| Intelligent Systems And Their Societies | Walter Fritz |
Artificial neural fields are used today in a number of applications and work well. However, while checking them out, we found that an IS of response rules, on the same computer, finds the answer faster. The reason is that each concept has attached a list of the numbers of response rules, in which that concept is used. The IS program finds all possible response rules for evaluation very fast since they are those mentioned in the concepts of the present situation.
Also, neural fields need many more examples before learning the correct answer. (It may be that they learn slower but better.) It looks to me like neural fields should be the way to build ISs if each neuron is represented by a different physical processing unit (for instance a transducer) and thus the neurons would be able to work in parallel. Then the speed of response to an input to a neural field should be excellent.
An expert system also uses response rules, but such a system does not learn these response rules by itself; a knowledge engineer determines them in meetings with the expert and then manually writes each individual response rule into the program. This turns out to be a great disadvantage because this manual work takes a lot of time for both the expert and the knowledge engineer. For instance, all of the knowledge that a person has accumulated by the time they are in high school could not be determined and written into an expert system program because of the astronomical amount of time such a project would require. An intelligent learning program, however, could learn and organize this volume of knowledge in a reasonable amount of time.
For continuos reading, like a book - continue
here.
Jump to the e-book Contents /
Artificial IS document /
top of this document.
Last Edited 6 Mar. 06 / Walter Fritz