Documenti di Didattica
Documenti di Professioni
Documenti di Cultura
National Safeware Engineering Network of Centres of Innovative Academia-Industry Handshaking Project TEMPUS-SAFEGUARD (Project 158886-TEMPUS-1-2009-1-UK-TEMPUS-JPCR)
Problem description
International standards (ISO/IEC 9126, ISO/IEC 15504, ISO/IEC 15939, ISO/IEC 25000) not good for AI software
Problem with human experts:
Person effect Personalization effect
Intelligence SW Lifecycle
Identification Experimental exploitation
Conceptual modeling
Expert system
Testing
Formalization
Prototyping
Identification stage
Identify main goal of AI software nonformal task description Identify team members and their roles
Identify available resources knowledge sources, time resources, machine resources, money
2012 Sergii Parfenov 5
Formalization stage
Get expert`s knowledge
Knowledge systematization Knowledge representation Result knowledge base.
2012 Sergii Parfenov 7
Prototyping stage
Identification
Experimental exploitation
Conceptual modeling
Expert system
Testing
Formalization
Prototyping
Identification
Experimental exploitation
Conceptual modeling
Expert system
Testing
Formalization
Prototyping
10
Identification
Experimental exploitation
Conceptual modeling
Expert system
Testing
Formalization
Prototyping
11
Requirements Competency
Desired Minimum
2012 Sergii Parfenov
Service
Desired
Minimum
14
Identification
Competency requirements
Expert system
Experimental exploitation
Conceptual modeling
Testing
Formalization
Prototyping
15
Identification
Experimental exploitation
Conceptual modeling
Testing
Formalization
Prototyping
16
Problems of consistency
Redundancy: two rules have the same antecedent, and the conclusions of one subsume those of the other (e.g., x -> y and x -> y ^ z). Conflict: two rules have the same antecedent, but their conclusions are contradictory (e.g., x -> y and x -> !y).
Subsumption: two rules have similar conclusions, but the antecedent of one subsumes that of the other (e.g., x -> y and x ^ z -> y).
Unnecessary IF rules: two rules have the same conclusions, and their antecedents contain contradictory clauses, but are otherwise the same (e.g., x ^ z -> y and x ^ !z -> y). Circularity: a set of rules forms a cycle.
17
Problems of completeness
Unreferenced attribute values: some values in the set of possible values for an object's attribute are not covered in the set of rules. Illegal attribute values: a rule refers to an attribute value that is not in the set of legal values. Unreachable conclusions: the conclusion of a rule should either match a goal or an if condition in some other rule. Dead-end goals and dead-end IF conditions: either the attributes of a goal must be askable (the system can request information form the user) or the goal must match the conclusion of a rule. Similar considerations apply to the IF conditions in each rule.
18
Thank you!!!
Sergii Parfenov sergiiparfenov@gmail.com Khmelnitsky National University System Programming Department, http://spr.khnu.km.ua
2012 Sergii Parfenov 19