Research Project Proposal

For CSC2228: Topics in Mobile and Pervasive Computing

Project Team members: Jim Cai, Shizhong Li

The Problem:

Nowadays most users would have to pay for their Mobile connections. This leads to the need of bandwidth adaptation. One of the most important classes of adaptation system is the iterative adaptation systems which adapt based on frequent interactions with users. By iterative, we refer to adaptation schemes that perform iterative refinement on adapted objects. This includes 1) Systems that perform iterative improvement without listening to user feedback and 2) Systems performing iterative improvement which, at each step of refinement, learns explicitly from user feedback and uses the feedback to determine future adaptations (e.g., CDA). By interactions, we refer to both the adaptation output interface that present to users and the user input mechanism (e.g., scroll mouse, right click on mouse, explicitly keyboard typing etc.) which adaptation system will gain feedback from. We anticipate a strong connection between the adaptation’s HCI scheme and the adaptation performance. If the hypothesis is proved to be true, we can take a step further to show what would be the most suitable HCI scheme for iterative adaptation systems in the sense that it optimizes the adaptation performance (in terms of less interactions and bandwidth usage). Will this selection of HCI scheme depends on mobile characteristics such as device type, bandwidth and/or screen size? Will the selection of HCI scheme increase the likelihood of users selecting the same fidelity (locality) and hence helps the system to predict future adaptations?

At this stage of the research, we will look at only iterative adaptation system that does not learn from user feedback. We will also only concentrate on comparing different user input mechanism (as opposed to adaptation output interface).

Experiment:

We will design a prototype adaptation system that consists of the following components:

• A content server that stores 2 websites, where each website has numbers of embedded JPEG images.
• A proxy that does JPEG Trans-coding
• Two client machines, each with a different user input mechanism for providing adaptation feedback.

When downloading JPEG images from the server, each client will initially request for the lowest possible fidelity version. We will have:

• One client browser implements iterative improvement of JPEG with right click of the mouse. Each time a mouse click in traced by the system, the fidelity will be increase by one level. This is the base case
• One client browser implements iterative improvement of JPEG with scrolling. Each time user will be asked to provide the next desirable fidelity level by scrolling to that level
• Enabling both click and scroll
• Allow users to select multiply objects to increase fidelity level at one time
• Present time to download as the scale, rather than fidelity level number.

We ask two groups of users to browse over same content and finish same tasks.

(Websites and tasks to be determined)

Questions to be answered:

Objective:

1. Will one user group consume significantly less bandwidth than the other?
2. Will one user group interact less with the interface than the other?
3. Will one user group take much more time in average than the other to finish the same task?
4. Will fidelity selection for a particular image more centralized for one group of user?
5. Other possible statistical results that reflect the differences in user behaviors

Subjective:

1. How do users like the feedback mechanism they are using?

Timeline:

Implementation (Oct.1^st – Nov 15^th)

Experiment (Nov.15^th – Dec 1^st )

Presentation Dec 10^th

Report Dec.10^th – Dec 20^th