Neural Network

   The Characteristics of Neural Networks That Make Them So Useful

   1. A neural network is composed of a number of very simple processing elements [("neurodes")] that communicate through a rich set of interconnections with variable weights or strengths.

   2. Memories are stored or represented in a neural network in the pattern of variable interconnection weights among the neurodes. Information is processed by a spreading, constantly changing pattern of activity distributed across many neurodes.

   3. A neural network is taught or trained rather than programmed. It is even possible to construct systems capable of independent or autonomous learning. . . .

   4. Instead of having a separate memory and controller, plus a stored external program that dictates the operation of the system as in a digital computer, the operation of a neural network is implicitly controlled by three properties: the transfer function of the neurodes, the details of the structure of the connections among the neurodes, and the learning law the system follows.

   5. A neural network naturally acts as an associative memory. That is, it inherently associated items it is taught, physically grouping similar items together in its structure. A neural network operated as a memory is content addressable; it can retrieve stored information from incomplete, noisy, or partially incorrect input cues.

   6. A neural network is able to generalize; it can learn the characteristics of a general category of objects based on a series of specific examples from that category.

   7. A neural network keeps working even after a significant fraction of its neurodes and interconnections have become defective.

   8. A neural network innately acts as a processor for time-dependent spatial patterns, or spatiotemporal patterns. (Caudill & Butler, 1990, pp. 7-8)