Optimal learning for sequential sampling with non-parametric beliefs (Q742143)

The authors consider the problem of maximizing an unknown function over a finite set of possible alternatives. They propose a sequential learning policy for ranking and selection problems, using a non-parametric procedure for estimating the value of a policy. Their estimation approach aggregates over a set of kernel functions in order to achieve a more consistent estimator. The final estimate uses a weighting scheme with the inverse \ mean square errors of the kernel estimators as weights. This weighting scheme is shown to be optimal under the independent kernel estimators. For choosing the measurement, the authors employ the knowledge gradient policy that relies on predictive distributions to calculate the optimal sampling point. The method allows a setting where the beliefs are expected to be correlated but the correlation structure is unknown beforehand. Moreover, the proposed policy is shown to be asymptotically optimal.