Zero-Order, Black-Box, Derivative-Free, and Simulation-Based Optimization
Friday, August 5, 2016 - 11:00am - 12:30pm
This lecture will focus on the situation when gradients of the objective function are not available to an optimization algorithm. We summarize algorithms for local optimization of a deterministic function, with particular attention directed to model-based trust-region methods. We provide foundations for the theory underlying these algorithms and highlight performance in practice. We conclude with an introduction to ways of building upon these algorithms to improve the performance on problems that involve composite functions of black boxes.