Variational Bayesian Reinforcement Learning with Regret Bounds

In reinforcement learning the Q-values summarize the expected future rewards that the agent will attain.