Uncertainty in Fuzzy Expert System
ÆÛÁöÈ®·üÀÌ Àü¹®°¡½Ã½ºÅÛ¿¡ »ç¿ëµÉ ¶§ ±âÁ¸ÀÇ È®·üÃ߷аú´Â Â÷À̰¡ ÀÖ´Ù. ÀüÇüÀûÀÎ fuzzy ruleÀ» »ý°¢Çغ¸ÀÚ. If X is F then Y is G (È®·ü B¸¦ °¡Áü) ¶ó´Â ruleÀ» Á¶°ÇÈ®·ü·Î ¾²¸é ´ÙÀ½°ú °°´Ù. P ( Y is G | X is F ) = B Áï ÀüÅëÀûÀÎ È®·üÀÌ·ÐÀ» »ç¿ëÇÑ ±âÁ¸ Àü¹®°¡½Ã½ºÅÛ¿¡¼´Â ´ÙÀ½°ú °°´Ù. P ( Y is not G | X is F ) = 1 - B
±×·¯³ª F °¡ fuzzy setÀ̶ó¸é À§ÀÇ °ÍÀº ´ÙÀ½°ú °°ÀÌ Ç¥ÇöµÈ´Ù. P ( Y is G | X is F ) + P ( Y is not G | X is F ) ¡Ã 1 ¿Ö³ÄÇϸé fuzzy number ¿¡¼´Â true, false °£ÀÇ ÁßøµÇ´Â ºÎºÐÀÌ Àֱ⠶§¹®ÀÌ´Ù. ÀϹÝÀûÀ¸·Î ÆÛÁö ½Ã½ºÅÛ¿¡¼´Â P(H | E) ÀÌ ¹Ýµå½Ã 1 - P(H¡Ç| E) ¿Í °°Àº °ÍÀº ¾Æ´Ï´Ù
ÆÛÁöÀü¹®°¡½Ã½ºÅÛ¿¡¼ 3 ¿µ¿ª¿¡¼ÀÇ fuzziness°¡ ÀÖÀ»¼ö ÀÖ´Ù.
1. rule¿¡¼ÀÇ Á¶°ÇºÎ and/or °á·ÐºÎ
If X is F then Y is G
If X is F then Y is G with CF = a
CF ´Â È®½Åµµ¸¦ ÀǹÌÇϸç 0.5 µîÀÇ ¼öÄ¡°ªÀ» °¡Áø´Ù
2. antecedent °ú fact »çÀÌÀÇ ºÎºÐÀûÀÎ match
ºñÆÛÁö Àü¹®°¡½Ã½ºÅÛ¿¡¼´Â Á¶°ÇºÎ°¡ fact¿Í Á¤È®ÇÏ°Ô ÆÐÅϸÅÄ¡µÇÁö ¾ÊÀ¸¸é ruleÀº fireÇÏÁö ¾Ê´Â´Ù. ±×·¯³ª ÆÛÁöÀü¹®°¡½Ã½ºÅÛ¿¡¼´Â ¸ðµç °ÍÀÌ Á¤µµÀÇ ¹®Á¦À̰í threshold°¡ ¼³Á¤µÇ¾îÀÖÁö ¾Ê´Ù¸é ¸ðµç ruleµéÀÌ ¾î´À¹üÀ§³»¿¡¼ fireµÉ¼ö ÀÖ´Ù.
3. most ¿Í °°Àº fuzzy quantifier, very likely,quite true, definitely possible µîµîÀÇ qualifier
¸íÁ¦ (proposition) ´Â ´ë°³ fuzzy quantifier¸¦ °¡Áø´Ù. (implicit and/or explicit)
¿¹¸¦µé¸é disposition (~°æÇâÀÌ ÀÖÀ½) À» µé ¼ö ÀÖ´Ù
dispositionÀº º¸Åë trueÀÎ ¸íÁ¦¸¦ ÀǹÌÇϸç ÀüÇüÀûÀÎ ÇüÅ´ ´ÙÀ½°ú °°´Ù
Usually (X is R)
¿©±â¼ Usually´Â ÇÔÃàµÈ fuzzy quantifierÀ̸ç X ´Â ±×°ªÀ» ÃëÇÏ´Â ÇѰ踦 °¡Áö´Â Á¦¾àº¯¼öÀ̰í R Àº º¯¼ö X ¿¡ ÀÛ¿ëÇÏ´Â Á¦¾à°ü°èÀÌ´Ù.
Àΰ£ÀÌ ½Ç¼¼°è¿¡¼ ¾Ë°í ÀÖ´Â ¸¹Àº heuristic ruleµéÀº dispositionµéÀÌ¸ç ¶ÇÇÑ commonsense knowledge´Â ±âº»ÀûÀ¸·Î ½Ç¼¼°è¿¡ ´ëÇÑ dispositionÀÇ ÁýÇÕÀÌ´Ù.
disposition Àº ´ÙÀ½°ú °°Àº ¸íÈ®ÇÑ ¸íÁ¦ÇüÅ·Πº¯È¯µÉ¼ö ÀÖ´Ù.
"desserts are wonderful" ¶ó´Â ¸íÁ¦¿¡ ´ëÇØ
p = usually desserts are wonderful
p = most desserts are wonderful
¶ÇÇÑ À̰ÍÀº ´ÙÀ½°ú °°Àº heuristic rule·Î Ç¥ÇöµÉ¼ö ÀÖ´Ù.
r = If x is a dessert
then it is likely that x is wonderful
fuzzy ¿¡¼ÀÇ rule of inference ¿¡´Â ´ÙÀ½°ú °°Àº °ÍÀÌ ÀÖ´Ù
1. entailment principle ( entail : (³í¸®Àû ÇÊ¿¬À¸·Î½á) ÀǹÌÇÏ´Ù )
X is F
F ¡ø G
-------
X is G
2. dispositional entailment : usually °¡ always °¡ µÇ´Â Á¦ÇÑµÈ °æ¿ì
usually ( X is F )
F ¡ø G
-----------------
usually ( X is G )
3. compositional rule
X is F
( X, Y ) is R : R Àº ÀÌÁøº¯¼ö (X, Y) ¿¡ ´ëÇÑ ÀÌÁø°ü°è
--------------
Y is F ¡Ý R
À̰ÍÀº sup [ min...] ¿¡ ÀÇÇØ¼ ±¸ÇØÁø´Ù.
sup Àº supremum ÀÇ ¾à¾î·Î¼ least upper bound ·Î Á¤ÀÇ µÈ´Ù. ´ë°³ sup ´Â max ¿Í °°Àº °³³äÀÌÁö¸¸ ¿¹¸¦µé¾î " 0 º¸´Ù ÀÛÀº maximum real number " ¸¦ ±¸ÇÒ ¶§ sup ´Â least upper bound·Î¼ 0À» ÃëÇÏ°Ô µÇ´Â °ÍÀÌ´Ù.
4. generalized modus ponens
X is F
Y is G if X is H
------------------
Y is F ¡Ý (H¡Ç¢Á G)
H¡Ç´Â H ÀÇ fuzzy negationÀÌ´Ù. ¿©±â¼ bounded sumÀÌ Á¤ÀÇµÉ ¼ö ÀÖ´Ù.....
generalize modus ponens´Â antecedent "X is H" °¡ premise "X is F" ¿Í µ¿ÀÏÇÒ °ÍÀ» ¿ä±¸ÇÏÁö´Â ¾Ê´Â´Ù.À̰ÍÀº Á¤È®È÷ ¸ÅÄ¡µÉ °ÍÀ» ¿ä±¸ÇÏ´Â ÀüÅëÀûÀÎ ³í¸®¿Í´Â ¸Å¿ì ´Ù¸£´Ù. generalized modus ponens´Â Ãß·ÐÀÇ compositional ruleÀÇ Æ¯º°ÇÑ °æ¿ìÀÌ´Ù. ±âÁ¸ÀÇ Àü¹®°¡½Ã½ºÅÛ¿¡¼´Â modus ponens°¡ Ãß·ÐÀÇ ±âº» ruleÀ̾úÁö¸¸ fuzzy Àü¹®°¡½Ã½ºÅÛ¿¡¼´Â Ãß·ÐÀÇ compositional ruleÀÌ ±âº» ruleÀÌ´Ù.
approximate reasoningÀ» »ç¿ëÇÑ Àü¹®°¡½Ã½ºÅÛµéÀº 2 °¡Áö ¹æ¹ýÁß Çϳª¸¦ »ç¿ëÇÑ´Ù.
Çϳª´Â truth value restrictionÀÌ°í ´Ù¸¥ Çϳª´Â compositional inferenceÀÌ´Ù. WhalenÀÇ survey¿¡ µû¸£¸é 11 °³ÀÇ ÆÛÁöÀü¹®°¡½Ã½ºÅÛÁß¿¡¼ °ÅÀÇ ÀüºÎ°¡ compositional inference¸¦ »ç¿ëÇÏ¿´´Ù
(Giarratano 1989)