Gradient and Hessian Approximations for Model-based Blackbox Optimization

Speaker: Warren Hare, Ph.D. Professor, Mathematics, UBC

Room: B501

Date: 22/11/2022 14:00

Abstract:

A blackbox is a function that provides output without 
explanation of how the output was constructed.  One common strategy for 
optimization involving blackbox functions is to numerically approximate 
gradients and/or Hessians and use these approximations in a classical 
method.  In this talk, we examine the mathematical theory behind such an 
approach.   We discuss classical and novel approximation techniques for 
blackbox functions. And, we illustrate the results on a case study from 
an application in Medical Physics.