Neural networks, surrogate models and black box algorithms: theory and applications

In this Ph. D. Thesis we will analyze some of the most used surrogate models, together with a particular type of line search black box strategy. After introducing these powerful tools, we will present the Canonical Duality Theory, the potentiality it has to improve these tools, and some of their applications. The principal contributes of this Thesis are the reformulation of the Radial Basis Neural Network problem in its canonical dual form in Section 2.2 and the application of the surrogate models and black box algorithms presented in this Thesis on various real world problems reported in Chapter 3.