Adaptive Interpolation Strategies in Derivative-Free Optimization: a case study

Abstract: Derivative-Free optimization (DFO) focuses on designing methods to solve optimization problems without the analytical knowledge of gradients of the objective function. There are two main families of DFO methods: model-based methods and direct search methods. In model-based DFO methods, a model of the objective function is constructed using only objective function values, and the model is used to guide the computation of the next iterate. Natural questions in this class of algorithms include how many function evaluations should be used to construct the model? And, should this number be fixed, or adaptively selected by the algorithm? In this paper, we numerically examine these questions, using Hare and Lucet's Derivative-Free Proximal Point (DFPP) algorithm [Hare, Lucet, 2014] as a case study. Results suggest that the number of function evaluations used to construct the model has a huge impact on algorithm performance, and adaptive strategies can both improve and hinder algorithm performance.