Online learning and optimization of Markov jump linear models

Sevi Baltaoglu; Lang Tong; Qing Zhao

ICASSP 2016accepted0 citations

Online learning and optimization of Markov jump linear models

Abstract

The problem of online learning and optimization of unknown Markov jump linear models is considered. A new online learning algorithm, referred to as Markovian simultaneous perturbations stochastic approximation (MSPSA), is proposed. It is shown that ν/ MSPSA achieves the minimax regret order of Θ(√T). Using the Van Trees inequality (stochastic Cramér-Rao bound), it is shown ν/ that Θ(√T) is the lowest regret order achievable. Simulation results show scenarios that MSPSA offers significant gain over the greedy certainty equivalent approaches.

BibTeX

@inproceedings{icassp2016_onlinelearningan,
  title = {Online learning and optimization of Markov jump linear models},
  author = {Sevi Baltaoglu and Lang Tong and Qing Zhao},
  booktitle = {ICASSP 2016},
  year = {2016}
}