Community-Driven Adaptation Iqbal Mohomed Department of Computer Science

Community-Driven Adaptation Iqbal Mohomed Department of Computer Science University of Toronto Wallach / de Lara

Mobility and Adaptation • Content/applications target the desktop • Resource rich environment • Stable • Mobile clients • Limited resource (nw, power, screen size) • Variable resources (Mbps to Kbps) • Adapt application/data to bridge gap

Manual/Static Adaptation • Publishers make available content for several classes of devices • e. g. , HTML and WAP versions of Web page • Disadvantages: • High cost • Several copies • Maintaining consistency and coherence • Continuous effort to support new types of devices • You can never cover all possible versions! • In practice: • Only done for few high-traffic sites • Limited number of devices

Automatic/Dynamic Adaptation • Adapt content on-the-fly • Optimize for device type, user preferences, context, etc. • Typically done using proxies Proxy

Existing Approaches • Rule-based adaptation • Convert images larger than 10 KB to JPEGs at 25% resolution • Constraint-based adaptation • Functions that relate "user happiness" to metrics (resolution, color depth, frame rate, latency) • Find point that meets all constrains and maximizes "happiness"

Limitations • Cannot have rules/constrains per-object per-device • Hard to define correlation between "user happiness" and metrics • In practice, rely on small sets rules/constrains • Based on broad generalizations • e. g. , "typical image is viewable at resolution X" • Content agnostic

Problem • User does not care equally about all objects • The fidelity at which an object is useful depends a lot on the task and the object's content (semantics) 10%

Problem • User does not care equally about all objects • The fidelity at which an object is useful depends a lot on the task and the object's content (semantics) 10% 50%

Observations • Computers have a really hard time judging if adapted content is good enough for a task • People can do this easily! Have the users decide how to adapt content!

Community-Driven Adaptation • System makes initial prediction as to how to adapt content (use rules and/or constrains) • Let user fix adaptation decisions • Feedback mechanism • System learns from user feedback • Improve adaptation prediction for future accesses

How it Works Server 1 Correct Application Proxy Mobile Server 2 Prediction

Draw Backs • User is integral part of adaptation loop • Significant burden on user • Iterative process is slow and frustrating • No way people are going to accept this for every access!

Hypothesis • User can be grouped into communities • Community members share adaptation requirements • Adapted content that is good for one member is likely to be good for other community members • By tracking a few users we can learn how to adapt content for the community as a whole

Research Questions • How good are CDA predictions? • What are good heuristics for learning how to adapt? • At what granularity should user accesses be tracked? (e. g. object, page, site, etc. ) • How do we classify users into communities? • Does this classification change over time? • Types of adaptations supported by this technique • Fidelity, page layout, modality (text to voice, video to image) • UI • Good UI for working with adapted data • Effects of UI on quality of adaptation prediction

Performance Evaluation • Goal: Quantify extend to which CDA predictions meet users’ adaptation requirements • Approach: • Step 1: User study • Create trace that captures levels of adaptation that users consider appropriate for a given task/content • Step 2: Simulation • Compare rule-based and CDA predictions to values in trace

Simples Meaningful Scenario • 1 kind of adaptation • 1 data type • 1 adaptation method • Fidelity • Images • Progressive JPEG compression • 1 community • Same device • Laptop at 56 Kbps • Same content • Same tasks

Prototype Proxy • Adaptation proxy • Transcode Web images into PJPEG • Split PJPEG into 10 slices • Client • Microsoft Internet Explorer 6. 0 • IE plugging enables users to request fidelity refinements • Network between client and proxy • Simulated at 56 Kbps

Prototype Operation Proxy • When loading page, provide just 1 st slice • When user clicks on image • Provide additional slice • Reload image in IE • Add request to trace

Web Site and Tasks Site Car show Task Find cars with license plates e. Store Buy a PDA with a camera Uof. T Map Name of all buildings between two BA and Queen Subway Goal: finish task as fast as possible (minimize clicks) Traces capture minimum fidelity level that users’ consider to be sufficient for the task at hand.

e. Store

Trace Characteristics • 77 different images • All tasks can be performed with images available at Fidelity 4 (3 clicks) • Average data loaded by users for all 3 tasks • 790 KB • 32 images are never clicked by any user

Metrics • Extra data • Measure of overshoot • Extra data sent beyond what was selected by user • Extra clicks • Measure of undershoot • Number of time users will have to click to raise fidelity level from prediction to what they required in trace

Results

Results For same clicks, 90% less extra data

Results For same data, 40% less extra clicks

e. Store Fidelity Breakdown

Summary • CDA adapt data tacking into account the content’s relationship to the user task • CDA outperforms rule-based adaptation • 90% less bandwidth wastage • 40% less extra clicks

Future Work • Comprehensive CDA evaluation • More bandwidths • More devices Next 7 months • Automatic classification of users into communities • Other data types • Stored video, audio • Other types of adaptation • Page layout, modality • UI • Good UI for working with adapted data • Effects of UI on quality of adaptation prediction 2 nd & 3 rd year

Research Team • Supervisor: Eyal de Lara • Grad. Students: Iqbal Mohomed Alvin Chin • Under. Students: Jim Cai Dennis Zhao iq@cs. toronto. edu www. cs. toronto. edu/~iq