Abstract
Reinforcement learning (RL) for a linear family of tasks is studied in this paper. The key of our discussion is nonlinearity of the optimal solution even if the task family is linear; we cannot obtain the optimal policy by a naive approach. Though there exists an algorithm for calculating the equivalent result to Q-learning for each task all together, it has a problem with explosion of set sizes. We introduce adaptive margins to overcome this difficulty.
Access this chapter
Tax calculation will be finalised at checkout
Purchases are for personal use only
Preview
Unable to display preview. Download preview PDF.
References
Jaakkola, T., et al.: Neural Computation 6, 1185–1201 (1994)
Sutton, R.S., Barto, A.G.: Reinforcement Learning. The MIT Press, Cambridge (1998)
Kaneko, Y., et al.: In: Proc. IEICE Society Conference (in Japanese), vol. 167 (2004)
Kaneko, N., et al.: In: Proc. IEICE Society Conference (in Japanese), vol. A-2-10 (2005)
Natarajan, S., et al.: In: Proc. Intl. Conf. on Machine Learning, pp. 601–608 (2005)
Hiraoka, K., et al.: The Brain & Neural Networks (in Japanese). Japanese Neural Network Society 13, 137–145 (2006)
Yoshida, M., et al.: Proc. FIT (in Japanese) (to appear, 2007)
Preparata, F.P., et al.: Computational Geometry. Springer, Heidelberg (1985)
Alexandrov, V.N., Dongarra, J., Juliano, B.A., Renner, R.S., Tan, C.J.K. (eds.): ICCS 2001. LNCS, vol. 2073. Springer, Heidelberg (2001)
Fukuda, K.: J. Symbolic Computation 38, 1261–1272 (2004)
Fogel, E., et al.: In: Proc. ALENEX, pp. 3–15 (2006)
Author information
Authors and Affiliations
Editor information
Rights and permissions
Copyright information
© 2008 Springer-Verlag Berlin Heidelberg
About this paper
Cite this paper
Hiraoka, K., Yoshida, M., Mishima, T. (2008). Parallel Reinforcement Learning for Weighted Multi-criteria Model with Adaptive Margin. In: Ishikawa, M., Doya, K., Miyamoto, H., Yamakawa, T. (eds) Neural Information Processing. ICONIP 2007. Lecture Notes in Computer Science, vol 4984. Springer, Berlin, Heidelberg. https://doi.org/10.1007/978-3-540-69158-7_51
Download citation
DOI: https://doi.org/10.1007/978-3-540-69158-7_51
Publisher Name: Springer, Berlin, Heidelberg
Print ISBN: 978-3-540-69154-9
Online ISBN: 978-3-540-69158-7
eBook Packages: Computer ScienceComputer Science (R0)