RL CH2 - Multi-Armed Bandit
In this Chapter:
Multi-Armed Bandit (MAB) problem
Exploitation vs Exploration
ε-greedy algorithm
Upper Confidence Bounds (UCB) algorithm
Thompson Sampling algorithm
Aim of this chapter:
Understand concepts of the Multi-Armed Bandit problem as a basic problem for Reinforcement Learning (RL) and discuss multiple approaches to deal with the problem.
*Note: correction on the slide 15*
In slide 15, minute: 37 the calculations for UBC and (2log 1000) is 13.8155, therefore 0.26, 0.25, 0.13 for bandits 1 to 3 respectively