Not only could innovating around scalable learning help to rebuild trust in our institutions, it could also lead to a profound shift in the nature of performance improvement. The scalable efficiency institutional model is inherently a diminishing returns model – the more efficient these institutions become, the longer and harder they will need to work to get the next increment of performance improvement. Scalable learning, on the other hand, for the first time offers the potential to shift to an increasing returns model where the more people who join together to learn faster, the more rapidly value gets created.
Previous rigorous approaches for this problem rely on dynamic programming (DP) and, while sample efficient, have running time quadratic in the sample size. As our main contribution, we provide new sample near-linear time algorithms for the problem that – while not being minimax optimal – achieve a significantly better sample-time tradeoff on large datasets compared to the DP approach. Our experimental evaluation shows that, compared with the DP approach, our algorithms provide a convergence rate that is only off by a factor of 2 to 4, while achieving speedups of three orders of magnitude.