ProblemSolving Methods in Perspective

1
Problem-Solving Methodsin Perspective

• Richard Benjamins
• University of Amsterdam

2
In perspective?

• Knowledge components
• PSMs and Ontologies
• A standard for characterizing PSMs?
• PSMs in Cyberspace
• From ideas to practice a community effort

3
IBROW

• An Intelligent Brokering Service for
• Knowledge-Component Reuse on the World-Wide Web

Richard Benjamins
4
Objectives of IBROW

• Develop an intelligent brokering service that
  enables third party knowledge-component reuse
  through the Web
• Configurable reasoning services on the web
• Plug Play of PSMs and Ontologies

5
The agents involved
6
Innovative aspects

• Most web services broker static information
• Metacrawler, Searchbroker, Ontobroker
• IBROW brokers dynamic (reasoning) knowledge
• Opens possibility for a new electronic market
  place

7
Motivation

• Topdown
• make KE technology on reuse more widely
  accessible at lower cost
• Bottom up
• several PSM libraries exist, but not accessible
  nor interoperable
• WWW is here to stay

8
IBROW project

• ESPRIT, Open Long Term Research
• basic research
• industrial potential
• unknown risk
• Partners
• University of Amsterdam, SWI (nl)
• University of Karlsruhe, AIFB (de)
• Open University, KMI (uk)
• Artificial Intelligence Research Institute (es)

9
Approach

• Standard Product Description Language
• Libraries of reusable PSMs and Ontologies
• Brokering problem-solving knowledge
• Interoperability
• User Interface (browser)

10
Overview

• Method description language (UPML)
• Brokering problem-solving knowledge
• Interoperability of heterogeneous components
• Conclusions

11
What is UPML?

• Starting point CML of CommonKADS
• Adds
• component-based software development
• component reuse
• Machine processable
• enables semi-automatic system development
• needed for broker

12
The Backbone of UPML
13
Ontology

• Explicit specification of a conceptualization
• can be shared by multiple reasoning components
• Provide definition of
• signatures
• axioms
• used by other parts of the architecture

14
Task

• Specifies problem to be solved
• by knowledge-based system
• Problem definition is domain independent
• enables reuse of generic problem definitions for
  different applications.
• contrary to most approaches in software
  engineering

15
Domain Models

• Domain knowledge required by
• problem-solving method
• task definition
• Three elements characterization of
• properties
• assumptions
• domain knowledge

16
Problem-Solving Methods

• Describe reasoning steps and types of knowledge
  needed to perform a task.
• Two different types of PSMs
• complex problem-solving methods
• primitive problem-solving methods

17
Problem-Solving Methods

• Complex problem-solving methods
• decompose a task into subtasks
• Primitive problem-solving methods
• make assumptions about domain knowledge to
  perform a reasoning step
• do not have an internal structure

18
Problem-solving method
19
Bridges, Refiners
20
Bridges

• Model relationships between two different parts
  of an architecture
• between domain and task
• task and problem-solving method

21
Refiners

• Used to express the stepwise adaptation of one
  type of element of a specification
• a task is refined into a more specific task
  (design --gt parametric design)
• a problem-solving method is refined into a more
  specific PSM (search --gt hill climbing)

22
Design Rules

• Architectural constraints ensure well formed
  specifications
• The individual components must fulfill certain
  properties (assumptions operational-spec --gt
  competence of PSM)
• Their connection via bridges and refiners must
  fulfill certain properties

23
Editor for UPML

• PROTEGE-II provides an environment for generating
  knowledge acquisition tools
• Use of PROTEGE-II for
• defining UPML meta ontology
• designing editor for UPML

24
Meta ontology of UPML
25
Editor for UPML
26
Browser for UPML

• On2broker for
• browsing UPML specifications
• query interface for UPML specifications
• Query Interface is a JavaTM Remote Method
  Invocation (RMI) Server which can be consulted by
  any software agent

27
The IBROW broker

• Acquire customers problem description
• Configuration
• Find candidate problem-solving methods
• Check their applicability wrt the KB
• Integrate the PSMs into a coherent reasoner
• Execution
• CORBA
• Web standard

28
Two problems
29
Characterizing PSMs

• UPML provides slots for
• competence of PSMs
• goal of tasks
• pragmatics (non functional)
• ease of use
• rate of success
• successfully used in
• assumptions of PSMs, tasks and domains

30
Finding PSMs

• Libraries have to register at the broker
• Global match in case of many PSMs
• pragmatics, keywords
• For promising PSMs, match competence with task
  goal (theorem prover)
• might imply mapping and renaming

31
Checking applicability

• Check assumptions of the PSM in the KB of the
  customer
• theorem prover
• If different ontologies, derive bridges to
  connect PSM and KB

32
Locating PSMs
33
Integrating PSMs

• Impose control knowledge on the selected PSMs
• chain inputs and outputs
• consider UPMLs operational descriptions
• exploit existing task structures

34
Execution

• CORBA
• the configured problem-solver is a collection of
  clients
• each PSM is a server
• the KB is a server
• Lisp Server (using Web standards)
• client can also query Lisp Server
• Broker controls which client sends requests

35
Overall picture
36
Small demo
37
Summary of Achievements

• Distributed knowledge-intensive applications
• Competence-oriented component libraries with
  explicit models based on a uniform framework
• (Semi-) automated construction of bridges between
  components and refiners
• Interoperability through CORBA and WWW standards
• Powerful brokering facilities

38
Open Issues

• Formalization (pragmatics)
• CORBA wrapping of (legacy) KBs
• Bridges and refiners (theory)
• Architectures