Reinforcement Learning of POMDPs using Spectral Methods

Abstract : We propose a new reinforcement learning algorithm for partially observable Markov decision processes (POMDP) based on spectral decomposition methods. While spectral methods have been previously employed for consistent learning of (passive) latent variable models such as hidden Markov models, POMDPs are more challenging since the learner interacts with the environment and possibly changes the future observations in the process. We devise a learning algorithm running through episodes, in each episode we employ spectral techniques to learn the POMDP parameters from a trajectory generated by a fixed policy. At the end of the episode, an optimization oracle returns the optimal memoryless planning policy which maximizes the expected reward based on the estimated POMDP model. We prove an order-optimal regret bound w.r.t. the optimal memoryless policy and efficient scaling with respect to the dimensionality of observation and action spaces.
Document type :
Conference papers
Complete list of metadatas

Cited literature [27 references]  Display  Hide  Download

https://hal.inria.fr/hal-01322207
Contributor : Alessandro Lazaric <>
Submitted on : Thursday, May 26, 2016 - 5:34:15 PM
Last modification on : Friday, March 22, 2019 - 1:35:31 AM
Long-term archiving on : Saturday, August 27, 2016 - 11:03:09 AM

File

master.pdf
Files produced by the author(s)

Identifiers

  • HAL Id : hal-01322207, version 1

Citation

Kamyar Azizzadenesheli, Alessandro Lazaric, Animashree Anandkumar. Reinforcement Learning of POMDPs using Spectral Methods. Proceedings of the 29th Annual Conference on Learning Theory (COLT2016), Jun 2016, New York City, United States. ⟨hal-01322207⟩

Share

Metrics

Record views

254

Files downloads

164