Adhoc frames conceptual graphs
0 likes1,896 views
This document discusses different knowledge representation techniques used in artificial intelligence systems, including ad-hoc methods, heuristic reasoning methods, frames, associative networks, and conceptual graphs. It provides examples of each technique and how they can represent knowledge with examples from early expert systems like MYCIN. It also describes how conceptual graphs can be converted to and from first-order predicate logic.
![Conversion from FOPL to CG
• Put FOPL formula into prenex normal form & convert
all logic connectives to negation & conjunction.
• Every occurrence of Universal quantification is
replaced by ~E~x
• Every variable x & every occurrence of existential
quantification is then replaced with most general type
concept denote as [T: *x]
• Implication in CG can be represented with negation &
conjuction
• Ex: P Q can be written as ~[P ~[Q]]
#](https://image.slidesharecdn.com/adhocframesconceptualgraphs-121216111126-phpapp01/85/Adhoc-frames-conceptual-graphs-15-320.jpg)
Adhoc frames conceptual graphs
- 1. Reasoning & Conceptual Graphs Ayaz Ahmed Shariff K Asst. Professor, Dept. Of CSE Birla Institute of Technology International Centre Ras Al Khaimah #
- 2. AD-HOC Methods • Ad Hoc are methods deals with uncertainty are methods which have no formal theoretical basis. • Different ad-hoc procedures have been employed successively in no. of AI systems, particularly Expert Systems like MYCIN system. • MYCIN System: Earliest Expert System (ES) to diagnose meningitis & infectious blood diseases. • MYCIN’s KB composed of “if.... Then rules” to assess various forms of patient evidence. #
- 3. Contents • Ad Hoc Methods • Heuristic Reasoning Methods • Frames • Associative Networks • Conceptual Graphs #
- 4. AD HOC Methods... • Typical rule of MYCIN ES has form: • This is a rule that would be used by Inference mechanism to help to identify the organism IF: The stain of the organism is gram +ve, and The morphology of the organism is coccus, and The growth conformation of the organism is chains THEN: There us suggestive evidence (0.`7) that the identity of the organism is Streptococcus • The Ad Hoc method measures both belief & disbelief to represent degrees of confirmation respectively given hypothesis. • Represented by MB (H,E), is measure of increased#
- 5. Heuristic Reasoning Methods • The Heuristic methods are based on use of procedures, rules and other forms of encoded knowledge to achieve specified goals under uncertainty. • Using heuristics , one of the several alternative conclusions may be chosen through the strength of positive v/s negative evidence presented in form of justifications or endorsements. • For ex: SOLOMON a prototype system developed by Paul Cohen in 1985, endorsements in form of heuristics are used to reason about uncertainties associated with client’s investment portfolio. • Thus belief is associated with the rule is a “Subjective conditional probability” – P(H | E1, E2, E3) ; H is hypothesis, E is evidence #
- 6. Heuristic Reasoning Methods... • IF: Client income need is high & net worth is medium to high; THEN: Risk-Tolerance level is medium. • IF: Client tax bracket is high and risk-tolerance is low; THEN: Tax-exempt mutual funds are indicated. • IF: Client age is high & income needs are high & retirement income is medium; THEN: Risk-Tolerance is low. • IF: Two +ve endorsements are medium or high and one -ve endorsement is high; THEN: Favor positive choice. #
- 7. Associative Networks (AN) • Network representations provide means of structuring & exhibiting the structure in knowledge. It provides more natural way to map NLs and pictorial representation of objects. • Associative Networks introduced by Quillion in 1996 to model semantics of English words • Associative Networks are – Directed graphs with labelled nodes & arcs – Ex: 7.2 p128 – Bird -> class of objects, tweety -> class, properties -> color #
- 8. Syntax & Semantics of AN • No generally accepted syntax nor semantics for AN • Designer dependent & vary from one to other • Based on PL & FOPL with extensions • Syntax for any given system is determined by object & relation primitives chosen & by any special rules if any to connect nodes • Language of AN is formed from letters of alphabet (upper case & lower case), relation symbols, set membership, decimals, square & oval nodes, directed arcs #
- 9. Associative Networks... • “ISA” is predicate has been used to exhibit following type of structures • Generic – Generic Relationships (subset – superset, Generalization – Specialization, AKO) • Generic – Individual Relationships (set membership, Abstraction) • PREDICATE shows relations in form of arcs • AKO Subset • Member of Attributes • ISA Instance of #
- 10. Semantics of Associative Networks • If a class A of objects has some property P, and “a” is a member of A, then we can infer that “a has property P” • Associative Networks can use in parallel inference rules like modus ponens, chain rule & resolution #
- 11. Conceptual Graphs (CG) • Conceptual Graphs (CG) may become de-facto standard for Associative Networks (AN) in future • CG may be regarded as primitive building block of AN is a formalism of knowledge representation • “CG is a graphical representation of a mental perception which consists of basic or primitive concepts & relationships that exist b/w the concepts” • A single CG is equivalent to a graphical diagram of natural language sentence where words are depicted as concepts & relationships #
- 12. CG.... PERSON : joe agent eat object FOOD: soup INSTRUMENT SPOON #
- 13. CG... • Concepts refers to entities, actions, properties or events in the world. • A concept can be Generic or Individual. Individual concepts have type field followed by referent field • Generic concepts have no referent field • Concepts are enclosed in boxes and relations between concepts in ovals. • Direction of arrow depends to the order of the arguments in the relation they occur • Edges do not have labels • Standard concepts: AGENT, INSTRUMENT, # OBJECT, PART etc
- 14. Conversion from CG to FOPL 1. Assign unique variable names to every generic concepts. Ex: EAT, SPOON will be given x & y 2. Labels like PERSON, FOOD converted into unary predicates with some name 3. Standard Conceptual relations like AGENT, OBJECT, INSTRUMENT are converted into predicates with many attributes if required 4. Concept referents like joe, soup become FOPL constants 5. Generic concepts with no quantifier in the referent field have an existential quantifier placed before the formula for each variable #
- 15. Conversion from FOPL to CG • Put FOPL formula into prenex normal form & convert all logic connectives to negation & conjunction. • Every occurrence of Universal quantification is replaced by ~E~x • Every variable x & every occurrence of existential quantification is then replaced with most general type concept denote as [T: *x] • Implication in CG can be represented with negation & conjuction • Ex: P Q can be written as ~[P ~[Q]] #
- 16. Frame Structures • Introduced by Marvin Minsky in 1975 as a data structure to represent a mental model of situation like “driving a car”, “attending meeting”, “eating in hotel”. • Frames are general record like structures which consists of slots & slot values. Slots can of any size & type. • Slots which have names & values (or) subfields are called facets. • Facets may have names & number of values #
- 17. Syntax: Frame Structures (<frame name>) (<slot 1> (<facet 1><value 1> ......... <value k1>) (<facet 2><value 2> ......... <value k2>) ... ... (<facet n><value n> ......... <value kn>) ... ... (<slot 2> (<facet 1><value 1> ......... <value k1>) ... # ...
- 18. Example: Frame structures (ford (AKO (VALUE car) (COLOR (VALUE silver) (GAS-MILEAGE (DEFAULT fget) (RANGE (VALUED undefined) .... .... Note: Gas-Mileage(fget function) to fetch default value from another frame #
- 19. Thank You !!! Reference: Artificial Intelligence & Expert Systems by Dav W Patterson #