Credibility Theory It is a new branch of mathematics that studies the behavior of fuzzy phenomena. Baoding Liu Uncertainty Theory Laboratory Department of Mathematical Sciences Tsinghua University Uncertainty Theory & Uncertain Programming UTLAB Fashion of Mathematics 2300 Years Ago: Euclid: “Elements”, First Axiomatic System 1899: Hilbert: Independence, Consistency, Completeness 1931: K. Godel: Incompleteness Theorem 1933: Kolmogoroff: Probability Theory 2004: B. Liu: Credibility Theory Baoding Liu Tsinghua University http://orsc.edu.cn/~liu Uncertainty Theory & Uncertain Programming UTLAB Why I do not possibility measure? (a) Possibility is not self-dual, i.e., Pos{ A}+Pos{ Ac } 1. (b) I will spend "about $300": (200,300, 400). In order to cover my expenses with maximum chance, how much needed? Pos{ x} 1 x 300 A self-dual measure is absolutely needed! Baoding Liu Tsinghua University http://orsc.edu.cn/~liu Uncertainty Theory & Uncertain Programming UTLAB Five Axioms Axiom 1. Cr{}=1. Axiom 2. Cr{ A} Cr{B} whenever A B. Axiom 3. Cr is self-dual, i.e., Cr{ A} Cr{ Ac } 1. Axiom 4. Cr i Ai 0.5=supi Cr{ Ai } if Cr{ Ai } 0.5 for each i. Axiom 5. For each A P (1 2 n ), we have sup min Crk { k }, if sup min Crk { k } 0.5 1 k n (1 , , n )A (1 , , n )A 1 k n Cr{ A} min Crk { k } 0.5, if sup min Crk { k } 0.5. 1 ( , sup (1 , , n )A 1 k n , n )Ac 1 k n 1 Independence (Yes) Consistency (?) Completeness (Absolutely No) Baoding Liu Tsinghua University http://orsc.edu.cn/~liu UTLAB Uncertainty Theory & Uncertain Programming Credibility Subadditivity Theorem Liu (UT, 2004) Credibility measure is subadditive, i.e., Cr{A B} Cr{A}+Cr{B}. A credibility measure is additive if and only if there are at most two elements in universal set. Baoding Liu Tsinghua University http://orsc.edu.cn/~liu UTLAB Uncertainty Theory & Uncertain Programming Credibility Semicontinuity Laws Liu (UT, 2004) Theorem: Let (, P(), Cr) be a credibility space, and A1 , A2 , P(). Then lim Cr{ Ai } Cr{ A} i if one of the following conditions is satisfied: (a) Cr{A} 0.5 and Ai A; (b) lim Cr{ Ai } 0.5 and Ai A; i (c) Cr{A} 0.5 and Ai A; (d) lim Cr{ Ai } 0.5 and Ai A. i Baoding Liu Tsinghua University http://orsc.edu.cn/~liu UTLAB Uncertainty Theory & Uncertain Programming Credibility Extension Theorem Li and Liu (2005) If Cr{ } satisfies the credibility extension condition, sup Cr{ } 0.5, Cr{ *}+ sup Cr{ } 1 if Cr{ *} 0.5, * then Cr{ } has a unique extension to a credibility measure on P(), sup Cr{ }, if sup Cr{ } 0.5 A A Cr{A}= Cr{ } 0.5, if sup Cr{ } 0.5. 1 sup c A A Baoding Liu Tsinghua University http://orsc.edu.cn/~liu Uncertainty Theory & Uncertain Programming UTLAB Fuzzy Variable Definition : A fuzzy variable is a function from a credibility space (,P(),Cr) to the set of real numbers. Membership function: ( x) 2Cr x 1. Sufficient and Necessary Condition : A function : [0,1] is a membership function iff sup ( x) 1. Baoding Liu Tsinghua University http://orsc.edu.cn/~liu Uncertainty Theory & Uncertain Programming UTLAB Credibility Measure by Membership Function Liu and Liu (IEEE TFS, 2002) Let be a fuzzy variable with membership function . Then 1 Cr{ A} sup ( x) 1 sup ( x) . 2 xA xAc Baoding Liu Tsinghua University http://orsc.edu.cn/~liu Uncertainty Theory & Uncertain Programming UTLAB Independent Fuzzy Variables Zadeh (1978), Nahmias (1978), Yager (1992), Liu (2004) Liu and Gao (2005) The fuzzy variables 1 , 2 , , m are independent if m Cr {i Bi } min Cr{i Bi } i 1 1i m for any sets B1 , B2 , , Bm of . Baoding Liu Tsinghua University http://orsc.edu.cn/~liu UTLAB Uncertainty Theory & Uncertain Programming Theorem: Extension Principle of Zadeh Let 1 , 2 , , n be independent fuzzy variables with membership functions 1 , 2 , , n , respectively. Then the membership function of f (1 , 2 , ( x) sup , n ) is min i ( xi ). x f ( x1 , x2 , , xn ) 1i n It is only applicable to independent fuzzy variables. It is treated as a theorem, not a postulate. Baoding Liu Tsinghua University http://orsc.edu.cn/~liu Uncertainty Theory & Uncertain Programming UTLAB Expected Value Liu and Liu (IEEE TFS, 2002) Let be a fuzzy variable. Then the expected value of is defined by E[ ] 0 0 Cr{ r}dr Cr{ r}dr provided that at least one of the two integrals is finite. Yager (1981, 2002): discrete fuzzy variable Dubois and Prade (1987): continuous fuzzy variable Baoding Liu Tsinghua University http://orsc.edu.cn/~liu UTLAB Uncertainty Theory & Uncertain Programming Why the Definition Reasonable? (i) Since credibility is self-dual, the expected value 0 0 E[ ] Cr{ r}dr Cr{ r}dr is a type of Choquet integral. (ii) It has an identical form with random case, 0 0 E[ ] Pr{ r}dr Pr{ r}dr. Baoding Liu Tsinghua University http://orsc.edu.cn/~liu UTLAB Uncertainty Theory & Uncertain Programming Credibility Distribution Liu (TPUP, 2002) The credibility distribution : ( , ) [0,1] of a fuzzy variable is defined by ( x) Cr{ | ( ) x}. Baoding Liu Tsinghua University http://orsc.edu.cn/~liu UTLAB Uncertainty Theory & Uncertain Programming A Sufficient and Necessary Condition Liu (UT, 2004) A function : [0,1] is a credibility distribution if and only if it is an increasing function with lim ( x) 0.5 lim ( x) x x lim ( y ) ( x) if lim ( y ) 0.5 or ( x) 0.5. y x y x Baoding Liu Tsinghua University http://orsc.edu.cn/~liu Uncertainty Theory & Uncertain Programming Entropy UTLAB (Li and Liu, 2005) What is the degree of difficulty of predicting the specified value that a fuzzy variable will take? n H [ ] S (Cr{ xi }) S (t ) t ln t (1 t ) ln(1 t ) i 1 Baoding Liu Tsinghua University http://orsc.edu.cn/~liu UTLAB Uncertainty Theory & Uncertain Programming Random Phenomena Fuzzy Phenomena (1654) (1965) Probability Theory (1933) Credibility Theory (2004) Probability Credibility Three Axioms Five Axioms Sum "+" Maximization " " Product " " Minimization " " Baoding Liu Tsinghua University http://orsc.edu.cn/~liu UTLAB Uncertainty Theory & Uncertain Programming Essential of Uncertainty Theory Probability Theory Measure Theory + Function Theory Credibility Theory Two basic problems? [1] Measure of Union: { A B} { A} {B} " " {A B} {A} {B} " " [2] Measure of Product: { A B} { A} {B} " " {A B} { A} {B} " " Baoding Liu Tsinghua University http://orsc.edu.cn/~liu UTLAB Uncertainty Theory & Uncertain Programming What Mathematics Made? (+,)-Axiomatic System: Probability Theory (,)-Axiomatic System: Credibility Theory (,)-Axiomatic System: Nonclassical Credibility Theory (+,)-Axiomatic System: Inconsistent Baoding Liu Tsinghua University http://orsc.edu.cn/~liu UTLAB Uncertainty Theory & Uncertain Programming Fuzzy Programming max f ( x, ) subject to: g ( x, ) 0, j j 1, 2, ,m x - Man proposes - God disposes It is not a mathematical model! Baoding Liu Tsinghua University http://orsc.edu.cn/~liu UTLAB Uncertainty Theory & Uncertain Programming The Simplest The Most Fundamental Problem Given two fuzzy variables and , which one is greater? Baoding Liu Tsinghua University http://orsc.edu.cn/~liu Uncertainty Theory & Uncertain Programming UTLAB Expected Value Criterion E[ ] E[ ]. Objective: max f ( x, ) max E[ f ( x, )] Constraint: g j ( x, ) 0 E[ g j ( x, )] 0 Baoding Liu Tsinghua University http://orsc.edu.cn/~liu Uncertainty Theory & Uncertain Programming UTLAB Fuzzy Expected Value Model Liu and Liu (IEEE TFS, 2002) Find the decision with maximum expected return subject to some expected constraints. max E[ f ( x, )] subject to : E[ g ( x, )] 0, j 1,2,, p j Baoding Liu Tsinghua University http://orsc.edu.cn/~liu Uncertainty Theory & Uncertain Programming UTLAB Optimistic Value Criterion sup ( ) sup ( ) Objective: max f ( x, ) x (Undefined) max max f : Cr{ f ( x, ) f } x Constraint: Baoding Liu f g j ( x, ) 0 Cr g j ( x, ) 0 Tsinghua University http://orsc.edu.cn/~liu Uncertainty Theory & Uncertain Programming UTLAB (Maximax) Chance-Constrained Programming Liu and Iwamura (FSS, 1998) Maximize the optimistic value subject to chance constraints. max max f max f x f subject to : Cr{ f ( x, ) f } Crg j ( x, ) 0, j 1,2,, p Baoding Liu Tsinghua University http://orsc.edu.cn/~liu Uncertainty Theory & Uncertain Programming UTLAB Pessimistic Value Criterion inf ( ) inf ( ) Objective: max f ( x, ) x max min f : Cr{ f ( x, ) f } x f Constraint: g j ( x, ) 0 Cr g j ( x, ) 0 Baoding Liu Tsinghua University http://orsc.edu.cn/~liu Uncertainty Theory & Uncertain Programming UTLAB (Minimax) Chance-Constrained Programming Liu (IS, 1998) Maximize the pessimistic value subject to chance constraints. max min f f x subject to : Cr{ f ( x, ) f } Crg ( x, ) 0, j 1,2,, p j Baoding Liu Tsinghua University http://orsc.edu.cn/~liu Uncertainty Theory & Uncertain Programming UTLAB Credibility Criterion Cr r Cr r Remark: Different choice of r produces different ordership. Baoding Liu Tsinghua University http://orsc.edu.cn/~liu UTLAB Uncertainty Theory & Uncertain Programming Fuzzy Dependent-Chance Programming Liu (IEEE TFS, 1999) Find the decision with maximum chance to meet the event in an uncertain environment. max Cr{hk ( x, ) 0, k 1, 2, , q} subject to: g j ( x, ) 0, j 1, 2, , p Baoding Liu Tsinghua University http://orsc.edu.cn/~liu UTLAB Uncertainty Theory & Uncertain Programming Classify Uncertain Programming via Graph Information Random Fuzzy Fuzzy random Fuzzy Stochastic Single-Objective P MOP GP DP MLP Philosophy EVM CCP DCP Maximax Minimax Structure Baoding Liu Tsinghua University http://orsc.edu.cn/~liu Uncertainty Theory & Uncertain Programming UTLAB Last Words [1] Liu B., Foundation of Uncertainty Theory. [2] Liu B., Introduction to Uncertain Programming. If you want an electronic copy of my book, or source files of hybrid intelligent algorithms, please download them from http://orsc.edu.cn/~liu Baoding Liu Tsinghua University http://orsc.edu.cn/~liu