Prof. Henry Lam (Columbia University)
Assessing solution quality in stochastic optimization under limited data
We study the construction of statistical confidence intervals for the optimality gap, as a measurement of solution quality, in stochastic optimization under limited data. We demonstrate how viewing an optimistic bound for such problems as a classical symmetric statistic allows inference with less order of data size than previously proposed. We also demonstrate how our machinery is related to the input uncertainty problem in Monte Carlo simulation where the output analysis attempts to capture both the Monte Carlo noise and the input data noise, which gives rise to some bootstrapping techniques in computing the estimates.
|Date||November 27, 2017 (Mon) 14:00 - 15:00|