Mobile Web Performance: Bottlenecks, Optimizations, and Metrics

Performance and energy optimization Several Web optimizations have been designed for the Web, but their effect on improving page load times have been mixed. The bottleneck during the page load process is not known, different optimizations affect pages differently, and the effect of an optimization on power consumption is not well-studied.
Our goal is to characterize page load bottlenecks and provide a microscopic analysis of the performance and energy effects of any given optimization.
For more details, see here

Gaze tracking: A critical problem is browsers is that the traditional performance metric does not always correlate well with user experience. We define a new user-perceived page load time metric called uPLT. Our goal is to conduct large scale user studies to characterize the relationship between user gaze and uPLT. Using this correlation, we are building new optimizations to optimize the uPLT metric.
For more details, see here.

Modeling HTTP, HTTP/2, and Quic There have been several network protocols designed to improve the page load performance: these include the new HTTP/2 protocol and the UDP-based Quic protocol. Our goal is to systematically model the performance of these new network protocols to understand the implications of deploying them widely.

Web performance in developing regions With collaborators in Cambridge University and Queen Mary University, we are working on the AIMO (Africa Internet Measurement Observatory) project. The goal of our work in AIMO is to significantly improve Web performance in developing regions.

Ubiquitous Mobile Systems

Private Intelligent Assistance (PrIA) Our goal is to build an intelligent assistant system for smartphones that greatly improves privacy, compared to existing systems. Current intelligent assistants such as Cortana and Google Now require users to fork over their private data. Instead PrIA works locally on the user device. PrIA uses cross-device data to build a user model and provide services locally, while reducing privacy leaks.
For more details, see here.

Wearable devices We are researching various techniques to improve the usability and performance of wearable devices. We are currently working on virtualizing the smartphone user interface so that any smartphone application can be controlled from any wearable device, without the need to write additional code.
For more details, see here.