English

Interpretable Gradient Descent for Kalman Gain

Optimization and Control 2025-07-23 v2 Systems and Control Systems and Control

Abstract

We derive a decomposition for the gradient of the innovation loss with respect to the filter gain in a linear time-invariant system, decomposing as a product of an observability Gramian and a term quantifying the ``non-orthogonality" between the estimation error and the innovation. We leverage this decomposition to give a convergence proof of gradient descent to the optimal Kalman gain, specifically identifying how recovery of the Kalman gain depends on a non-standard observability condition, and obtaining an interpretable geometric convergence rate.

Keywords

Cite

@article{arxiv.2507.14354,
  title  = {Interpretable Gradient Descent for Kalman Gain},
  author = {M. A. Belabbas and A. Olshevsky},
  journal= {arXiv preprint arXiv:2507.14354},
  year   = {2025}
}
R2 v1 2026-07-01T04:08:45.046Z