Knowledge-based

systems

Rozália Lakner
University of Veszprém
Department of Computer Science
An overview
 Knowledge-based systems, expert systems
 structure, characteristics
 main components
 advantages, disadvantages
 Base techniques of knowledge-based systems
 rule-based techniques
 inductive techniques
 hybrid techniques
 symbol-manipulation techniques
 case-based techniques
 (qualitative techniques, model-based techniques, temporal reasoning
techniques, neural networks)

Engineering Application of AI - PhD Course - 2

Knowledge-based systems

Engineering Application of AI - PhD Course - 3

Structure and characteristics 1
 KBSs are computer systems
 contain stored knowledge
 solve problems like humans would
 KBSs are AI programs with program structure of new type
 knowledge-base (rules, facts, meta-knowledge)
 inference engine (reasoning and search strategy for solution, other
services)
 characteristics of KBSs:
 intelligent information processing systems
 representation of domain of interest  symbolic representation
 problem solving  by symbol-manipulation
  symbolic programs

Engineering Application of AI - PhD Course - 4

Structure and characteristics 2

Explanation Case specific

subsystem database

User
User
interface

Inference
engine

Knowledge
base
Knowledge
Knowledge Developer's
acquisition
engineer interface
subsystem

Engineering Application of AI - PhD Course - 5

 Main components 1
 knowledge-base (KB)
 knowledge about the field of interest (in natural language-like formalism)
 symbolically described system-specification
 KNOWLEDGE-REPRESENTATION METHOD!
 inference engine
 „engine” of problem solving (general problem solving knowledge)
 supporting the operation of the other components
 PROBLEM SOLVING METHOD!
 case-specific database
 auxiliary component
 specific information (information from outside, initial data of the concrete problem)
 information obtained during reasoning

Engineering Application of AI - PhD Course - 6

Main components 2
 explanation subsystem
explanation of system’ actions in case of user’ request
typical explanation facilities:
 explanation during problem solving:
 WHY... (explanative reasoning, intelligent help, tracing information about the actual
reasoning steps)
 WHAT IF... (hypothetical reasoning, conditional assignment and its consequences,
can be withdrawn)
 WHAT IS ... (gleaning in knowledge-base and case-specific database)
 explanation after problem solving:
 HOW ... (explanative reasoning, information about the way the result has been
found)
 WHY NOT ... (explanative reasoning, finding counter-examples)
 WHAT IS ... (gleaning in knowledge-base and case-specific database)

Engineering Application of AI - PhD Course - 7

Main components 3
 knowledge acquisition subsystem
 main tasks:
 checking the syntax of knowledge elements
 checking the consistency of KB (verification, validation)
 knowledge extraction, building KB
 automatic logging and book-keeping of the changes of KB
 tracing facilities (handling breakpoints, automatic monitoring and reporting the
values of knowledge elements)
 user interface ( user)
 dialogue on natural language (consultation/ suggestion)
 specially intefaces
 database and other connections
 developer interface ( knowledge engineer, human expert)

Engineering Application of AI - PhD Course - 8

Main components 4
 the main tasks of the knowledge engineer:
 knowledge acquisition and design of KBS: determination,
classification, refinement and formalization of methods, thumb-rules
and procedures
 selection of knowledge representation method and reasoning
strategy
 implementation of knowledge-based system
 verification and validation of KB
 KB maintenance

Engineering Application of AI - PhD Course - 9

Expert Systems

Engineering Application of AI - PhD Course - 10

Structure and characteristics 1
 expert systems  knowledge-based systems
 employ expert’ knowledge
 applied in a narrow specific field
 solve difficult problems (must be demand on special knowledge)
 specialized human experts are needed
 experts must be agreed on the fundamental questions of professional field
 learning examples and raw data are needed
 expectations from an ES (like a human expert):
 make intelligent decision: offer intelligent advice and explanations
 question/ answer (“treated as an equal conversation partner”)
 explanation of questions
 acceptable advice even in case of uncertain situation

Engineering Application of AI - PhD Course - 11

Structure and characteristics 2

AI programs

Knowledge-based systems

Expert systems

 AI programs:
intelligent problem solving tools
 KBSs
AI programs with special program structure
separated knowledge base
 ESs
KBSs applied in a specific narrow field
Engineering Application of AI - PhD Course - 12
Expert system shells 1
 „empty” ESs, contain all the active elements of an ES
 empty KB, powerful knowledge acquicition subsystem
 contain services for construction and operation of ES
independently of the field of interest
 support the development of rapid prototype and the
incremental construction
 examples: CLIPS, GoldWorks, G2, Level5

Engineering Application of AI - PhD Course - 13

Expert system shells 2

Explanation Case specific

subsystem database

User
User
interface

Inference
engine

Knowledge
base
Knowledge
Knowledge Developer's
acquisition
engineer interface
subsystem

Engineering Application of AI - PhD Course - 14

Advantages of KBSs and ESs
 make up for shortage of experts, spread expert’ knowledge
on available price (TROPICAID)
 field of interest’ changes are well-tracked (R1)
 increase expert’ ability and efficiency
 preserve know-how
 can be developed systems unrealizabled with tradicional
technology (Buck Rogers)
 self-consistents in advising, equable in performance
 are available permanently
 able to work even with partial, non-complete data
 able to give expanation

Engineering Application of AI - PhD Course - 15

Disadvantages of KBSs and ESs
 their knowledge is from a narrow field, don’t know the
limits
 the answers are not always correct (advices have to be
analysed!)
 don’t have common sence (greatest restriction)  all of
the self-evident checking have to be defined
(many exceptions  increase the size of KB and the
running time)

Engineering Application of AI - PhD Course - 16

Base techniques of KBSs

Engineering Application of AI - PhD Course - 17

Techniques of KBSs
based on the knowledge-representation methods and
reasoning strategies applied in the implementation
 rule-based techniques

 inductive techniques

 hybrid techniques

 symbol-manipulation techniques

 case-based techniques

 (qualitative techniques, model-based techniques, temporal

reasoning techniques, neural networks)

Engineering Application of AI - PhD Course - 18

Rule-based techniques
(a short review)

Engineering Application of AI - PhD Course - 19

Reasoning with rules 1
 knowledge-representation form: rule
 rule-base can be according to the structure of KB
 simple/unstructured
 structured (contexts)
 reasoning strategies:
 according to the control direction
 data-driven/forward chaining
 goal-driven/backward chaining

Engineering Application of AI - PhD Course - 20

Reasoning with rules 2
 aim: proving a goal statement or achieving a goal state
 the reasoning algorithm:
 pattern matching
 finding applicable rules (watching condition/conclusion part of rules)
 fireable rules  conflict set (match condition/conclusion part of rules)
 conflict resolution
 selecting the most appropriate rule from conflict set
 conflict resolution strategies
 firing
 executing the selected rule  new knowledge (new facts or new subgoals to
be proved)
 watching termination conditions
 restart of the cycle

Engineering Application of AI - PhD Course - 21

Inductive techniques

Engineering Application of AI - PhD Course - 22

Inductive reasoning
 a type of machine learning technics
 inferring from individual cases to general information
 given a collection of training examples (x, f(x))
 return a function h that approximates f
 h is called hypothese

f(x)

x h y p o t h e s e s

 aim: finding the hypothese fits well on the training examples

 h is used for prediction the values of the unseen examples

Engineering Application of AI - PhD Course - 23

Decision tree 1
 one of the most known methods of inductive learning:
learning decision trees
 decision tree: simple representation for classifying
examples
 elements of the decision tree:
 nonleaf (internal) nodes are labelled with attributes (A)
 arcs out of a node are labelled with possible attribute values of A
 leaf nodes are labelled with classifications (Boolean values –
yes/no - in the simplest case)

Engineering Application of AI - PhD Course - 24

Decision tree 2

Country Age Engine Colour Easy to sell

We want to classify new examples on
1. Germany 3-6 diesel white yes property Easy to sell based on the
2. Japan 6-10 diesel red yes examples’ Country, Age, Engine and
3. Japan 3-6 diesel blue no Colour.

Country

Germany Japan

yes Colour

red blue

yes no

Engineering Application of AI - PhD Course - 25

Decision tree 3
 a decision tree under construction contains:
 nodes labelled with attributes
 nodes labelled with classifications (yes/no values)
 unlabelled nodes
 arcs labelled with attribute values outlet only form nodes labelled
with attributes
 every unlabelled nodes possess:
 a subset of training examples
 eligible attributes

Engineering Application of AI - PhD Course - 26

Decision tree 4
 some questions about decision tree:
 Given some data (set of training examples and attributes), which
decision tree should be generated?
 A decision tree can represent any discrete function of the inputs.
Which trees are the best predictors of unseen data?
 You need a bias (preference for one hypothesis over another).
Example, prefer the smallest tree.
 Least depth?
 Fewest nodes?
 How should you go about building a decision tree? The space of
decision trees is too big for systematic search for the smallest
decision tree.

Engineering Application of AI - PhD Course - 27

Learning decision trees 1
 learning decision tree  ID3 algorithm:
1. initially decision tree contains an unlabelled node with all of the training
examples and attributes
2. selecting an unlabelled node (n) with non-empty set of training
examples (T) and non-empty set of attributes (A)
 if T is homogen class  n leaf node, label with the classification
 otherwise
 choosing the „best” attribute (B) from A
 extension of the tree with all of the possible attribute values of B (devide into
subclasses)
 classification of T to the children nodes according to the attribute values (assign
the elements of T to subclasses)
 continue with step 2.
 building the tree top-down

Engineering Application of AI - PhD Course - 28

Learning decision trees 2
 how to choose the „best” attribute?
 attribute divides the examples into homogen classes
 otherwise attribute makes the most progress towards this
 hill-climbing search on the space of decision trees
 searching for the smallest tree  heuristics (maximum information gain)
 information gain of an attribute test
 measures the difference between the original information requirement and the
new requirement (after the attribute test)
 information gain (G) it is based on information contents (entropy, E)

n Si
S , A) S:Eset
G (where:   examples,
( Sof) classified E ( Si ) A: attribute
1 n
S
S , … , S : subsets of iS1according to A
E: entropy

S S S S
E(S )   log 2  log 2
S S S S
Engineering Application of AI - PhD Course - 29
Learning decision trees 3
Author Thread Length Reads
1 known new short true
2 unknown new long true
3 unknown old short false
4 known old short true
5 known new long true
6 known old long true
7 unknown old long false
8 unknown new long true
9 known new short true
10 unknown old long false
11 known new short true
12 known old long true
13 known new long true

Engineering Application of AI - PhD Course - 30

Using decision trees 1
 major problem with using decision tree: overfitting
 occurs when there is a distinction in the tree that appears in the
training examples, but it doesn’t appear in the unseen examples
 handling overfitting:
 restricting the splitting, so that you split only when the split is
useful
 allowing unrestricted splitting and pruning the resulting tree
where it makes unwarranted distinctions:
 examples are devided into two sets: training set and test set
 constructing a decision tree with the training set
 examining all of the nodes with the test set: whether the subtree
under the node is replaceable with a leaf node

Engineering Application of AI - PhD Course - 31

Using decision trees 2
 supporting knowledge acquisition/ fast prototype-making
(rule-based/ hybrid systems with inductive services)
 each one row in the matrix of training examples is a rule
Author Thread Length Reads IF (Author = known) and (Thread = new) and (Length = short)
THEN (Reads = true)
1 known new short true
IF (Author = unknown) and (Thread = new) and (Length = long)
2 unknown new long true THEN (Reads = true)
… …

 better: each one path (root  leaf) on the decision tree is a rule

IF (Author = known)
THEN (Reads = true)
IF (Author = unknown) and (Thread = new)
THEN (Reads = true)
IF (Author = unknown) and (Thread = old)
THEN (Reads = false)

Engineering Application of AI - PhD Course - 32

Main components of inductive
systems

Knowledge representation:
The matrix of training examples:
attributes, values

Reasoning and control:

Algorithm, which constructs a
decision tree using the matrix of
training examples and operates
the generated system.

Engineering Application of AI - PhD Course - 33

Main steps of inductive systems
 problem definition (knowledge representation):
 attributes (head of the matrix, generate coloumns, define object
classes)
 training examples (fill the raws of the matrix, define instances)
 reasoning (generating a hypothese)
 checking the contradiction freeness of the training examples
 learning optimal decision tree (DT)  knowledge base
 control (operating the system)
 classification of user’ (unknown) examples (traversing DT)
 analysis of user’ examples (with the help of DT)

Engineering Application of AI - PhD Course - 34

Hybrid techniques

Engineering Application of AI - PhD Course - 35

Characteristics of hybrid systems
 supporting various programming techniques:
 frame-based techniques
 rule-based techniques
 data-driven reasoning
 goal-driven reasoning
 inductive techniques
 realization:
 using of object-oriented tools

Engineering Application of AI - PhD Course - 36

Frames
 knowledge-representation unit developed on epistemology
foundations
 formal tool using for description of structured objects or events
or notions
 characteristics of frames:
 a frame contains:
 the name of the object/event
 its important properties (attributes)  stored in slots (slot identifier, type, value
– it can be another frame)
 classes, subclasses, instances
 hierarchical structure (is_a, instance_of relations)
 inheritance (classes - subclasses, classes - instances)
 procedures controlled by events: daemons

Engineering Application of AI - PhD Course - 37

 Formalization of frames 1
 directed graph

f_name name
Person l_name Subject preconditions

is_a is_a instance_of

name Expert_
status systems
subjects
subjects Teacher Student ES
preconditions
AI
instance_of instance_of
f_name f_name Peter
l_name subjects
status Rozália Peter
l_name
subjects
Kis

Engineering Application of AI - PhD Course - 38

Formalization of frames 2
 description in frame-based environment
frame person frame student frame subject
is_a class is_a person is_a class
f_name: subjects: collection_of subject name:
l_name: end precond: collection_of
end subject
end

frame Peter frame ES

instance_of student isnstance_of subject
f_name: Peter name: Expert_systems
l_name: Kis precond: AI
subjects: ES end
end

Engineering Application of AI - PhD Course - 39

Formalization of frames 3
 object-attribute-value triplets

<Peter, f_name, Peter>

<Peter, l_name, Kis>
<Peter, subjects, [ES]>
<ES, name, Expert_systems>
<ES, preconditions, [AI]>

Engineering Application of AI - PhD Course - 40

Daemons 1
 active elements of a frame system
 standard built-in procedures
 assigned to the attributes of the classes and instances
 automatically invoked in case of predefined changing in the
value of the slot
 usual daemons are as follows:
 when-needed: describes the steps to be performed when the value of
slot is read
 when-changed: is invoked when the value of the slot is changed
 when-added: contains the actions to be performed when the slot gets
its first value
 when deleted: is executed when the value of the slot is deleted

Engineering Application of AI - PhD Course - 41

Daemons 2
 the executable part of the daemons is determined by the
user or it may even be empty
 execution is controlled by events
 daemons can invoke (call) each other via changing slot
values  spread over and over
 the operation of a frame system is described in an
indirect way (embedded in the daemons)
 daemons can be used for restricted data-driven
reasoning

Engineering Application of AI - PhD Course - 42

Daemons versus rules
Daemons Rules

Faster and more independent than rules. A rule is invoked by another rule or in case of
„Reason/action” is connected to the changes in presence of a certain data.
values and the system’ responses. The execution depends on the situation and
They act in autonomous way. cannot be seen in advance.
Less readable than rules. Easy to read.
(daemons are defined on the implementation (symbolic formalism, natural-language like)
language of the given tool)
They handle the pre-defined changes of the The built-in knowledge of the rules steams
given attribute-values. freely to all of the rules.

The range of a deamon is bounded statically in The range of a rule is stand out dynamically in
advance. (more or less flexible) run-time. (flexible, creative problem solving)

Engineering Application of AI - PhD Course - 43

Hybrid techniques
 rules: used for description of heuristic knowledge
 frames: contains both descriptive and procedural
knowledge of the given objects/ events/ notions (altogether
in one place!  easy to read and modify, the effects of
modifications can be held easily)
 inference engine of hybrid techniques can contain:
 mechanisms insuring inheritance and handling of daemons
 mechanisms insuring message changing (object-oriented)
 data-driven and/or goal-driven reasoning mechanism
 can support the organization of rules and/or frames into hierarchical
modules
 can support making and using of meta-rules

Engineering Application of AI - PhD Course - 44

Symbol-manipulation
techniques

Engineering Application of AI - PhD Course - 45

Programming languages of AI
 high-level symbol-manipulation languages are used to
support the implementation of AI methods
 LISP (LISt Processing)
 based on the notion and operations of lists
 all of the problems can be described in the form of function calls
 PROLOG (PROgramming in LOGic)
 high-level declarative language
 define relationships between various entities with the help of logic
 special type of clause (A  B1 …  Bn): fact, rule, question
 reasoning environment with a built-in inference engine
 answer to a question with the help of logical reasoning
 goal-driven (backward) reasoning

Engineering Application of AI - PhD Course - 46

Comparison of symbol-manipulation
and traditional techniques
Traditional programming languages LISP PROLOG

numeric calculus symbol-manipulation symbol-manipulation

Neumann-principle languages functional approach relation approach

consist of sequence of commands sequence of evaluation of function- based on mathematical logic
executed in a predefined order expressions (predicate-calculus)
(-calculus)
main elements: commands main elements: functions (procedures) main elements: predicates (relations
among objects)

procedural (executing in a predefined procedural declarative (defining only the description

order) of the problem)

executing mechanism have to be defined executing mechanism have to be defined built-in executing mechanism
by the programmer by the programmer (goal-driven reasoning with backtracking
search strategy)
the structure of program and data is the sructure of program and data is the the sructure of program and data is the
different same (can produce, execute other same (can produce, execute other
programs, can modify themselves) programs, can modify themselves)

readability: LISP-like hard to read easy to read

Engineering Application of AI - PhD Course - 47

Case-based techniques

Engineering Application of AI - PhD Course - 48

Case-based reasoning (CBR) 1
 basic assumption: like was the past like will be the future
 the „really” observation can be describe hard with the help of
classical rules
 it consists of interconnected relationships of more or less
generalized events
 idea:
 solving problems based on solutions for similar problems solved
in the past
 requires storing, retrieving and adapting past solutions to similar
problems

Engineering Application of AI - PhD Course - 49

Case-based reasoning 2
 solve a new problem by making an analogy to an old one
and adapting its solution to the current situation
 retrieving a case starts with a problem description and ends
when a best matching case has been found
 all case-based reasoning methods have in common the
following process:
 identifying a set of relevant problem descriptors
 retrieve the most similar case (or cases) comparing the case to the
library of past cases
 reuse the retrieved case to try to solve the current problem
 revise and adapt the proposed solution if necessary
 retain the final solution as part of a new case

Engineering Application of AI - PhD Course - 50

Case
 a case represents specific knowledge in a particular context
 there are three major parts in any case:
 a description of the problem/situation
the state of the world when the case is available
 solution
the chain of operators that were used to solve the problem (solving
path)
 outcome/consequence
the state of the world after the supervention of the case (description of
the effect on the world)
 in addition to specific cases, one also has to consider the
case memory organisation

Engineering Application of AI - PhD Course - 51

Case - indexing
 the most important problem in CBR
 how do we remember when to retrieve what?
 essentially, the indexing problem requires assigning
labels to cases to designate the situations in which they
are likely to be useful
 indexing of cases - issues
 indexing should anticipate the vocabulary a retriever might use
 indexing has to be by concepts normally used to describe the
items being indexed
 indexing has to anticipate the circumstances in which a retriever
is likely to want to retrieve something

Engineering Application of AI - PhD Course - 52

Main components of case-based
systems 1
 case-base (library of cases)
 tools for determining of key-elements of actual case and
for retrieving of most-similar cases
 for speeding of data-retrieval  indexing
 for finding suitable cases  pattern, similarity-estimation
 tools for the solution’ adaptation according to the
specialities of the new case
 finding the deviations, implementation of alterations in the
suggested solution (ex. null-adaptation, parameter adjustment)
 supervision (solution after the adaptation is suitable or not)
 learning (finding the reason of failure or enclosing the case to the
case-base)

Engineering Application of AI - PhD Course - 53

Main components of case-based
systems 2
new problem

indexing

retrieving
(case-matching)
similar cases

case-base selecting

proposed solution

adapting

learning checking

solution
Engineering Application of AI - PhD Course - 54
Advantages and disadvantages
 advantages:
 case-base is more objective and formal than the expert’s interpretation
(knowledge of expert’s)
 knowledge are represented in an explicit way
 case can be defined for incomplete or badly-defined notions
 CBR is suitable for domains for which a proper, theoretical foundations do
not exist
 CBR is applicable in default of algorithmic method
 easy knowledge acquisition (get well during usage)
 disadvantages:
 CBR solves only the problems covered by cases
 CBR might use a past case blindly without validating it in the new situation
 solution is time-demanding (also in case of proper indexing)

Engineering Application of AI - PhD Course - 55

Rule-based systems versus case-
based systems
Rule-based systems Case-based systems
Rule: symbolic pattern Case: collection of data, constants
Rule: individual unit, independent of the other rules, Case: depends on the other cases (often overlap each
consistent piece of field of interest other), individual unit of the field of interest

Retrieving rule: exact matching Retrieving case: partial matching

Using of rules: general iterativ cycle Using of cases: several steps (approximate retrieval,
adaptation, refinement)
The model of the problem have to be developed The model of the problem needn’t be developed
(sometimes it is hard or impossible)
The knowledge-acquisition of field of interst is hard and The knowledge-acquisition of field of interst is limited to
time-demanding collecting and analysing the past cases

Development time is long Development time is short

Slow, handling of many data is difficult Many data is treatable with the useing of database –
handling techniques
Enlargement is hard (the validation have to be repeated Enlargement and development is easy.
after enlargement)
Learning is not supported It is able to learn (preserving new cases)

Engineering Application of AI - PhD Course - 56

Summary
 Knowledge-based systems, expert systems
 Base techniques of knowledge-based systems
 rule-based techniques
 inductive techniques
 hybrid techniques
 symbol-manipulation techniques
 case-based techniques

References
 K. M. Hangos, R. Lakner and M. Gerzson: Intelligent Control Systems. An
Introduction with Examples. Kluwer Academic Publishers, 2001. Chapter 5.
 D. Poole, A. Mackworth, R. Goebel: Computational Intelligence. A logical
Approach. Oxford University Press, 1998. Chapter 6.
 I. Futó (Ed.): Mesterséges intelligencia. Aula Kiadó, 1999. Chapter 12. (in
hungarian)

Engineering Application of AI - PhD Course - 57