Download presentation
Presentation is loading. Please wait.
1
Knowledge Component 6: Knowledge Utilization
Systems for Decision Support 2nd Edition
2
Module Information Intended audience Key words Author
Novice Key words Decision support, knowledge, representation, reasoning, maintenance, user interfaces Author Ian Smith, EPFL, Switzerland Reviewer (1st Edition) Ni-Bin Chang, U of Central Florida 2
3
What there is to learn The quizzes at the end summarize important aspects. General ideas are: Knowledge systems may help engineers reason with knowledge Due to knowledge maintenance challenges, rules sets should be kept small and causal directions should be maintained. A review of the knowledge in the knowledge base is necessary at regular intervals
4
Outline Knowledge Systems Important Characteristics
Representation of Knowledge Reasoning with Knowledge Importance of User Interfaces Knowledge Maintenance
5
Introduction A Knowledge System (KS) is broadly defined as a computer-based system that aids the process of decision making. KSs are most useful in situations where traditional algorithms have exponential complexity. Knowledge Systems can also be of assistance when dealing with ill-structured tasks (Module 1.3).
6
Introduction (cont’d.)
KSs are used as stand alone systems and also in combination with other components such as object oriented programs, data bases (Module 4.1) and search algorithms (Module 5.2.3). Knowledge systems have evolved within the context of successes and failures with expert systems since the 1970s.
7
KS ≠ Engineer For important tasks, KSs cannot replace engineers. Their usefulness depends on the abbilities of engineers who use them. They can make good engineers perform better and vice versa (bad engineers may become worse!). The engineer bears the legal responsibility for the decisions. KSs are similar to books in that they can only guide the engineer.
8
Important Characteristics
Outline Knowledge Systems Important Characteristics Representation of Knowledge Reasoning with Knowledge Importance of User Interfaces Knowledge Maintenance
9
Characteristics of a KS
Provides support for tasks with many possible solutions Helps users identify good solutions. More generally, a good KS encourages user participation. Has an explicit representation of knowledge Provides greater separation between knowledge and control than with traditional approaches
10
Diagram Engineer developer Chunks of knowledge and data
Control Human-Computer Interface Algorithms Engineer User
11
Representation of Knowledge
Outline Knowledge Systems Important Characteristics Representation of Knowledge Reasoning with Knowledge Importance of User Interfaces Knowledge Maintenance
12
What is Knowledge? There are many definitions of ‘knowledge’. Some have broad coverage with vague sociological definitions while others are very restrictive. For the purposes of this course, we will use a rather restrictive definition that is sufficiently general from a computer science perspective. Note that dictionary definitions may not be completely compatible with our definitions.
13
Information Knowledge is part of information. Information is a “class” that includes data and knowledge. Data are facts about something. For example, this car is red; that building has five stories; the temperature is 20°C; his name is Paul, etc. These are attribute-value pairs. In relational databases, attributes are columns and values are the entries (numbers, symbols) in tables (Module 4.1).
14
Knowledge Knowledge is created by expressing relationships between data. Following are examples of possible links. classification (this car is a kind of vehicle) compositional (this building is part of a school) empirical (the exchange rate is CHF 1.50/$) physical principles (F=m∙a)
15
Knowledge (cont'd.) Knowledge can also be
semantic (Paul is the name of a boy) heuristic (very long bridges are suspension bridges) causal (if the temperature is 100°C then water boils)
16
Causal Knowledge Causal knowledge, or rules, are important elements of engineering reasoning. Rules may represent links such as physical principles, empirical knowledge and heuristics. Causal knowledge is also easy to implement in computers because most computer languages have if … then … statements.
17
Control Knowledge Knowledge is also concerned with linking procedures, rather than data, together. This is sometimes called control knowledge. An example of control knowledge is a set of steps to study for an exam. The topic of control knowledge is out of the scope of this course.
18
Knowledge Engineering
Knowledge Engineering (KE) refers to the building, maintaining and development of knowledge systems. In civil engineering, much knowledge is already organized in codes, standards and other conventions. Also, experience has shown that successful knowledge systems are built and maintained exclusively by people who are experts in their civil engineering sub-field. Therefore, this course does not cover aspects related to knowledge elicitation and also, strategies related to knowledge representation are only summarized.
19
Explicit Representation
In computer science, the some fields of artificial intelligence deal with explicit representation and use of knowledge. Many researchers believe that any intelligence must include knowledge as defined here. This requirement is currently a topic of debate.
20
Representing Knowledge
How is knowledge represented in knowledge systems? The most common frameworks for representing knowledge are Data bases – relations and functional dependencies (Module 4.1) Object hierarchies and other types of semantic networks
21
Representing Knowledge (cont'd.)
Rules (if...then... statements) Models – models can be mathematical formulas, numerical representations (such as finite-element models) and simulations. Models are used to transform causes into effects through explicit use of geometry, physical principles, empirical relationships and other relationships between data.
22
Representing Knowledge (cont'd.)
Traditional knowledge systems (1970s) employed data and rules. This approach is still useful for small systems in well-defined domains where knowledge seldom changes. Since implementation of rules often requires a transformation of knowledge into if… then… statements, important information may be lost. Also, unwanted dependencies may be added (See Module 4.1).
23
Representing Knowledge (cont'd.)
Current systems use data, objects, rules, models and cases. The number of rules is kept to a minimum. Models are often the most natural and robust way to represent knowledge. They are usually easier to modify and control than rules.
24
Reasoning with Knowledge
Outline Knowledge Systems Important Characteristics Representation of Knowledge Reasoning with Knowledge Importance of User Interfaces Knowledge Maintenance
25
Reasoning with Knowledge
In this section, we limit our discussion to reasoning with rules. More sophisticated methods exist to support reasoning with other representations of knowledge such as models and semantic networks. However, this is out of the scope of this course.
26
Reasoning with Rules There are many ways to use rules to support engineering reasoning. The two principal methods are Forward chaining Backward chaining Bi-directional chaining is employed to have rules operate in both directions. This is not in the scope of this course.
27
Forward Chaining Forward chaining uses the data that are currently available to satisfy the conditions of the rules (if…). If conditions are satisfied, the rule is fired. This means that consequences (then…) are added to the currently available data. This new augmented data set is then compared with other conditions in other rules to see if they can further augment the data set. Chaining stops when no more rules can be fired.
28
Backward Chaining Backward chaining requires an additional element. This is a goal or a question regarding a fact. For example, a question may be whether a data value is “true”. In backward chaining, a rule is identified that has as its consequence the affirmation of the question. The conditions of this rule are examined to see if they satisfy the data set. If they don’t, the conditions of this rule are taken as sub-goals for subsequent iterations. Chaining stops when there are no more rules or when a pre-defined stopping condition is reached.
29
Forward or Backward Chaining?
Both strategies often provide useful results for the same task. From the viewpoint of computational efficiency, forward chaining is best when rule antecedents (the “if” part) are simpler than the rule consequents (the “then” part). This is typically the case for control tasks, process monitoring and some configuration activities. When rule antecedents are longer than the rule consequents, backward chaining is best. This occurs for tasks such as diagnosis, evaluation and certification.
30
Exercise Task You are a building inspector for a large insurance company. You are responsible for evaluating their buildings through on-site inspections. This insurance company has hundreds of buildings scattered around the country. You cannot visit each one on a regular basis. A knowledge system has been created to help you decide which buildings require on-site inspections. Apply forward and backward chaining to the following rules.
31
Exercise (cont'd.) Rules
R1: If water damage is possible and if the building has not been renovated for 30 years Then on-site inspection is necessary R2: If the plumbing system is suspect Then water damage is possible R3: If the building is more than 40 years old Then the plumbing system is suspect
32
Exercise (cont'd.) Data Building ‘D450’ is more than 40 years old and it has not been renovated for 30 years. Question Is an on-site inspection necessary? Fill in the following tables for this exercise.
33
Exercise (cont'd.) Forward chaining Rules examined Rules fired
Data used Building is more than 40 years old Building has not been renovated for 30 years
34
Exercise (cont'd.) Backward chaining Rules examined Rules fired
New goal
35
Exercise (cont'd.) Forward chaining: Solution Rules examined
Rules fired Data used R1, R2, R3 R3 Building is more than 40 years old R1, R2 R2 The plumbing system is suspect R1 Water damage is possible Building has not been renovated for 30 years On-site inspection is necessary
36
Exercise (cont'd.) Backward chaining: Solution Rules examined
Rules fired New Goal R1 Water damage is possible R2 The plumbing system is suspect R3
37
Number of combinations
Discussion The number of combinations of buildings to visit in a given time period is exponential in terms of the total number of buildings. Therefore, this is a good task for the application of knowledge systems for decision support. Number of combinations Total number of buildings
38
Discussion (cont'd.) The information in the data set has to be compatible with the conditions and consequences of the rules. For example if the building has never been renovated, this information has to be transformed a priori into the fact that the building has not been renovated for 30 years. When such transformations are performed, there may be a loss of information.
39
Discussion (cont'd.) Addition of rules may require modifications to existing rules. For example, a rule R4 is added. R4: If the building was built by XYZ contractor Then the plumbing is ok With this new rule R4, rule R3 should be modified to add the condition “not built by XYZ contractor” in order to avoid inconsistencies. Thus, the addition of a rule created an unwanted dependency that lead to modifications elsewhere.
40
Review Quiz - 1 What are the main characteristics of KSs?
What is the difference between forward chaining and back ward chaining? List the ideal applications.
41
Answers to Review Quiz - 1
What are the main characteristics of KSs? Provides support for tasks with many possible solutions Helps users identify good solutions. More generally, a good KS encourages user participation. Has an explicit representation of knowledge Provides greater separation between knowledge and control than with traditional approaches
42
Answers to Review Quiz - 1
What is the difference between forward chaining and back ward chaining? Forward chaining approach starts with some facts and applies rules to find all possible conclusions, whereas backward chaining starts with the desired conclusions and works backwards to find supporting facts. These approaches can be viewed as two variations on search.
43
Answers to Review Quiz - 1
List the ideal applications Forward chaining are better for control tasks, process monitoring and some configuration activities. Backward chaining systems are better for diagnosis, evaluation and certification tasks.
44
Importance of User Interfaces
Outline Knowledge Systems Important Characteristics Representation of Knowledge Reasoning with Knowledge Importance of User Interfaces Knowledge Maintenance
45
How should the multiple solutions be presented to the user?
User Interfaces Many solutions are possible for most engineering tasks and often, there is no unique best solution, especially when it is known that information is incomplete. This is where user interfaces are useful to help the user choose the appropriate solution. A well-conceived user interface (UI, User Interface) is essential for engineering applications. How should the multiple solutions be presented to the user? It depends on the task, the user and the context of the task.
46
Well-conceived UI Following are example guidelines for a well conceived user interface (UI). Introduce information that is specific to the task Present multiple solutions to the user in an understandable fashion Allow exploration of solutions within the space of good solutions Provide for addition of knowledge
47
Well-conceived UI (cont'd.)
Explain reasoning so that the user is better able to understand why certain solutions have been proposed and why others have been eliminated. Proactively “cue” users to make good choices Indicate inconsistent information Evaluate different combinations of choices Teach users how to use the system
48
Well-conceived UI (cont'd.)
Different combinations of task characteristics, user needs and contexts result in a wide range of appropriate user interfaces. Formal models of engineer-computer interaction are emerging. However, in many engineering areas, developers are currently crafting solutions without explicit development frameworks.
49
Evaluating Usability of a UI
The quality of a UI may be measured according to many factors. For example, Ease of learning Efficiency of use Error frequency, severity, and recovery Ability to propose multiple solutions Subjective satisfaction The importance of each factor depends on the goal of the application.
50
Knowledge Maintenance
Outline Decision Support Systems Important Characteristics Representation of Knowledge Reasoning with Knowledge Importance of User Interfaces Knowledge Maintenance
51
Maintenance Difficulties associated with knowledge maintenance contributed to the demise of early expert systems. As rule sets grew, the necessity to check different combinations of fired rules grew exponentially. The maintenance costs also grew in the same manner.
52
Difficulties It has already been discussed that rule sets may not be able to include all information. Moreover, they may create unwanted dependencies that complicate knowledge maintenance. While such difficulties cannot be avoided altogether, other representations of knowledge require less maintenance.
53
Causal Inversion Much of the maintenance difficulties that were associated with early knowledge systems arose out of a lack of understanding of task analysis and fundamental logic (Module 1.3). Often, abductive tasks (for example, design and diagnosis) were supported by rules that did not maintain the correct causal direction. The correct rules were transformed using causal inversion.
54
Causal Inversion (cont'd.)
More specifically, the causal direction If cause Then effect was inverted to If effect Then cause
55
Causal Inversion (cont'd.)
Developers often did not realize that this inversion hid the importance of the closed world hypothesis for abductive tasks (all rules and data are known). Although rule chaining implements deduction, it is deduction only when rules are expressed correctly.
56
Causal Inversion (cont'd.)
Stated otherwise, rules were formulated for abductive tasks so that deductive mechanisms (chaining) could process them. Such formulation makes an implicit closed world assumption. Since any modification or addition to the rule set means that all rules are not known, difficulties with knowledge maintenance is an obvious consequence.
57
Modern Approaches Modern approaches to knowledge representation involve formulations that maintain the correct causal direction. In this way, when abductive tasks are supported, the close world assumption is explicit. In general, the number of rules are kept to a minimum. This facilitates knowledge maintenance as fewer unwanted dependencies are created.
58
Examples Examples of correct formulations are:
if environment then behavior if loads then deformations if deterioration then less strength
59
Review Quiz - 2 How can the usability of a user interface be evaluated? What approaches can help minimize difficulties in knowledge maintenance?
60
Answers to Review Quiz - 2
How can the usability of a user interface be evaluated? The quality of a UI may be measured according to many factors. For example, Ease of learning Efficiency of use Error frequency, severity, and recovery Ability to propose multiple solutions Subjective satisfaction The importance of each factor depends on the goal of the application.
61
Answers to Review Quiz - 2
What approaches can help minimize difficulties in knowledge maintenance? Difficulties in knowledge maintenance arise due to causal inversion and because of unwanted dependencies being created. Maintaining the correct causal direction helps keep the number of rules to a minimum and avoids the risk of unwanted dependencies.
62
Further Reading M. Stefik. Introduction to Knowledge Systems, Morgan Kaufman Publishers, San Fransisco, California, 1995 M. Ginsberg. Essentials of Artificial Intelligence, Morgan Kaufman Publishers, San Fransisco, California, 1993 Raphael, B. and Smith, I.F.C. Fundamentals of Computer-Aided Engineering, Wiley, 2003
Similar presentations
© 2025 SlidePlayer.com. Inc.
All rights reserved.