Arificial neural networks (ANNs) lie in a sort of machine-learning middle ground, somewhere between engineering and artificial intelligence. They use mathematical techniques, such as mean-square error minimization, but they also rely on heuristic methods, since very often there is no theoretical background to support decisions about ANN implementation.
Multilayer perceptrons (MLPs) are a widely used ANN class for nonlinear modeling. Their greatest advantage is that a priori knowledge of the specific functional form is not required. Most applications of feedforward MLP have been concerned with the estimation of relationships between input and target variables of interest and the superiority of the performance of this approach in comparison to more classical methods, but they are not only a 'black box' tool. In fact, they have the potential to significantly enhance scientific understanding of empirical phenomena subject to neural network modeling. In particular, the estimates obtained from network learning can serve as a basis for formal statistical inference. Statistical tests of specific scientific hypothesis of interest become possible. Because of the ability of MLP to extract complex nonlinear and interactive effects, the alternatives against which such tests can have power may extend usefully beyond those within reach of more traditional methods, like linear regression analysis.
Was this article helpful?
Do You Want To Learn More About Green Living That Can Save You Money? Discover How To Create A Worm Farm From Scratch! Recycling has caught on with a more people as the years go by. Well, now theres another way to recycle that may seem unconventional at first, but it can save you money down the road.