Performance comparison of non-RNN and RNN in Emergence of Discrete Decision Making through Reinforcement Learning
Samsudin Mohamad Faizal, Katsunari Shibata · Frontiers in artificial intelligence and applications · 2012
Using a neural network in the task that requires discrete decision making suffers from the problem of discrete decision making. On the other hand, using a lookup table suffers from the problem in generalization and the curse of dimensionality. In this paper, simple localized inputs in neural network are used in order to overcome this problem. Furthermore, by utilizing the internal dynamics in RNN, it is expected that quick discrete decision making can be obtained through learning.