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

Download:

Current browse context:

math.OC

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 > Optimization and Control

Title: Towards a Dimension-Free Understanding of Adaptive Linear Control

Abstract: We study the problem of adaptive control of the linear quadratic regulator for systems in very high, or even infinite dimension. We demonstrate that while sublinear regret requires finite dimensional inputs, the ambient state dimension of the system need not be bounded in order to perform online control. We provide the first regret bounds for LQR which hold for infinite dimensional systems, replacing dependence on ambient dimension with more natural notions of problem complexity. Our guarantees arise from a novel perturbation bound for certainty equivalence which scales with the prediction error in estimating the system parameters, without requiring consistent parameter recovery in more stringent measures like the operator norm. When specialized to finite dimensional settings, our bounds recover near optimal dimension and time horizon dependence.
Comments: presented at COLT 2021
Subjects: Optimization and Control (math.OC); Machine Learning (cs.LG); Machine Learning (stat.ML)
Cite as: arXiv:2103.10620 [math.OC]
  (or arXiv:2103.10620v2 [math.OC] for this version)

Submission history

From: Juan Carlos Perdomo [view email]
[v1] Fri, 19 Mar 2021 03:59:15 GMT (204kb,D)
[v2] Thu, 15 Jul 2021 17:40:16 GMT (1160kb,D)

Link back to: arXiv, form interface, contact.