Re: "Rules" to form rules??
William Siler (williamsiler@delphi.com)
Fri, 30 Jun 95 20:28:08 -0500
Jive Dadson <jdadson@ix.netcom.com> writes:
>>Lets say I have the following rules:
>>
>>if high(a) OR high(b) then high(c)
>>
>>if high(a) AND med(b) then med(c)
>>
>>
>>Are the above two rules VALID at the same time??
Apparently nobody involved here has any experience with a fuzzy rule-based
system.
I assume that A, B and C are fuzzy sets; that a has at least one member, high,
and
that b and c have at least two members, high and med. I assume further that the
grades of membership involved are high enough so that both rules are
concurrently fireable.
There are two ways of handling this situation, depending on whether you are
firing your rules sequentially (one at a time), or firing your rules in
parallel.
The first (and worst) way is to pick one rule for firing by some algorithm like
the MEA, and fire it, storing the antecedent confidence as the grade of
membership
of "high" (or "med", depending on which rule you pick) in fuzzy set b. At that
point, if no rules are newly fireable, you backtrack and fire the other rule,
storing its antecedent confidence as the grade of membership of "med" (or
high) in fuzzy set b.
The second (and best) way is to fire the two rules effectively in parallel.
Then the proper grades of membership in "high" and "med" in fuzzy set b
are both stored effectively simultaneously.
The point is that fuzzy sets have no problem at all with ambiguities or, for
that matter, with contradictions. From this simple fact comes much of the
strength of fuzzy systems theory.
>You win the prize! You have hit on the reason "fuzzy logic" is not
>logic.
Friend, perhaps fuzzy logic is not yours, but that's too bad. Many of the
criticisms of fuzzy logic apply with equal force to ANY multivalued logic, from
Lukasiewicz to Zadeh. The whole point to multivalued logics of any kind is to
help us deal with the re
al world more effectively, and to reason in terms less machine-like and more
human.
------------------------------