Model-based Bayesian reinforcement learning for enhancing primary user performance under jamming attack.