Chandrasekaran 1988: Generic Tasks as Building Blocks for Knowledge-Based Systems: The Diagnosis and Routine Design Examples

[ CogSci Summaries home | UP | email ]
http://www.jimdavies.org/summaries/

Chandrasekaran, B. (1988) Generic Tasks as Building Blocks for Knowledge-Based Systems: The Diagnosis and Routine Design Examples. Knowledge Engineering Review, 3 (3).

@InCollection{Chandrasekaran88o.89,
  author =       "B. Chandrasekaran",
  title =        "Generic Tasks as Building Blocks for Knowledge-Based
                 Systems: The Diagnosis and Routine Design Examples",
  booktitle =    "The Knowledge Engineering Review",
  publisher =    "?",
  pages =        "183--210",
  year =         "1988o.89",
}

Author of the summary: Jim Davies, 1999, jim@jimdavies.org

Cite this paper for:

SYSTEM: CSRL (conceptual structures Representation Language)

SYSTEM: HYPER (HYPothests matchER)

SYSTEM: IDABLE (Intelligent DAta Base Language)

SYSTEM: DSPL (Design Specialists and Plans Language)

SYSTEM: PEIRCE

SYSTEM: AIR-CYL

generic tasks

compiled knowledge systems (p7)

Knowledge abstraction helps efficiency.

Uniform representations mean a level of abstraction problem. (p26)

Knowledge based systems based on frames, logic, or rules are too primitive for diagonsis. This paper presents generic tasks, which are to frames and rules, ect., an high-level programming languages are to assembly.

Examples of generic tasks (GTs)

SYSTEM: CSRL (conceptual structures Representation Language)
creates a hierarchal classification
SYSTEM: HYPER (HYPothests matchER)
decides how well a hypothesis matches data
SYSTEM: IDABLE (Intelligent DAta Base Language)
finds attribute values by predicting based on examples.
SYSTEM: DSPL (Design Specialists and Plans Language)
designs objects
SYSTEM: PEIRCE
makes a composite hypothesis from hypotheses and data

The above building blocks can be used to make a general diagnosis system. Diagnosis is made up of many tasks that differ in input and output. Abstractly it is finding a set of causes for observations. Compiled knowledge systems include, at a minimum, a case memory and knowledge that helps map hypotheses to observations.

The proposed GT arch. classifies plausible hypotheses then takes that and makes a good composite. This helps computationally whenever the knowledge is in the right form: hierarchical, e.g.

There are 4 components to the architecture:

hierarchical classifier:
hierarchical matchers:
abductive assembly: finds the best explaning and most descriptive set of explanations
knowledge-directed data abstraction and inference

In the classification hierarchy, each node has a confidence. More confident nodes refine themselves by activating sub-nodes. (p10) Knowledge abstraction helps efficiency.

Design hierarchies can be component-subcomponent or function-subfunction based. Specialists in the design hierarchy are responsible for their part, according to constraints. The top node is responsible for the whole design. Each specialist has a collection of plans it can use. deep specialists have fewer, more straightforward plans (p22). Selectors choose plans; Sponsors give opinions.

Properties of the GT:

Multiformity: Uniform representations mean a level of abstraction problem. (p26)
Modularity
Each task has its own kind of knowledge aquisition.
Each GT integrates a particular way of representing knowledge with a particular way of using it.
Tractability

Summary author's notes:

page numbers are from a preprint version, I believe

Back to the Cognitive Science Summaries homepage
Cognitive Science Summaries Webmaster:

JimDavies (jim@jimdavies.org)

Last modified: Wed Mar 22 12:52:28 EST 2000