Detection Methods: part 3

Method 3: Neural Network

The training of nets is a difficult matter; It isn't just one of your everyday games. You may think at first I'm as mad as a hatter when I tell you each node must have three different names.
-- with apologies to T. S. Elliot

Artificial neural networks suggest that appropriate discrimination planes can be built to distinguish between different classes. Previously unseen members of a class (that has been previously observed) can be identified through extrapolation. However, artificial neural networks also involve some hit-or-miss work. The selection of an architecture and training set, for example, relies more on intuition and iteration than on a theoretical background. Also, the training process for neural networks is subject to becoming trapped in local minima of the error function; the goal of training the network is to find the global minimum, which may often be difficult. Finally, the training of artificial neural networks is time-consuming.

Previous Page Return to Main Page

jchen@micro.ti.com
tdorney@ti.com
sparr@owlnet.rice.edu