3.24 - Non-linear Modeling: Neural Networks

Artificial neural networks are mathematical models originally inspired by the idea of reproducing the functioning of human brain. In particular, from their biological counterpart, they have inherited the feature that data processing is distributed through a large quantity of networked processing units. This allows an high versatility and the possibility of implementing any arbitrary functional relationship. In this chapter, the fundamentals of artificial neural networks and their use in nonlinear regression are covered, focusing on the two architectures most used in chemometrics: multilayer feed-forward and supervised Kohonen networks.