home *** CD-ROM | disk | FTP | other *** search
- Newsgroups: comp.ai.fuzzy
- Path: sparky!uunet!gatech!rpi!batcomputer!elias
- From: elias@fitz.TC.Cornell.EDU (Doug Elias)
- Subject: Fuzzy Logic and \"George is 'smart'" -- 0.7/
- In-Reply-To: arms@cs.UAlberta.CA's message of Wed, 20 Jan 1993 17:42:13 GMT
- Message-ID: <ELIAS.93Jan22102643@fitz.TC.Cornell.EDU>
- Sender: news@tc.cornell.edu
- Nntp-Posting-Host: wonton.tc.cornell.edu
- Organization: Software and Consulting Support Group, Cornell Theory Center,
- C.U.
- Date: Fri, 22 Jan 1993 15:26:43 GMT
- Lines: 160
-
-
- arms@cs.UAlberta.CA (Bill Armstrong) comments that the statement:
-
- George is smart -- 0.7
-
- ..."express(es) in fuzzy terms something like 70% of people would apply
- this *description* to George."
-
- Instead, this statement, if viewed as a proposition in fuzzy logic, is
- saying that:
- "i-the-speaker feel that this claim, "George is smart", has a truth
- value, within the range of 0 [totally false] to 1 [totally true], of
- .7" (this is only a simple interpretation of the statement; the more
- normally accepted interpretation is described a few paragraphs down);
- that means that, with all of the evidence available, the observer
- subjectively feels that the claim is more-true-than-not, but isn't
- completely true.
-
- The two are very different in interpretation:
- *) the former assumes that each observer makes a classical "is/is-not"
- decision, and then you sum over all of those binary 1-or-0's;
- *) the latter says that (for a single observer) the strength-of-belief-in,
- or the weight-of-evidence-for the statement is being reported.
-
- Translating this into fuzzy set theory provides a different perspective:
- *) classically, the logical claim "George is smart (is true)" becomes the
- set theoretic "George is a member of the set 'smart'";
- *) the fuzzy-logical claim "George is smart (is true) to the .7
- membership-grade-level" becomes the fuzzy-set-theoretic "George is
- a member of the fuzzily-defined subset 'smart' (of the overall
- classification-set "intelligence", which consists of
- ('rock-stupid', 'real-dumb', 'dumb', 'dim', 'normal', 'quick',
- 'smart', 'real-smart', 'genius')) to the .7 degree of validity".
-
- In the former, "smart" is a characteristic you are considered to
- either have or not have; in the latter you can "have" it to varying
- degrees over a range of possible values -- for example, you may be
- 'real-smart' when it comes to things like programming, but
- 'rock-stupid' when it comes to keeping yourself in shape, and 'dim'
- when it comes to finances, etc,: a fuzzy classification scheme allows
- you to specify how "well" you represent each category all-at-once, and
- your overall "intelligence" classification is the join of your
- representativeness in each one of the possible categories.
-
- Getting back to fuzzy logic, "George is smart" would be normally be
- applied to the fuzzy set consisting of possible truth-values, say:
- (totally-false, false, slightly-false, neutral, slightly-true, true,
- totally-true)
- and the "0.7" would be the membership grade associated with one of
- those categories. This assignment would result, then, in a fuzzy-logical
- proposition which could, for example, be combined with other such
- claims using fuzzy-logical operators such as "not", "and", "or", and
- the logical formulas that can constructed from them. The "classical"
- fuzzy operator-functions "1 - m()" for "negation", "max" for "or", and
- "min" for "and" are only single members of essentially an infinite
- number of operator-functions that can be used to implement the symbols
- "~", "&", and "|".
-
- Armstrong uses the term "linguistic" for what is called "ambiguous",
- or "not completely assignable to a single category", within fuzzy set
- theory. "Ambiguous" classifications differ from probabilistic ones
- precisely because probability theory requires that a given element of
- the universal set be found in one-and-only-one subset of the
- classification (there must be no overlap among the possible
- alternatives), while fuzzy theory allows an element to maintain
- membership in multiple categories at the same time -- i.e.,
- probability theory enforces the "Law of the Excluded Middle", while
- fuzzy theory relaxs it (another indication of how probability theory
- is a proper subset of fuzzy theory, obtained when the "degree-of-relax-
- ation of the Law of the Excluded Middle" is set to 0).
-
- Applying these concepts to Armstrong's example propositions of "'George
- is smart' == S" and "'George is rich' == R", using the fuzzy operators
- ~, &, and |, would then result each full fuzzy theorem (e.g., "S & ~R")
- being evaluated and that result then being graphed onto the various
- "strength-of-truthity" classes described previously. Let's take this
- one case as an example, and set:
- m(S) = .7
- m(R) = .4
- Then m(S & ~R) = .7 & .6 = .7, using the classical fuzzy negation and
- conjunction operaters. Now, this simply states that George is a
- member of the set of people who are both smart and not rich to the .7
- degree, i.e., more-a-member-than-not. We can now evaluate its various
- truthvalues by using the following:
-
- 1 ____(At_Least)_Very Weakly True__________________
- M | / | - - -|
- e |*/**************************** - | - |
- m |/ - | - - |
- b || (At Least) - | - |
- e.75| Weakly True | - - |
- r || - | - |
- s ||******************************** - - |
- h || - - | |
- i || - | - |
- p .5| - - | |
- || - | - |
- G || - True- | |
- r || - | - |
- a.25|**-*********-********Strongly True- |
- d || - - | /
- e ||- - - | /
- || - - | Very Strongly True
- || ----------------------------------------------/
- 0/0________.25___________.5________|__.75_ __________1
- |
- Propositional Truth Value .7
- |
- M 1--------(At-Least)-Very-Weakly-False----------- \
- e || - - | \
- m || - - | \
- b ||- - - | \
- e || - (At Least) | |
- r.75| _ - | |
- s || - Weakly | |
- h ||****************-************False |
- i || False - | - |
- p || - - | |
- .5| _ | |
- G || Strongly False _ | - |
- r || _ | |
- a ||*********-********************** - |
- d .25 | - - |
- e |\ - | - |
- | \ | - - |
- |Very*************************** -*| - |
- | Strongly False | - - |
- |____\_______________________________________________|
- 0/0________.25___________.5________|__.75_ __________1
-
-
- The final result, then, is the determination that
- m((George is smart) & (George is not rich)) == .7
- is a member of the fuzzy-set "Truth Value" with membership grades:
- 0/Very Strongly True
- .25/Strongly True
- .70/True
- .85/At Least Weakly True
- 1/At Least Very Weakly True
- 1/At Least Very Weakly False
- .65/At Least Weakly False
- .30/False
- .25/Strongly False
- 0/Very Strongly False
-
- Now, for using this conjunction within a fuzzy logical evaluation, you
- first determine what level of "truth value" you want it to have, then
- plug in the appropriate membership grade for that level of "truth-ity".
-
-
- Whew. "Fuzzy Sets, Uncertainty and Information Theory", by Klir and
- Folger (Plenum Press or Prentice Hall, can't remember which), 1988,
- covers this in much more detail, and with much better graphics (sic).
-
- doug
- --
- # ____ |Internet: elias@tc.cornell.edu
- #dr _|_)oug|USmail: Sci.Comp.Support/Cornell Theory Center
- # (_| | 737 TheoryCtrBldg/C.U./Ithaca/N.Y./14853-3801
- # (_|__ |MaBelle: 607-254-8686 Fax: 607-254-8888
-