%0 Report %1 beleznay1999 %A Beleznay, F. %A Gr\Hobler, T. %A Szepesvári, Cs. %C Budapest 1121, Konkoly Th. M. u. 29-33, Hungary %D 1999 %K convergence, learning of rate reinforcement theory, %N TR-99-02 %T Comparing Value-Function Estimation Algorithms in Undiscounted Problems %X We compare scaling properties of several value-function estimation algorithms. In particular, we prove that Q-learning can scale exponentially slowly with the number of states. We identify the reasons of the slow convergence and show that both TD($łambda$) and learning with a fixed learning-rate enjoy rather fast convergence, just like the model-based method.

@techreport{beleznay1999, abstract = {We compare scaling properties of several value-function estimation algorithms. In particular, we prove that Q-learning can scale exponentially slowly with the number of states. We identify the reasons of the slow convergence and show that both TD($\lambda$) and learning with a fixed learning-rate enjoy rather fast convergence, just like the model-based method.}, added-at = {2020-03-17T03:03:01.000+0100}, address = {Budapest 1121, Konkoly Th. M. u. 29-33, Hungary}, author = {Beleznay, F. and Gr{\Ho}bler, T. and Szepesv{\'a}ri, {Cs}.}, biburl = {https://www.bibsonomy.org/bibtex/2f46b67f4af6ffb5d3a1c61b83a90beea/csaba}, date-added = {2010-08-28 17:38:14 -0600}, date-modified = {2010-09-02 13:09:15 -0600}, institution = {Mindmaker Ltd.}, interhash = {63cbb1a604baf335259adb81f90bc9da}, intrahash = {f46b67f4af6ffb5d3a1c61b83a90beea}, keywords = {convergence, learning of rate reinforcement theory,}, number = {TR-99-02}, pdf = {papers/slowql-tr99-02.ps.pdf}, timestamp = {2020-03-17T03:03:01.000+0100}, title = {Comparing Value-Function Estimation Algorithms in Undiscounted Problems}, year = 1999 }