Sample-Efficient Reinforcement Learning of Undercomplete POMDPs

In many sequential decision making settings, the agent lacks complete information about the underlying state of the system, a phenomenon known as partial observability .