Institutional Repository [SANDBOX]
Technical University of Crete
EN  |  EL

Search

Browse

My Space

Reinforcement learning as classification: leveraging modern classifiers

Lagoudakis Michael, Parr, R.

Full record


URI: http://purl.tuc.gr/dl/dias/78C8B833-D841-436A-82B4-676C1B860269
Year 2003
Type of Item Conference Full Paper
License
Details
Bibliographic Citation M.G. Lagoudakis and R. Parr. (2003, Aug.). Reinforcement learning as classification: leveraging modern classifiers. [Online]. Available: http://www.aaai.org/Papers/ICML/2003/ICML03-057.pdf
Appears in Collections

Summary

The basic tools of machine learning appear inthe inner loop of most reinforcement learning algorithms,typically in the form of Monte Carlomethods or function approximation techniques.To a large extent, however, current reinforcementlearning algorithms draw upon machine learningtechniques that are at least ten years old and,with a few exceptions, very little has been doneto exploit recent advances in classification learningfor the purposes of reinforcement learning.We use a variant of approximate policy iterationbased on rollouts that allows us to use a pure classificationlearner, such as a support vector machine(SVM), in the inner loop of the algorithm.We argue that the use of SVMs, particularly incombination with the kernel trick, can make iteasier to apply reinforcement learning as an “outof-the-box”technique, without extensive featureengineering. Our approach opens the door tomodern classification methods, but does not precludethe use of classical methods. We presentexperimental results in the pendulum balancingand bicycle riding domains using both SVMs andneural networks for classifiers

Services

Statistics