The CommonKADS Design ModelIntroduction to CommonKADSSlide 3Application DesignFunctional DecompositionObject-oriented DecompositionAI ParadigmArchitectural commandArchitectural DesignPlatform DesignWorked Example I: IMPRESSWorked Example II: X-MATEConclusionThe CommonKADS Design Model•Introduction to CommonKADS•CommonKADS Design Model•Worked Example•ConclusionIntroduction to CommonKADS•Problem faced by Knowledge Engineers --designing good KBS relies on knowledge engineer abilities•Approach to the problem--CommonKADS --represent the expert knowledge and design specification in form of text or diagrams•Important components of CommonKADS --Expertise Model: domain knowledge,inference knowledge, task knowledge --Design ModelThe CommonKADS Design Model•Intends to support knowledge engineers in choosing knowledge representations and programming techniques•Three-stage transformation process --Application design --Architectural design --Platform designApplication Design•Purpose is to decompose the knowledge into manageable “chunks”•Three approaches --Functional decomposition --Object-oriented decomposition --AI paradigmsFunctional Decomposition•Treat each inference step from the Expertise Model as a “chunk” of functionality•Benefits --replicate expert problem solving process --canonical inference has its expected functionality already defined --preserve task structureObject-oriented Decomposition•Treat each concept from the domain model as a “chunk” of data--an object class•Preserve the structure of the domain models in the expertise model•inference and task structure may be maintained •Have difficulty in assigning production rules because they refer to more than one objectsAI Paradigm•Include blackboard system, constraint-based programming, qualitative simulation, or model-based reasoning•“chunk”of knowledge may be constraints, knowledge sources, or whatever appropriate for the chosen approach•little of the structure of the expertise model will be maintainedArchitectural command•Characterize the contents of each “chunk”after decomposition to specify further design requirements•A function name that describe the action which the function performs --e.g. an inference step of type selec t -subset can be implemented by a subset command•Help validate the Expertise ModelArchitectural Design•Task is to define a computational infrastructure capable of implementing the architecture commands defined in the application design•Emphasis is on choosing ideal techniques•Experience of knowledge engineer can play an important rolePlatform Design•Consider how the ideal knowledge representation and inference techniques should be implemented in chosen software•Knowledge representation is not a problem•Some programming techniques may be awkward to implementWorked Example I: IMPRESS•Diagnose faults in machine•Application design: functional decomposition•Architectural design --knowledge representation: objects --inference step: production rules, object-based operations•Platform design: KAPPA-PC on Compaq 386•Flow of control: based on task structure from the Expertise ModelWorked Example II: X-MATE•Assess the ability of mortgage applicants•Application design: functional decomposition•Architecture design: select data source, use production rules, choose rule set, use arithmetic functions•Platform design: KAPPA-PC 1.1 on HP Vectra 386 PC•Flow of control: repeat running through the whole inference structureConclusion•CommonKADS provides useful documentation of system design process•Encourage greater modularity and reusability of designs•Help to validate the Expertise Model •Weakness --lack of guidance on selection of techniques --lack of a defined set of architectural