To refer to this page use:
http://arks.princeton.edu/ark:/88435/pr1g85n
Full metadata record
DC Field | Value | Language |
---|---|---|
dc.contributor.author | Plevrakis, Orestis | - |
dc.contributor.author | Hazan, Elad | - |
dc.date.accessioned | 2021-10-08T19:50:56Z | - |
dc.date.available | 2021-10-08T19:50:56Z | - |
dc.date.issued | 2020 | en_US |
dc.identifier.citation | Plevrakis, Orestis, and Elad Hazan. "Geometric Exploration for Online Control." Advances in Neural Information Processing Systems 33 (2020). | en_US |
dc.identifier.issn | 1049-5258 | - |
dc.identifier.uri | https://papers.nips.cc/paper/2020/file/565e8a413d0562de9ee4378402d2b481-Paper.pdf | - |
dc.identifier.uri | http://arks.princeton.edu/ark:/88435/pr1g85n | - |
dc.description.abstract | We study the control of an \emph{unknown} linear dynamical system under general convex costs. The objective is minimizing regret vs the class of strongly-stable linear policies. In this work, we first consider the case of known cost functions, for which we design the first polynomial-time algorithm with n 3 √ T -regret, where n is the dimension of the state plus the dimension of control input. The √ T -horizon dependence is optimal, and improves upon the previous best known bound of T 2 / 3 . The main component of our algorithm is a novel geometric exploration strategy: we adaptively construct a sequence of barycentric spanners in an over-parameterized policy space. Second, we consider the case of bandit feedback, for which we give the first polynomial-time algorithm with p o l y ( n ) √ T -regret, building on Stochastic Bandit Convex Optimization. | en_US |
dc.language.iso | en_US | en_US |
dc.relation.ispartof | Advances in Neural Information Processing Systems | en_US |
dc.rights | Final published version. Article is made available in OAR by the publisher's permission or policy. | en_US |
dc.title | Geometric Exploration for Online Control | en_US |
dc.type | Conference Article | en_US |
pu.type.symplectic | http://www.symplectic.co.uk/publications/atom-terms/1.0/conference-proceeding | en_US |
Files in This Item:
File | Description | Size | Format | |
---|---|---|---|---|
GeometricExploreOnline.pdf | 319.96 kB | Adobe PDF | View/Download |
Items in OAR@Princeton are protected by copyright, with all rights reserved, unless otherwise indicated.