An Empirical Investigation into Function Approximation with Reinforcement Learning

@inproceedings{Gbel2005AnEI,
title={An Empirical Investigation into Function Approximation with Reinforcement Learning},
author={Max C. G{\"o}bel},
year={2005}
}

In the reinforcement learning framework, standard, table-based look-up methods for value functions converge to the optimal solution, yet unfortunately these methods are intractable for complex real-world control problems. A common approach to overcome this problem are so-called function approximation techniques that generalise over their input spaces. In this paper we study the capabilities of two machine learning frameworks, the neural network and the self-organising map, in the light of their… CONTINUE READING