《信息科学原理PIS-07(2)》由会员分享,可在线阅读,更多相关《信息科学原理PIS-07(2)(36页珍藏版)》请在金锄头文库上搜索。
1、Principles of Information ScienceChapter 7(2)Principles of Information Conversion (2)- Strategy Creation TheoryList of Contents1. Information Conversion (2): from Knowledge to Strategy2. Classical Model of Decision-Making3. Information Theory of Strategy-Creation4. Unified TheoryFrom Knowledge to St
2、rategy (Information Conversion: No.2)Strategy Creation1.Model of Information Conversion (2)Information Conversion (2)Knowledge about the Problem and EnvironmentStrategy for solving problem under the given environmentGoal to be soughtConcepts: Knowledge and StrategyKnowledge: A set of descriptions on
3、 the states and the laws of states varyingfor certain categories of objects. Strategy: A description on sequence of well organized actions for solvingspecific problems under certain environment toward certain goal. Strategy can also be regarded as a highest level of information indicating the states
4、 and the manner of states varying during problem solving.Knowledge EcologyCommonsenseKnowledgeRegular KnowledgeEmpirical KnowledgeInnate KnowledgeInfoUnder-maturedNormally-maturedOver-MaturedKnowledge grows from under-matured stage to normal-matured stage and further to over-matured stage.Training t
5、he output is the one desired. Otherwise back to Step 3, till the distance is acceptable.This algorithm has closed relation to Neural Networks Reasoning: Regular Knowledge Regular StrategyPInitial State of DatabaseOperationNew StateGoalMatch GRule BaseKnowledge BaseRule SequenceNYStrategyDistance Ind
6、icationControlCConceptualized Algorithm: Regular Knowledge Regular Strategy1.Given P(the Initial State of the Problem), C (the Knowledge and the Rule Bases) and G(the Final State of the Problem) 2, Select the best Rule, from the Rule Base, so that whose left side matches the Initial State while whos
7、e right side leads to such a New State whose distance to the Goal is the minimum compared with other selections by using the Knowledge 3, Check the New State thus obtained. If its distance to the Goal is the smallest one but unequal to zero, do the same thing from the New State as did in step 2 4, O
8、therwise, re-select a rule at step 2 5, Until the distance between the New State and the Goal equal to zero, or sufficiently small, go loop from step 2 to 4.The sequence of the rule applications is the strategy sought.Conceptualized Model: Commonsense Knowledge Commonsense StrategyInput PatternsSens
9、or-Motor CorrespondenceOutput ActionsPattern Recognition: Commonsense Knowledge Commonsense StrategyStep 1 Given a class of problems in which the strategy has fixed relation to the pattern of input. Store the relations P A in KB.Step 2 Input a problem and recognize its pattern.Step 3 Produce the act
10、ion which is related to the pattern recognized.This is the Sensor-motor System AlgorithmClassical Model of Decision-Making- Strategy Selection2.An Example: Umbrella TrickyBenefits TableWeather (Info)BenefitAction (Strategy)SunnyRainingCarrya(1)Not Carry a(2)p1-pc(1,1)c(1,2)c(2,1)c(2,2)Decision RuleC
11、(1) = p c(1,1) + (1-p) c(1,2)C(2) = p c(2,1) + (1-p) c(2,2)Calculate the average benefit for each action:Choose the action with bigger benefit as the strategy:If C(1) C(2) then a(1) is chosen; Otherwise, a(2) is chosen.Decision-Making based on Comprehensive Information Theory (CIT)3.The Problem X ha
12、s L possible states: x(n), n(1,L)The certainty distribution: c(n), n(1,L)The possible action strategy: a(k), k(1,K)The possible outcomes for each a(k): b(k,l) c(1) c(l) c(L)b(1,1) b(1,l) b(1,L)b(k,1) b(k,l) b(k,L)b(K,1) b(K,l) b(K,L)a(1)a(k)a(K).B(1)B(k)B(K)CIT-based Decision MatrixCIT-based Decisio
13、n TreeA. .a(1)a(k)a(K)c(1)c(L)c(l)c(1)c(L)c(l)b(1,1)b(1,l)b(1,L)b(k,1) b(k,l) b(k,L)b(K,1) b(K,l)b(K,L)CIT-based Decision RulesFor each a(k), define x(k) = c(l) tb(k,l) ub(k,l)=c(l)t(k,l)u(k,l), l (1,L) as the integrative utility of the action strategy a(k), and then we have I( x(k) ) as the measure
14、 of integrative pragmatic information of a(k).The decision rule can thus be set to be the following:If I( x(k) ) = I( x(k) ), then a(k) is chosenMax k k4.Unified Theory of Decision-MakingCIT-based Rule and Bayes RuleLet t(k,l) = 1, k, l, we then havex(k) = c(l) u(k,l)I( x(k) ) can be reduced to c(l)
15、 u(k,l)l (1,L)The ruleIf I( x(k) ) = I( x(k) )Max k kThen a(k) is chosenreduces to the well-known Bayes Rule. CIT-based Rule and Min-Max RuleIf, further, let c(l) = 1, l, then only u(k,l) needs to be considered. Min-Max, Max-Min, Min-Min, Max-Max are all special cases of this kind of rules.u(1,1) u(1,l) u(1,L)u(k,1) u(k,l) u(k,L)u(K,1) u(K,l) u(K,L)Unified Theory of Decision-Making1, Classic rules of DM is special case of CIT-based DM Rule, as seen above.2, If CIT-based Rule, which is Integrative Pragmatic Information as seen onp.17, is taken as the Goal in algorithm of Decision-