sparse identification of nonlinear dynamics with library optimization mechanism, which proposes a novel approach to optimize the design of basis functions for improved model accuracy.

Sources

Numerical Artifacts in Learning Dynamical Systems

Data-driven optimal approximation on Hardy spaces in simply connected domains

Structure-preserving deflation of critical eigenvalues in quadratic eigenvalue problems associated with damped mass-spring systems

Sparse identification of nonlinear dynamics with library optimization mechanism: Recursive long-term prediction perspective

Built with on top of