COM362 Knowledge Engineering

1
Blackboard ArchitecturesandProblem Solving
Methods

• John MacIntyre
• 0191 515 3778
• john.macintyre_at_sunderland.ac.uk

2
Content

• Introduction to Blackboard Architectures
• Introduction to Other Representations
• Introduction to Problem Solving Methods

3
Background to Blackboards

• Grew out of attempts to build automatic speech
  understanding mechanisms
• the HEARSAY project (Erman et al, 1980)
• required many different knowledge sources to
  process the waveform generated by speech
• difficult problem!
• Noise in the data
• subjective interpretations and context
• I scream vs ice cream
• introduced the idea of triggers for different
  knowledge sources

4
Background to Blackboards

• Further developed by others
• ACCORD (1986)
• solving arrangement problems
• BB1 (1986)
• first attempt at a generic blackboard
  architecture
• PROTEAN (1987)
• deriving protein structures from constraints
• Generic Blackboard Builder (1988)
• further attempt to develop a blackboard shell
• ERASMUS (1989)
• aircraft design - developed by Boeing

5
The Blackboard Metaphor
Blackboard architectures are a metaphor because
they organise and process knowledge in a fashion
similar to a group of people working around a
blackboard
The blackboard is used as a repository for
knowledge
The group leader provides a control function,
guiding and focusing the activities of the
knowledge sources
Each person represents a knowledge source
6
The Blackboard Metaphor

• A group of experts meet in a classroom (with a
  blackboard) trying to solve a problem
• Each expert has relevant (but different)
  expertise for solving the problem
• The problem, and any initial data are written up
  on the blackboard
• All experts watch the blackboard to see how the
  development of a solution is progressing
• If any expert feels they can make a contribution,
  they perform some analysis and write the results
  up on the blackboard

7
The Blackboard Metaphor

• New information added from one expert might spur
  another expert to produce more new information
• The process is collaborative, using many experts,
  and the progressive development of the solution
  is stored on the blackboard
• Experts continue to make contributions until a
  satisfactory solution has bee reached
• Early expert systems tried to model this
  methodology for large-scale systems

8
Blackboard Architectures

• Provide a problem-solving model for organising
  knowledge and potentially a strategy for applying
  that knowledge
• Allows a range of knowledge representation
  methods to be applied
• Segments the knowledge base making it more
  maintainable and making the implementation more
  efficient
• It can be argued that blackboard architectures
  are flexible and powerful - especially where many
  knowledge sources are used

9
Requirements forBlackboard Architectures

• The knowledge sources MUST all use the same
  language when putting information on to the
  blackboard
• There must be some way of deciding who puts
  information on the blackboard and when
• resolution of multiple requests to post
  information need a means of deciding who will go
  first!
• We therefore need three things
• the blackboard itself
• the knowledge sources
• and a SCHEDULER to CONTROL information posting

10
Flexibility of Representation

• Because the knowledge sources are independent of
  each other, they can use their own, individual
  knowledge representation and inferencing schemes

Knowledge Sources
Blackboard
Working Memory
IE
KB
11
Blackboard Architectures

• Each module must provide information in an agreed
  format so that the other modules can use it
• Working memory is subdivided into regions and
  structured appropriately - this is called the
  Blackboard
• Communication between modules takes place only
  via the blackboard
• Each knowledge source contributes to the solution
  whenever the data it requires is available on the
  blackboard

12
Example KBS House Design

• Imagine a KBS to design a house, segmented into
• a KBS to design the general layout of the house,
  the grounds and the paths
• a KBS that, given general layout, will design
  gardens
• a KBS that, given the general layout, will design
  the internal structure of a house, allocate
  rooms, set room sizes and specify windows, doors
  etc

13
House Design (continued)

• a KBS that given specific room details, size
  shape etc, will design the kitchen
• similarly other KBSs that will respectively
  design the dining room, living room, bathroom and
  bedrooms
• Imagine also a blackboard designed to develop and
  communicate these designs

14
Pictorially
15
The system in operation.

• Initially the blackboard will be empty
• While all knowledge sources will have access to
  it, only the KBS to design the general layout
  will have sufficient information to proceed
• After it has finished the KBS for garden design
  and the KBS to design the internal structure will
  both be able to start their work
• The scheduler will determine which of these can
  post information to the blackboard, and when

16
House Design (continued)

• As the internal design progresses and provides
  general details on the individual rooms to the
  blackboard, the KBSs responsible for the detailed
  design of those rooms will begin their work.
• Thus the overall problem is broken down into
  specific tasks
• aids maintenance
• improves efficiency
• allows flexibility of representation and
  inferencing

17
Disadvantages ofBlackboard Architectures

• Blackboard architectures do not explicitly
  specify the control of the individual knowledge
  sources
• The internal structure of each knowledge source
  has a profound effect on the quality of the
  reasoning
• Control is the most important aspect of
  blackboard architectures
• difficult to control dynamic issues of
  concurrency and parallelism

18
Other Representations

• Semantic Nets
• Logic
• propositional
• predicate
• Many other forms of representation!!

19
Semantic Nets

• One of the oldest and easiest to understand
  knowledge representation schemes
• first introducted by Ross Quillian in 1968
• Represents DECLARATIVE knowledge
• A graphical description of knowledge that shows
  relationships between objects
• Essentially, a set of nodes and links
• nodes represent objects or concepts
• links represent relationships between objects

20
Semantic Nets
Example semantic net, depicting relationships
KEY
Nodes which represent objects
Links show the relationship between objects
21
Advantages ofSemantic Nets

• Semantic networks are a powerful, flexible and
  graphical way of representing knowledge
• Often used as communication tool between
  knowledge engineer and expert during knowledge
  acquisition phase of project
• can be used to develop domain model
• can be used to identify objects and relationships
  previously unknown to KE

22
Disadvantages ofSemantic Nets

• However very difficult to inference as IE must
  understand each type of link
• Less reliable than rules as inferring becomes a
  process of searching across diagram
• Diagrams can become complex especially when they
  need to represent exception clauses

23
Propositional Logic

• Based on the idea that knowledge can be
  represented as a set of statements which are
  either true or false
• John is a fantastic teacher
• Students love Monday mornings
• Knowledge can then be represented symbolically
• A John is a fantastic teacher
• B John teaches students on Monday mornings
• C Students love Monday mornings
• this leads us to IF A and B then C

24
Propositional Logic

• This can be re-written as follows
• A ? B ? C
• Note that C is inferred from the presence of A
  and B, it is not asserted directly
• Propositional logic is limited because it only
  allows us to represent knowledge with
  propositions which can be either true or false
• A more powerful representation is known as
  Predicate Logic

25
Predicate Logic

• Predicate logic allows us to break down
  propositions into two components
• arguments
• predicates
• Also allows use of variables, as well as
  propositional inferencing
• Example
• IS_A(John,fantastic teacher)
• IS_A is the PREDICATE
• John, fantastic teacher are the ARGUMENTS

26
Predicate Logic

• Predicate logic allows much more powerful use of
  symbolic representation
• ? x, Man(x) ? ? y, s.t. Woman(y) ? Loves (x,y)
• For any object x in the world, if x is a Man,
  then there exists an object y, such that y is a
  Woman and x Loves y
• Development of predicate logic led to the
  development of special languages for developing
  expert systems
• LISP, Prolog, etc etc

27
Problem Solving Methods

• Experts develop particular methods for solving
  particular types of problem - this is part of
  their expertise
• These methods differ for different types of
  problem, eg
• Design, Diagnosis, Scheduling
• Modelling the problem solving behaviour of
  experts we can develop methods for generic tasks

28
Problem Solving Methods

• PSMs specify the sequence of inference tasks that
  are needed to solve a problem and the domain
  knowledge required by the method

29
Advantages of PSMs

• Two or more PSMs for solving the same generic
  task can be compared and we can use which ever is
  more suited to our task
• The PSM is open to inspection and can be
  improved, without necessarily affecting the
  domain knowledge
• Separating the PSM from the domain knowledge
  enables re-use the PSM can be reused on other
  similar problems in different domains
• the general process of designing a car may be
  very similar to the general process employed when
  designing a house

30
Complexities in Developing PSM

• To enable reuse we need to develop a library of
  PSMs
• These are difficult to classify as we need to
  specify the
• task independence (ie how generic)
• formality (informal descriptions or implemented
  software)
• granularity (size)
• of the PSMs contained within the library.
• Reusability - usability trade off to consider
• Task independent PSMs will require refinement and
  adaptation before they can be used. They can
  however be reused in a range of situations. Task
  dependent PSMs require little adaptation before
  use, however they are less easily used elsewhere