ABSTRACT
One of the oft-quoted advantages of neural systems is that they can be used as a black box, able to learn a task without the user having a detailed understanding of the internal processes. While this is undoubtedly true, it is also the case that many errors and cases of poor performance are created by users who use inappropriate networks, architectures or learning paradigms for their problems, and that having a grasp of what the network is trying to do and how it is going about it will inevitably result in the more appropriate and effective use of neural systems.
