Given a particular objective function to be optimized, it would be useful to know which optimization method will perform best on that objective. Indeed, the entire purpose of studying the static search generators is to provide tools to answer this very question. To this end, different categories of performance criteria are analyzed theoretically in this chapter. Several performance criteria are shown to be continuous and non-linear with respect to static search generators, implying that similar optimization methods perform similarly and that linearly interpolated generators produce optimization methods may outperform the methods being interpolated. These facts are demonstrated experimentally in Chapter 11. Further, the categories of performance criteria described in this chapter make it possible to identify the conditions under which No Free Lunch theorems hold in infinite-dimensional spaces, to be undertaken in Chapter 12.
CITATION STYLE
Lockett, A. J. (2020). Performance analysis. In Natural Computing Series (pp. 239–262). Springer. https://doi.org/10.1007/978-3-662-62007-6_10
Mendeley helps you to discover research relevant for your work.