Intelligent Control Methods Lecture 7: Knowledge representation Slovak University of Technology Faculty of Material Science and Technology in Trnava
2 The quality of AI-system is determined by used knowledge. The knowledge is explicit formulated and stored in system (exception: NN and GA) Knowledge representation: Formulation of knowledge in the symbolic form (so that they can be effective stored, manipulated and exploited)
3 Knowledge representation schemes: The scheme determines the way of knowledge organization in system. The manner of knowledge manipulation is by the scheme determined, too (but not single-valued). Used schemas: declarative production rules frames associative nets procedural
4 Declarative representation Knowledge is given (declared) as description of discrete states of problem environment description of possible changes of states to another states (operators) Description is separated from solution process Solution methods: resolution method, searching in state space, GPS, STRIPS,... Typical representative: predicate calculus of the 1st order
5 Predicate calculus (example of declarative knowledge representation) The base element is a predicate (fact about an object or about relationship between objects; in case of object concretization the predicate = statement (preposition)). P(x) – x is tall (or x wears hat,...) L(x,y) – x > y (or x loves y) – according to interpretation Knowledge is formulated like logical expression consisting of quantifiers, predicates and logical operators: (x,y,z): parent(x,y) parent(y,z) man(x) grandfather(x,z)
6 Declarative representation Advantages: simplicity theoretical background good connection to solving methods Disadvantages: without time and qualitative changes without quantitative information (4 from 5, for more then ) without uncertainty (only yes - no) the complexity of solution grows exponential with formulas number
7 Production rules the most used schema Form: IF antecedent THEN consequent IF assumption THEN conclusion (declarative interpretation) IF condition THEN action (procedural interpretation) Antecedent is 1 or more assertions, connected by logical operations. Consequent is a fact or an action. If the liquid is white, then the liquid is milk. If the liquid is white, then it is necessary to call firemen.
8 Production rules - chaining Consequent in 1 rule can be antecedent in another one – rules are chained. R1:If the patient worked before 1985 like tiler(A1) then the patient worked with asbestos.(C1) R2:If the patient worked with asbestos(A2) and the patient worked in closed spaces(A3) then his lungs contain asbestos dust.(C2) A1 C1=A2 A3 C2
9 Production rules Chained rules create graph (tree). Solving methods: forward chaining (new facts deduction till the goal fact is derived) backward chaining (verification of the goal fact truth by verification its antecedent)
10 Production rules Advantages: simple creation simple supplement (to the width, to the depth)) simple modification simple inference Disadvantages: ineffective inference differ from human knowledge
11 Frames Minsky 1975, beginning of OOP. The frame is a structure, which represents some stereotypic situation. blank (form) with couples (columns) attribute - value Empty: prototype frame With at least one completed value: frame instance
12 Frames Rubriky: hodnota instruction for value determination (function) default inherent value frame Task solution: Value determination in some column.
13 Semantic nets Oriented graphs nodes – objects, concepts, actions,... arcs – relations STU belongs to MTF Trnava seats MTF belongs to STU
14 Semantic nets: Solution Knowledge: Trnava seats MTF belongs to STU Task (problem): To what does MTF belong? X belongs to MTF Knowledge: Trnava seats MtF belongs to STU X=STU
15 Semantic nets difficulties with logical operations representation difficult quantification used by memory modeling and by natural language analysis
16 Procedural representation Knowledge in form of procedures description not important (what is it), but the way of manipulation (how) the solution means the steps sequence to answer (algorithm) typical for procedural („ordinary“) programming languages In AI: Automatic procedures call in PROLOG and PLANNER