Download File - Md. Mahbubul Alam, PhD

Lecture 6:
Knowledge Application Systems
Md. Mahbubul Alam, PhD
Associate Professor
Dept. of AEIS
Sher-e-Bangla Agricultural University
Chapter Highlights
• Describe knowledge application mechanisms
• Explain knowledge application technologies,
– expert systems
– decision support
– advisor systems
– fault diagnosis (or troubleshooting) systems
– help desk systems.
Knowledge Application Systems
• Artificial Intelligence
• Computers to perform tasks that resemble human thinking ability.
 Systems that act like humans
 Systems that think like humans
 a computer program whose input to output behavior matches those of humans.
 Systems that think rationally
 Systems that follow a specific logic to solve a problem.
 Systems that act rationally
 Systems that are expected to have specific characteristics that enable them to operate
autonomously within their environments, and even adapt to change in he face of
uncertainty.
Computer Vs. Human Vs. AI
• Computer
– Perform repetitive and logical tasks extremely well, e.g., complex arithmetic calculations
and database retrieval & storage.
– Computer tasks are mainly algorithmic in nature.
– Operated by precise and logically designed instructions.
• Human
– Solve problems using symbols (i.e., a specific meaning)
– e.g., understanding the meaning of a poem.
• AI
–
–
–
–
Intelligence and knowledge are tightly intertwined.
Knowledge is associated with cognitive symbols.
Intelligence refers to our ability to learn and communicate in order to solve problems.
“The science that provides computers with the ability to represent and manipulate
symbols so they can be used to solve problems not easily solved through algorithmic
models.”
Technologies for Knowledge Application Systems
Technology
Domain Characteristics
Rule-based
systems
Applicable when the domain knowledge can be defined by a
manageable set of rules or heuristics.
Case-based
reasoning
Applicable in weak-theory domains, that is, where an expert
either doesn’t exist or does not fully understand the domain. Also
applicable if the experience base spans an entire organization, rather
than a single individual.
ConstraintApplicable in domains that are defined by constraints, or what cannot
based reasoning be done.
Model-based
reasoning
Applicable when designing a system based on the description of the
internal workings of an engineered system. This knowledge is typically
available from design specifications, drawings, and books, and can be
used to recognize and diagnose its abnormal operation.
Diagrammatic
reasoning
Applicable when the domain is best represented by diagrams and
imagery, such as when solving geometric problems.
Rule-based Systems
• Use rules or models to represent the domain knowledge.
• Requirements:
– Subject-matter expert, generating or contributing knowledge.
– Knowledge engineer, responsible for the elicitation & representation of the expert’s
knowledge.
• Process: (Knowledge Engineering)
– Elicit the knowledge from the expert
– Represent it in a form that is usable by computers.
• Rules:
– IF-THEN statements.
– IF  condition (truth-value), THEN  fact (action, conclusion or consequence)
• Disadvantages
– The number of rules may be quite large.
– Difficulty in coding, verifying, validating, & maintaining the rules.
– Poor efficiency of the inference engine executing the rules.
Case-based Reasoning Systems (CBR)
• An artificial intelligence technique designed to mimic human problem
solving.
• Based on Schank’s (1982) model of dynamic memory.
• Goal: Mimic the way humans solve problems.
• When faced new problem, humans search their memories for past
problems resembling the current problem and adapt the prior solution to
“fit” the current problem.
• CBR is a method of analogical reasoning that utilizes old cases or experiences
in an effort to solve problems or interpret solutions.
CBR: Processes
1. Search the case library for similar cases
–
This implies utilizing a search engine that examines only the appropriate cases and not the entire case
library, as it may be quite large.
2. Select and retrieve the most similar case(s)
–
New problems are solved by first retrieving previously experienced cases. This implies having a means
to compare each examined case to the current problem, quantifying their similarity, and somehow
ranking them in decreasing order of similarity.
3. Adapt the solution for the most similar case
–
If the current problem and the most similar case are not similar enough, then the solution may have to be
adapted to fit the needs of the current problem. The new problem will be solved with the aid of an old
solution that has been adapted to the new problem.
4. Apply the generated solution and obtain feedback
–
5.
Once a solution or classification is generated by the system, it must be applied to the problem. Its effect
on the problem is fed back to the CBR system for classification of its solution (as success or failure).
Add the newly solved problem to the case library
–
The new experience is likely to be useful in future problem solving. This step requires identifying if the
new case is worth adding to the library and placing it in the appropriate location in the case library.
CBR: Classification
1. Exemplar-based reasoning
– These systems seek to solve problems through classification, that is, finding the right class
for the unclassified exemplar.
2. Instance-based reasoning
– These systems require a large number of instances (or cases) that are typically simple;
that is, they’re defined by a small set of attribute vectors. The major focus of study of these
systems is automated learning, requiring no user involvement (Aha et a1. 1991).
3. Analogy-based reasoning
– These systems are typically used to solve new problems based on past cases from a
different domain.
– Analogy-based reasoning focuses on case reuse, also called the mapping problem, which is
finding a way to map the solution of the analogue case to the present problem.
Constraint-based Reasoning
• An AI technique that uses essentially “what cannot be done”.
• It reflects what constraints restrict possible solutions.
• This technique is useful in naturally constrained tasks such as
planning and scheduling.
• e.g., To schedule a meeting all the individuals that need to
attend must be available at the same time, otherwise the
“availability constraint” will be violated.
Model-based reasoning
• Knowledge about the internal workings of a target system.
• It can be used to recognize and diagnose its abnormal
operation (discrepancies).
• Incorporates generic troubleshooting procedures common to
diagnosing many types of systems.
• Can help diagnose faults not previously experienced.
Diagrammatic Reasoning
• Understanding of concepts and ideas through the use of
diagrams and imagery, versus linguistic or algebraic
representations.
• Instrumental in developing systems such as Gelernter’s
Geometry Machine.
• These technologies are radically different from rule-based or
CBR systems and have very specific application areas.
Developing KAS: The Case-Method Cycle
1.
2.
3.
4.
5.
6.
System development process
–
To develop a knowledge application system that will store new cases and retrieve
relevant cases.
Case library development process
–
To develop and maintain a large-scale library that will adequately support the domain in
question.
System operation process
–
To define the installation, deployment and use support of the knowledge application
system. This process in based on standard software engineering and relational database
management procedures.
Database mining process
–
Uses rule-inferencing technique and statistical analysis to analyze the case library.
Management process
–
Describes how the project task force will be formed and what organizational support
will be provided.
Knowledge transfer process,
–
Describes the incentive systems to encourage user acceptance and support.
Sub-processes of developing case library
1.
Case collection
– Collection of seed cases.
– Provides an initial view of the application.
2.
Attribute-Value Extraction and Hierarchy Formation
– Process of indexing and organizing the case library.
– Goal is to extract the attributes that define the case representation and indexing.
3.
Feedback
– Provide necessary feedback to those supplying the cases to the CBR system, so that the
quality of the cases can be improved.
• CASE Method in CBR
– Significant reduction in system development workload and costs
• Knowledge Application Systems
– Apply a solution to a similar problem.
– Serve as a framework for creative reasoning.
Types of KAS
• Expert systems
• Decision support
• Advisor systems
• Fault diagnosis (or troubleshooting) systems
• Help desk systems
KAS: Limitation
• Typically developed to serve a task-specific domain problem, and not
integrated with the organization’s enterprise systems.
• Security: cases may include sensitive information.
• Scalability: must represent a large enough number of cases.
• Speed: as the size of the case library grows to a more comprehensive
representation of real environments, computing and searching costs will
also increase.
• May not be able to solve all the problems that come across, in particular,
increasingly complex environments