https://5dok.net/document/y960mj5l-maximizing-information-gain-partially-observable-environments-prediction-rewards.html