Towards an Adaptive and Flexible Metamodeling Toolbox

In the process of designing and understanding complex electrical, mechanical, structural, biological, economic and social systems metamodeling techniques have become indispensable. Due to the increasing computational complexity of current simulation codes multivariate approximation models (polynomials, Artificial Neural Networks, Support Vector Machines, ...) are often used to gain insight into the system behaviour over the the often high dimensional design space. Thus a lot of research has gone into methods for quickly building accurate metamodels as cheaply as possible. This report discusses the motivations behind metamodeling, algorithms for building metamodels and the design and implementation of the M3 metamodeling toolbox.