of Bayesian Network Diagrams Dr Kamaran Fathulla University of Essex International Academy kamaranessexacuk June 2011 Clay Tablet map from GaSur Kirkuk 2500 BC Some 2500 years ago people have used clay tablets to express boundaries groupings and routes ID: 788601
Download The PPT/PDF document "A Richer Understanding" is the property of its rightful owner. Permission is granted to download and print the materials on this web site for personal, non-commercial use only, and to display it on your personal computer provided you do not modify the materials and that you retain all copyright notices contained in the materials. By downloading content from our website, you accept the terms of this agreement.
Slide1
A
Richer Understanding
of Bayesian Network Diagrams
Dr.
Kamaran
Fathulla
University of Essex
International Academy
kamaran@essex.ac.uk
June 2011
Slide2Clay Tablet map from Ga-Sur, Kirkuk, 2,500 B.C.
Some 2500 years ago people have used clay tablets to express boundaries, groupings, and routes
Slide3Diagrams are everywhere
Applied psychology
Cognitive scienceLinguisticsVisual programmingData visualisationGraphic designEducationHistory and philosophy of scienceArchitecture
Blackwell and Engelhardt (1999)
Eppler, 2003
Slide4But.... Diagrams continue to be
difficult
to understand and work withThere has been a substantial growth in the use of diagrams in earlier stages of the research process to collect data. Despite this growth, guidance on this technique is often isolated within disciplines.Numerous references
are cited on people expressing difficulty or discomfort with diagramming.
Muriah J Umoquit, et al (2011) A multidisciplinary systematic review of the use of diagrams as a means of collecting data from research subjects: application, benefits and recommendations
Slide5Sources of the difficulties
1.
Diversity of types of diagramsBoxes and linesContour mapsBar charts etc..
2. Diversity of types of change (dynamics)Different types of diagrams require different editing operationsDifferent diagram types have different rules of well formedness
Need to relax these rules but maintain well formedness3.
All of the above in semantically mixed diagrams
Even single type diagrams have mixed semantics
Next..... Bayesian type diagrams
Slide6Influence Diagrams ID
First introduced in the mid 70s.
ID’s have become a de facto standard for representation of Bayesian decision problems.There is not too much feedback from analysts and experts about their experiences using IDs for building decision-making models.Constructing ID is considered as an art.
Concha Bielza, Manuel Gómez, Prakash P. Shenoy (2009)
Modeling Challenges with Influence Diagrams: Representation IssuesSCHOOL OF BUSINESS WORKING PAPER NO. 319
An Influence Diagram
Slide7James E. Corter et al (2009) Bugs and Biases: Diagnosing Misconceptions in the Understanding of Diagrams. Proceedings of the 31st Annual Conference of the Cognitive Science Society
Misconceptions:
No order misconception
: Failure to understand that pathsare ordered (e. g., commits CYB).LAN-as-path misconception: Failure to see that a path
through a LAN visits intervening nodes in sequence, likea physical path (e.g., commits YRBMC)Omit bridge misconception/error
: Failure to recognize or
list a true bridge node (e.g., commits BC or omits BMC)
Commit bridge misconception/error
: Inferring a
nonexistent bridge node, or selectively committing the
LAN-as-path error (e.g., commits YRB)
Reading-order misconception/bias
: Listing only forward
paths (misconception); or omitting backwards paths more
often (bias)
Generate
all possible shortest paths of information flow for Network
1 and Network 2
On the difficulty of Bayesian Network
Diagrams BND
Where to go from here?
It has been argued that Bayesian reasoning is
counterintuitive
. People do not employ Bayesian reasoning intuitively, find it very difficult to learn Bayesian reasoning when tutored, and rapidly forget Bayesian methods once the tutoring is over. This limitation seems to hold equally true for novices and highly trained professionals in a field .
Yudkowsky (2009) An Intuitive Explanation of Baye's Theorem
Slide8Pointers for a better understanding diagrams
“Knowledge Representation for people”
Brachman (1985)
“Making Bayesian networks more accessible to the probabilistically unsophisticated”
Eugene
Charniak
(1991) Bayesian Networks without Tears. IA Magazine
Slide9Towards a new approach
1 - Pointers
Central role of human functioning D. Watson, Nobel prize winner (1968) (who discovered the structure of DNA): "drawing and thinking are frequently so simultaneous that the graphic image appears almost an organic extension of the thinking process".
Everyday diagramming
Composed of visual elements which each have some symbolic meaning
Expressing several different types of symbolic information
Contain a number
of
mixed visual styles
to express these different types respectively
Growing rather than finished
Having parts already drawn being replaced by others
Part of the thinking is inherent in communication
The diagramming process is fluid
Need a meta approach
Slide10Existing frameworks of understanding diagrams
Indexing
Synopsie
Free rides Useful awkwardnessUnevenness Lability
PermissivenessProgramming Salience Provisionality Story Content
Modifiability
Controllability
Referencability
Focus
Coordination
Documentation
Consistency
Accountability
Traceability
Visual immediacy
Visual impetus
Visual impedance
Plastic-robust
Weakly-strongly structured
Abstract-concrete
Different meaning - common structure
General-specific
Conventionalizedcustomized
Abstraction
Hidden dependencies Premature commitment Secondary notation
Viscosity Visibility Closeness of mapping Consistency Diffuseness Error-proneness Hard mental operations
Progressive evaluation Provisionality Role-expressiveness Creative Ambiguity Specificity
Detail in context
Too fine grained
Too broad/general
Bertin, Peirce, etc
Eppler et al (2008)
Reductionist
Spatial, visual, cognitive, etc..
Slide11Dooyeweerd's Meaning-Oriented Approach
Dooyeweerd said (
1955), "Meaning is the being of all that has been created and the nature even of our selfhood. It has a religious root and a divine origin."
This led him to presuppose that Meaning is the primary property of that is, and that Existence emerges from Meaning. Meaning is better grasped intuitively, not theoretically.
The ordinary person can understand it in the full, holistic sense of the word
Towards a new approach
2 – Philosophy
Slide12Meaning is given to the cosmos (by its Creator) as a framework.
The framework is a framework of law which provides guidance for how all entities function.
Diversity of types of existing based on a diversity of
types of meaningThis leads into the notion of
Aspects
Towards a new approach
3 – Aspectual Philosophy
Aspects
What do they mean
Numeric
Discrete quantity
Spatial
Continuous extension
Kinematic
Motion
Physical
Energy and matter
Biotic
Life and vitality
Sensitive
Seeing and feeling
Analytic
Distinction
Formative
Formative power
Lingual
Symbolic representation
Social
Social interaction and institutions
Economic
Frugality
Aesthetic
Harmony
Juridical
What is due
Ethical
Self-giving love, generosity
Pistic
Faith, vision, commitment
Aspects are irreducible
Aspects are related
Aspects are not absolute
Aspects are independent of things or entities
Aspects are rich
Slide13Numeric
Concerned with the number of shapes, etc.
Spatial
Concerned with things like shapes in the diagram, spatial arrangement, connectivity, etc.
Kinematic
This concerns the activity of diagramming itself when conveying symbolic meaning
Physical
This concerns the specific physical medium or substrate used to display the diagram
Biotic
This aspect concerns life functions that are a necessary precondition for the sensory and nervous systems to function well
enough for diagram creation or reading.
Sensitive
This concerns perceiving color, texture, etc.. and also emotions
Analytic
This aspect concerns the activity of distinguishing what is meaningful in a diagram from its background
Formative
The laws of this aspect govern the creation or formation of the symbol structure being expressed, and also of the method by which the drawing is composed.
Lingual /
Symbolic
This concerns conveyance of meaning
Social
This concerns social impact and norms used when diagramming
Economic
This concerns the cost of producing the diagram
Aesthetic
This concerns how well a diagram is presented
Juridical
This concerns issues such as copyrights
Ethical
Diagrams should not be offensive
Pistic
Diagrams could express faith
Qualifying Aspects of Diagrams
Multi Aspectual nature of diagrams
Slide14What is a SySpM?
A distinct collection of Sy elements, or as
A distinct collection of Sp elements, or asA distinct collection of M
No.
The SySpM
Mapping
1
Boxes and Arrows
Item mapped onto Box,
Relationship mapped onto Arrow
2
Communicating Similarity
A collection of items mapped onto a collection of shapes
3
Map of Objects
Item location mapped onto Icon position
4
Set Membership
A shape inside a loop mapped onto Member of a set
5
Bar Charts
Magnitude mapped onto Length of a bar
6
Route Maps
Route mapped onto Curvilinear Line
7
Contour Maps
The set of location with the same quantitative value mapped onto Closed continuous curve
8
Surface Coverage
Region mapped onto Area
Slide15Symbolic Aspect
Primary
Sy1. Items, entities: meaning deliberate intention of creator to symbolize a distinct concept2. Relationship: deliberate intention of creator to symbolize the idea that two items are related in a meaningful
way Secondary Sy
3. Direction of relationship4. Relationships can be treated as items
5. Types of items
6. Types of relationship
7. Names to identify the items or relationships
Box and Arrow type of diagram
Slide16Spatial Aspect
Primary Sp
1. Boxes: as being present rather than absent.2. A line which is straight, curved, or bent. Secondary Sp3. Boxes shape: this could be a pictogram, a diamond, a square, etc. Box shapes are considered as a spatial sub type of a thing and in this case it is a visual characteristics which may also include color, texture, etc.
4. Box visibility: distinction between 1 and 3 becomes important in software generated diagrams. A box could be present but hidden. Software should be able to understand that an invisible box does not necessarily mean it is not present.
5. Edges of boxes6. Position of box
7. Size of box
8. Boundary: outside rather than inside. This refers to the spatially defined vicinity of a thing. This is important in cases where a line ends in the vicinity of a box rather than on one of its edges.
9. Route of line.
10. Line texture: this refers to the use of dotted, dashed, thick/narrow features of lines.
11. Connect/Attach: this is about the attachment of a line end to a box in an unambiguous way as opposed to a line being in the vicinity of one box rather than another.
12. Arrow head. These may be of various kinds. They indicate direction of a line.
13. Text. These are used to identify boxes.
14. Empty background (white space): this may actually be decorated as in backdrop.
15. Width of a line.
Slide17Sy
Sp
M
Item
Box
Presence of a box
Relationship
Line
Presence of a line
The potential for creating new items
Background space (backdrop)
Empty = Non present
Item type
Shape of a box
Changing the shape of a box results in changing its item type
Relationship type
Texture
Changing the shape of a line results in changing its line type
Direction
Arrow head or other directionally specific visual effect or target
Redirecting the direction of an arrow results in redirection of its relationship
Name
Text
Text express name. Proximity of text to box/line links name to item/relationship
1. Sp Mapped to Sy
Sp
Size of a boxes (if not seen as different from shape)
Exact route of a line
Position of boxes
Relative alignment/proximity of boxes
Boxes/lines crossing
Thickness of a line does not map to anything in this
SySpM
except where thickness is treated as texture.
3. Sp Not Mapped to Sy
Sy reason
Sp constraint
Relationships must always be between items
Both ends of a line must always be connected to a box.
Items and its relationships move together
When a box moves all lines connected to it move with it
Items are self contained things
A box is a closed shape
2. Constraints
Putting it all together Continued
Slide18Sy
Sp
M
Create item
Box appears in empty space
Empty space indicate no items created thus far
Delete an item
Erase box
The space of a box becomes empty
Hide an item
Make box invisible
Removal of a box’s visual properties expresses hiding an item
an item
Divide a box into two boxes of similar size and shape to original box
Divide a box into two, ensure all lines remain joined to the two created boxes
Merge two items
Combine two boxes into one box of similar size and shape to original boxes
Bringing together two boxes indicates intention to combine
Create a relationship
Draw a line between two boxes
Growth of line indicates intention to relate and direction suggests which item will eventually relate
Detach a relationship
Erase a line
Detachment of both ends of a line indicates intention to erase
Delete an item with all of its relationships
Erase a box and lines connected to it
The space of the deleted item and its relationship becomes empty
Redirect a relationship
Disconnect a line from the edge of a box and change its position
Detachment of line end from box indicates intention to redirect, direction of movement of line can suggest which item, attachment o item indicates fulfillment of intention to redirect
Split a relationship to form two parallel relationships
Thicken the line, then split the line, then bend them apart, but ensure both lines continue to touch the original two boxes
Delete a line and create two other lines relating the same items
4. Handling Change
Slide19What does
this mean
for BND’s ?
Bayesian networks are complex diagrams
A definition of BND: “Bayesian networks are directed acyclic graphs whose:
nodes
represent
variables
, and whose
edges
represent
conditional relationships
between two variables”
Consider the use of the Box and Arrows diagram type and modify its constraints to reflect BND
Develop BND
a
s a new diagram type.
OR
1
2
Slide20A typical BND
Box and Arrows diagram
Tabular display