Transcript Presentation - Department of Computer Science

Fine-Grain Adaptation
Using Context Information
Iqbal Mohomed
Department of Computer Science
University of Toronto
Advisor: Prof. Eyal de Lara
HotMobile 2007: Doctoral Consortium
Challenge
One size does not fit all
Challenge
One size does not fit all
• Adaptation can help!
• Challenge:
• How to pick appropriate adaptation?
• Existing techniques based on rules/constraints do not
consider relevance of content
Thesis
Use context information to determine relevance
of content and adapt based on this information
• We investigate two domains:
• Web Adaptation
• Remote Health Monitoring
Web Adaptation:
Factors to Consider
• Usage Context
Web Adaptation:
Factors to Consider
• Usage Context
• Varying Relevance
Web Adaptation:
Factors to Consider
• Usage Context
• Varying Relevance
• Multiple Usage
Web Adaptation:
Factors to Consider
• Usage Context
• Varying Relevance
• Multiple Usage
• Situational Content
• E.g. Type of device, characteristics of
available wireless link, user’s location
Web Adaptation:
Factors to Consider
• Usage Context
• Varying Relevance
• Multiple Usage
• Situational Content
• E.g. Type of device, characteristics of
available wireless link, user’s location
For fine-grain adaptation, content
must be tailored for both
usage context and situational context!
Taking Usage Context Into Account
40KB
Server 1
Improve Fidelity
Application
Adaptation
Proxy
Mobile 1
Server 2
10KB
20KB
Application
Mobile 2
Prediction
Tailoring Content to Situational Context
70
# of Users
60
50
40
30
20
10
Content
0
1
2
3
4
5
6
7
Image Fidelity
8
9
10
Tailoring Content to Situational Context
70
# of Users
60
50
40
30
20
10
0
1
2
3
4
5
6
7
8
9
10
Image Fidelity
# of Users
90
80
70
60
50
40
30
Content
20
10
0
1
2
3
4
5
6
7
8
9
10
8
9
10
Image Fidelity
70
# of Users
60
50
40
30
20
10
0
1
2
3
4
5
6
7
Image Fidelity
Remote Health Monitoring
Bluetooth,
ZigBee, etc.
Wifi,
GPRS, etc.
Remote Health Monitoring
Bluetooth,
ZigBee, etc.
Wifi,
GPRS, etc.
• Context-Aware Filtering can significantly reduce the
amount of data transmitted
• Use context information to judge what sensor
readings are expected
• Vary fidelity of transmitted data based on whether
sensor readings conform to expectations
Next Steps
• Web Adaptation
• Can we reduce the amount of interaction required,
while still providing fine-grain adaptation?
• How well will our techniques work on a large scale
in the real-world, over an extended period of time?
Next Steps
• Web Adaptation
• Can we reduce the amount of interaction required,
while still providing fine-grain adaptation?
• How well will our techniques work on a large scale
in the real-world, over an extended period of time?
• Remote Health Monitoring
• Can we use context-information to save energy (in
ways other than reducing the amount of data)?
Next Steps
• Web Adaptation
• Can we reduce the amount of interaction required,
while still providing fine-grain adaptation?
• How well will our techniques work on a large scale
in the real-world, over an extended period of time?
• Remote Health Monitoring
• Can we use context-information to save energy (in
ways other than reducing the amount of data)?
• Graduate! And live happily ever after …
Conclusions
• Use context information to determine relevance of
data in a given situation
• When resources are constrained, optimize based on
relevance
Examples:
• When bandwidth is costly, or low link-throughput:
• Perform aggressive fidelity reduction on less
relevant images
• Transmit averages when sensor readings conform
to norms
• When screen real-estate is limited:
• Simplify web page by removing irrelevant images
Conclusions
• Use context
information
to
determine
relevance
of
Collaborators:
data @
in UofT;
a given
situation
Prof. Eyal de Lara, Jin Zhang,
• When
resources
constrained,
optimize
Jim
Cai, Sinaare
Chavoshi
and Alvin
Chin based on
relevance
@ IBM Watson: Dr. Maria Ebling,
William Jerome, Dr. Archan Misra
Examples:
• When bandwidth is costly, or low link-throughput:
• Perform aggressive fidelity reduction on less
relevant images
• Transmit averages when sensor readings conform
to norms
• When screen real-estate is limited:
• Simplify web page by removing irrelevant images
Conclusions
• Use context information to determine relevance of
data in a given situation
• When resources are constrained, optimize based on
relevance
Examples:
• When bandwidth is costly, or low link-throughput:
• Perform aggressive fidelity reduction on less
relevant images
• Transmit averages when sensor readings conform
to norms
• When screen real-estate is limited:
• Simplify web page by removing irrelevant images