Re: "Most appropriate terms"
William Siler (williamsiler@delphi.com)
Fri, 19 May 95 20:57:25 -0500
Linguistic variables such as Temperature in your message have
usually some ambiguity - i.e. more than one term applies at
once with varying degrees of validity. In any kind of s system,
be it fuzzy control or fuzzy reasoning, that ambiguity is a
GOOD THING and lends sturdiness to a rule-based system which
operates on things like Temperature as an input set. Similarly
for output: a linguistic variable (say Control) which appears
as output is likely to have some ambiguity also, and it is not
only unecessary but undesirable to decide which term is "best",
since all terms with non-zero truth values will enter into the
defuzzification process.
There is a type of output fuzzy set which does need resolution
of terms; this occurs in fuzzy reasoning, when the output fuzzy
set is one of classifications. Now we don't have ambiguities,
but contradictions, and we have to find out which term is
correct. Some believe in taking the most likely term; we don't
beleve that is best unless the truth values of the most and
next most likely have a wide gap. We think that the best way to
handle this situation is to write rules using more data or
looking at existing data more critically to decide which term
is the proper one to use.
------------------------------