Jonathan Lee, Alekh Agarwal, Christoph Dann, Tong Zhang: Learning in POMDPs is Sample-Efficient with Hindsight Observability. ICML 2023: 18733-18773