Answers are grounded in this paper and cite their sources.
Discounted UCB1-tuned was proposed as one of the methods to choose the action in a multi-armed bandit problem. This algorithm is an optimized selection method for balancing between the exploration and the exploitation, by using weighted value and weighted variance. In this paper, we proposed the method to apply Discounted UCB1-tuned to Q-learning, and experimentally evaluated its performance in the continuous state spaces shortest path problem.
No takes yet. Share an insight, caveat, or question.
Saito et al. (2014) studied this question.