Name
Abstract | ||
|---|---|---|
The main challenge for adaptive regulation of linear-quadratic systems is the tradeoff between identification and control. An adaptive policy needs to address both the estimation of unknown dynamics parameters (exploration), as well as the regulation of the underlying system (exploitation). To this end, optimism-based methods that bias the identification in favor of optimistic approximations of the true parameter are employed in the literature. A number of asymptotic results have been established, but their finite-time counterparts are few, with important restrictions. This article establishes results for the worst-case regret of optimism-based adaptive policies. The presented high probability upper bounds are optimal up to logarithmic factors. The nonasymptotic analysis of this article requires the following very mild assumptions: stabilizability of the system's dynamics, and limiting the degree of heaviness of the noise distribution. To establish such bounds, certain novel techniques are developed to comprehensively address the probabilistic behavior of dependent random matrices with heavy-tailed distributions. |
Year | DOI | Venue |
|---|---|---|
2021 | 10.1109/TAC.2020.2998952 | IEEE Transactions on Automatic Control |
Keywords | DocType | Volume |
Certainty equivalence (CE),exploration-exploitation,optimism in the face of uncertainty (OFU),reinforcement learning,regret bounds | Journal | 66 |
Issue | ISSN | Citations |
4 | 0018-9286 | 0 |
PageRank | References | Authors |
0.34 | 4 | 3 |
Name | Order | Citations | PageRank |
|---|---|---|---|
| Mohamad Kazem Shirani Faradonbeh | 1 | 24 | 5.96 |
| Ambuj Tewari | 2 | 7 | 3.11 |
| George Michailidis | 3 | 0 | 4.73 |