We present a new algorithm for general reinforcement learning where the true environment is known to belong to a finite class of N arbitrary models. The algorithm is shown to be near-optimal for all but O(N log^2 N) time-steps with high probability. Infinite classes are also considered where we show that compactness is a key criterion for determining the existence of uniform sample-complexity bounds. A matching lower bound is given for the finite case.
@article{arxiv.1308.4828,
title = {The Sample-Complexity of General Reinforcement Learning},
author = {Tor Lattimore and Marcus Hutter and Peter Sunehag},
journal= {arXiv preprint arXiv:1308.4828},
year = {2013}
}