W&M Featured Events
This calendar presented by
William & Mary
[PAST EVENT] Xin Qi - Thesis Defense - Computer Science
March 31, 2015
3pm - 5pm
Abstract
Various sensors compose a unique and indispensable component of modern mobile devices. They digitize physical information into sensor readings, with which contexts of mobile users and devices, such as user activities and device motions, can be recognized. Context recognition stimulates the boom of context-aware applications, such as human activity recognition, location-based service and health monitoring.
While context recognition brings context awareness to mobile systems, how to improve context recognition in terms of accuracy, energy and privacy under the constraints of mobile devices, and how to leverage context awareness for mobile resource saving are still challenging problems. Our research examines the former problem by taking human activity as a sample context and with the goal of avoiding trading one metric for another. Our research examines the latter problem by taking energy and bandwidth as sample mobile resources and leveraging context awareness for energy and bandwidth savings. We hope our research will shed light on the design of general context recognition and context-aware resource saving in mobile systems.
To simultaneously achieve high accuracy, low energy and privacy risks for human activity recognition, we propose RadioSense that utilizes wireless communication features instead of sensing-based features. Through benchmarking we find that communication features are the most discriminative when radio power is low. Low radio power also results in a small communication range, which in consequence reduces privacy risks.
To effectively reduce energy overhead as well as maintain high accuracy for human activity recognition, we propose AdaSense, an activity-aware framework that systematically combines lower power activity binary classification that demands lower sensor sampling rate and higher power multi-activity classification that demands higher sensor sampling rate. To amplify the energy saving, we utilize Genetic Programming to further reduce sensors sampling rate.
To balance the video quality and bandwidth usage of mobile video chats, we propose LBVC (Low-bandwidth Video Chats), a vibration-aware frame rate adaption framework. LBVC saves bandwidth through reducing frame rate at the sender and interpolates the `missing' frames at the receiver. Additionally, the sender dynamically adapts frame rate with respect to inertial sensor readings in order to keep the scene change between consecutive frames small and prevent strong artifacts from frame interpolation.
Bio
Xin Qi received his B.Sc. degrees in computer
science from Nanjing University, China, in
2007 and M.E. Degree from LIAMA, a joint lab between Chinese Academy of Science and INRIA,
in 2010, respectively. He is currently pursuing
Ph.D. degree in the Department of Computer
Science, the College of William & Mary. His
research interests are mainly in ubiquitous computing
and mobile systems.
Various sensors compose a unique and indispensable component of modern mobile devices. They digitize physical information into sensor readings, with which contexts of mobile users and devices, such as user activities and device motions, can be recognized. Context recognition stimulates the boom of context-aware applications, such as human activity recognition, location-based service and health monitoring.
While context recognition brings context awareness to mobile systems, how to improve context recognition in terms of accuracy, energy and privacy under the constraints of mobile devices, and how to leverage context awareness for mobile resource saving are still challenging problems. Our research examines the former problem by taking human activity as a sample context and with the goal of avoiding trading one metric for another. Our research examines the latter problem by taking energy and bandwidth as sample mobile resources and leveraging context awareness for energy and bandwidth savings. We hope our research will shed light on the design of general context recognition and context-aware resource saving in mobile systems.
To simultaneously achieve high accuracy, low energy and privacy risks for human activity recognition, we propose RadioSense that utilizes wireless communication features instead of sensing-based features. Through benchmarking we find that communication features are the most discriminative when radio power is low. Low radio power also results in a small communication range, which in consequence reduces privacy risks.
To effectively reduce energy overhead as well as maintain high accuracy for human activity recognition, we propose AdaSense, an activity-aware framework that systematically combines lower power activity binary classification that demands lower sensor sampling rate and higher power multi-activity classification that demands higher sensor sampling rate. To amplify the energy saving, we utilize Genetic Programming to further reduce sensors sampling rate.
To balance the video quality and bandwidth usage of mobile video chats, we propose LBVC (Low-bandwidth Video Chats), a vibration-aware frame rate adaption framework. LBVC saves bandwidth through reducing frame rate at the sender and interpolates the `missing' frames at the receiver. Additionally, the sender dynamically adapts frame rate with respect to inertial sensor readings in order to keep the scene change between consecutive frames small and prevent strong artifacts from frame interpolation.
Bio
Xin Qi received his B.Sc. degrees in computer
science from Nanjing University, China, in
2007 and M.E. Degree from LIAMA, a joint lab between Chinese Academy of Science and INRIA,
in 2010, respectively. He is currently pursuing
Ph.D. degree in the Department of Computer
Science, the College of William & Mary. His
research interests are mainly in ubiquitous computing
and mobile systems.
Contact
[[vlthompsondopp]]