Batch reinforcement learning with state importance
- Lihong Li
- Vadim Bulitko
- Russ Greiner, Dept of Computing Science; PI of AICML
We investigate the problem of using function approximation in reinforcementlearning where the agent\'s policy is represented as a classifier mappingstates to actions. High classification accuracy is usually deemed to correlate withhigh policy quality. But this is not necessarily the case as increasing classificationaccuracy can actually decrease the policy\'s quality. This phenomenon takes placewhen the learning process begins to focus on classifying less \
Citation
L. Li, V. Bulitko, R. Greiner. "Batch reinforcement learning with state importance". European Conference on Machine Learning (ECML), Pisa, Italy, pp 566-588, September 2004.Keywords: | machine learning, reinforcement learning, importance |
Category: | In Conference |
BibTeX
@incollection{Li+al:ECML04, author = {Lihong Li and Vadim Bulitko and Russ Greiner}, title = {Batch reinforcement learning with state importance}, Pages = {566-588}, booktitle = {European Conference on Machine Learning (ECML)}, year = 2004, }Last Updated: May 29, 2007
Submitted by Staurt H. Johnson