Rule Induction
- One long standing hypothesis is that people are symbol
processing machines (Newell)
- This, and some other things, led to the advancement of
symbolic AI which was pretty much the only working AI
paradigm in 1980.
- One train of thinking is that all human reasoning can be
described by rules.
- A lot of rule based systems have been developed (e.g. Clips),
and rule based systems are Turing complete.
- Consequently, there has been a lot of work in learning rules.
-
Association rules are mechanism to find interesting correlations
between data items when there are large number of data items. An
example is finding which items to stock next to each other at Tescos.
- Decision Trees can be automatically generated using a form of
supervised learning.
- Some standard algorithms for building these rules are C4.5 and ID3.
- Finally, there are some learning algorithms for cognitive architectures
(like Soar and ACT) which can generate new rules given an existing
rule based system.
