The subdifferential of measurable composite max integrands and smoothing approximation (Q2189440)

The authors develop a framework for the study of smoothing methods for the expectation of random integrands \(F(x) = E[ f (\xi, x)]\) based on the smoothing of the integrand \(f\), and then a smoothing approach to the approximation of the Clarke subgradients \(\partial F(x)\) of the expectation \(F(x) = E[ f (\xi, x)]\), and the application of these techniques to the class of measurable composite max\,(CM) integrands. CM integrands arise in several important applications including stochastic nonlinear complementarity problems and optimization problems in statistical learning.