Adaptive Sampling Trust Region Optimization for Derivative-free Stochastic Functions and Deterministic Constraints
AustMS 2026, University of Western Australia
01 December 2026
Motivated by problems from simulation-based optimization, we consider the problem of minimizing a stochastic, black-box objective subject to deterministic nonlinear constraints. In this setting, we only have access to stochastic estimates (and no derivatives) of the objective, but we assume access to exact function and gradient evaluations for the constraints. We extend an adaptive sampling variant of a model-based derivative-free optimization algorithm to this setting, and establish almost sure convergence of an SQP-based algorithm, and present numerical results on a stochastic activity network problem. This is joint work with Nicole Felice and Sara Shashaani (North Carolina State University).
