Upload
diana-bennett
View
220
Download
0
Tags:
Embed Size (px)
Citation preview
25 juin 2010
Interactive Genetic Algorithmsfor Creative Enhancement in UI design
Dimitri MassonAlexandre DemeureGaelle Calvary
1
2
Problem statement
• User Interface (UI) design is a creative process
• General question: How to support designers’ exploration of the design space?
Instantmessenger
…
…Design spaceof possible UIs
My research proposal
• Presentation of generated design solutions for the application under study
• Motivation– Compliance with designers’ practice: samples browsing– Sources of inspiration
3
Outline
• State of the art
• Approach
• Prototype
• Early evaluation
• Conclusion and future work
4
State of the art
• Creativity– Creative person
– Creative product
– Creative process
• Creative evolutionary systems– Interactive Genetic Algorithms (IGA)
– IGA & HCI
• UI design process– Design process
– Tools
– Model Based UI
5
State of the art
• Creativity– Creative person
– Creative product
– Creative process
• Creative evolutionary systems– Interactive Genetic Algorithms (IGA)
– IGA & HCI
• UI design process– Design process
– Tools
– Model Based UI
6
State of the art
• Creativity– Creative person
– Creative product
– Creative process
• Creative evolutionary systems– Interactive Genetic Algorithms (IGA)
– IGA & HCI
• UI design process– Design process
– Tools
– Model Based UI
7
State of the art: Creative process
• A Darwinian process
8
Variation & recombination
Selective retention
State of the art: Creative process
• A Darwinian process
9
Variation & recombination
Selective retention
State of the art: Creative process
• A Darwinian process
10
Variation & recombination
Selective retention
State of the art
• Creativity– Creative person
– Creative product
– Creative process
• Creative evolutionary systems– Interactive Genetic Algorithms (IGA)
– IGA & HCI
• UI design process and places of creativity– Design process
– Tools
– Model Based UI
11
State of the art: IGA
Interactive Genetic Algorithms: Suitable for large problem space for which
– no analytical method can be applied– but candidate solutions can be evaluated
12
Variation & recombination
Selective retention
Performed by the system Performed by the designer
13
State of the art: IGA
Init a population of candidate solutions
Evaluate candidate solutions
Apply mutation and crossbreedings
Select promising candidate solutions
Design space
0.1
0.2
0.2
0.22
State of the art
• Creativity– Creative person
– Creative product
– Creative process
• Creative evolutionary systems– Interactive Genetic Algorithms (IGA)
– IGA & HCI
• UI design process– Design process
– Tools
– Model Based UI
14
• Three contributions: [Monmarché 1999 & Oliver 2002] [Quiroz 2007] [Plessi 2008]– Input: one UI sample– Dimensions of the design space: fonts, colors, layout– Output: definitive UIs for the end-user not for the
designer
15
State of the art: IGA & HCI
State of the art: IGA & HCI
The exploration of the design space is limited
• Evolution is limited to predefined elements
• The UI remains basically the same
Absence of high level description of the UI.
16
State of the art: IGA & HCI
Absence of high level description of the UI.
As a result:
• Evolution is limited to predefined elements
• The UI remains basically the same – No widget substitution
17
dd/mm/yy
State of the art: IGA & HCI
Absence of high level description of the UI.
As a result:
• Evolution is limited to predefined elements
• The UI remains basically the same – No widget substitution– No UI enrichment
18
dd/mm/yy
dd/mm/yySelect a date:
State of the art: coverage
• Creativity– Creative person
– Creative product
– Creative process
• Creative evolutionary systems– Interactive Genetic Algorithms (IGA)
– IGA & HCI
• UI design process– Design process
– Tools
– Model Based UI
19
State of the art: Model based UI
• High level description of the UI: concept & task model
20
Transformation 1
Transformation 2
Transformation N
• Transformations
…
Outline
• State of the art
• Approach
• Prototype
• Early evaluation
• Conclusion and future works
21
My approach
• Three key points:#1 Combination of IGA and Model Based UI at C&T.
#2 Evolution of transformations from C&T UI
#3 Support from a knowledge database
22
Concept and task (C&T) Transformations(C&T UI)
IGA
Evolve Transformations(C&T UI)
uses
23
My approach: #2 transformation
• Transformation = one tree– Edges: Select C&T elements– Nodes: Translate C&T elements into UI elements
Root of transformation
Select choices
Translate into radiobuttons
Select abstract tasks
Translate into windows
Select descendants abstract tasks
Translate into frameswith a red background
Root of the task
24
My approach: #2 Genetic operations
• Mutation– Edge: changing selector– Node: Changing translation– Add/remove nodes
Select choices
Translate into radiobuttons
Select abstract tasks
Translate into windows
Select descendants abstract tasks
Translate into framesTranslate into comboboxes
Select children abstract tasks
24
25
My approach: #2 Genetic operations
• Crossbreeding
My approach: #3 Knowledge database
• Widget representations for each task
• Attributes for each widget and manipulation function
26
Choice of date
Entry textFormatBackground colorForeground color
CalendarWidthHeight
23/11/08
Outline
• State of the art
• Approach
• Prototype
• Early evaluation
• Conclusion and future works
27
• VIDEO
28
Prototype: Implementation
• TCL
• COMET: a widget toolkit [Demeure 2008]– Task level interactors supporting multiple presentations– Database storing design knowledge
29
Outline
• State of the art
• Approach
• Prototype
• Early evaluation
• Conclusion and future works
30
Early evaluation
• The purpose is to– assess the effectiveness of the approach by collecting as
many ideas and practices as possible– identify new areas for improvement
• The purpose is not to– assess the creativity improvement– evaluate the usability of the prototype
31
Early evaluation: Protocol
• Protocol defined with N. Mandran, sociologist
• Face-to-face semi-structured interviews– Questionnaire about design activity
• Personal activity, Design practices, tools, etc.
– Sketch design of an Instant Messenger• e.g.: Sketches done by E1, Appendix C
– Manipulation of Magellan• Thinking aloud technique
• 11 participants: Designers, HCI experts, Students
• 2h/interview + 4h/post-analysis (> 2 weeks)
32
Early evaluation: Results
I conducted an experiment to test 3 hypothesis
1. Designers find benefits from the IGA approach.– “Not perfect but interesting”.
– The process enable to reject bad UIs.
2. Designers find the evolution of widgets valuable.– Presentation of elements in the actual UI.
3. Designers find the UI enrichment valuable. – Few even notice the enrichment process.
As well as eliciting unforeseen needs about such approach.– Need for layout support.
– Need for User Interaction.
33
Outline
• State of the art
• Approach
• Prototype
• Early evaluation
• Conclusion and future work
34
Conclusion: My contributions
• Increment on previous approaches: IGA & MBUI– Transformation as individual representation– Widget substitution– UI enrichment
• Early inquiry of designers’ needs – Previous work focus more on the IGA than on the HCI
• Magellan
• [Masson 2010]: ACM EICS, Berlin, June 20-23
35
Conclusion: Difficulties
• Large scope of the state of the art
• Automatic generation of UI is complex
• Conducting experiment is complex– Very long to set up– Require experience
• IGA requires a good modeling of individuals and fine tuning of the process
36
Conclusion: Future work
• Direct edition of the samples presented – Integration of user sketches
• Support several level of fidelity
• Partial UI evaluation and evolution
• New evaluation
37
Questions
38
Research direction
• Presentation of generated design options.
39
• A faire nb de ligne de code.
40
Early evaluation: findings
• Purpose 1: Assess the effectiveness of the approach by collecting as many ideas and practices as possible
• H1: Designers find benefits from the IGA approach
– “Not perfect but interesting”.
– The process enable to reject bad UIs• H2
• H3
– identify new areas for improvement
41