This project will provide algorithms and toolkits to empower end users for better application experience when using open-platform smartphones over dynamic wireless access networks. A novel model-based diagnosis engine helps users to understand the access problems as application performance degrades. The cognition-aware feedback completes an execution-evaluation loop for the users to construct correct conceptual perceptions of the troubleshooting process and the root cause. Smartphones with customized software will be distributed to conduct user studies and evaluate the user-interaction effectiveness when dealing with wireless access problems.
The combination of formal causality and state models is leveraged to troubleshoot complex wireless access problems. A unique contribution is to integrate application-level user context with network-level diagnostic models, as user mobility and physical environment are important factors of wireless access performance. Machine-learning algorithms are used for robust context inference using noisy data from on-device sensors. Norman's cognition framework is used for iterative presentation and perception of the troubleshooting process on small-screen smartphones with limited input methods.
This work fills in the gap of providing diagnostic tools to the mobile users. The success of this project will result in new perception tools, which can be disseminated directly to the public through mobile application stores, to significantly enhance the often-misleading signal indicator on smartphones today. An interactive Web-based education portal will be developed to better involve students and online course modules will be introduced through UMass Lowell's Division of Continuing Studies and Corporate Education to reach IT professionals.
With the support of this award, we have studied algorithms and developed toolkits to empower end users for better application experience when using open-platform smartphones over dynamic wireless access networks. We studied several approaches for network troubleshooting, including causality-and-state model, Bayesian networks, and decision trees. Through empirical evaluation, we found decision trees work best for fault diagnosis of mobile networks due to the limitation of mobile client measurements and self-adaptation behavior of cellular systems. The results lead to a prototype implementation of Android application (MobiDiag) available for download through Google Play market. This application provides understandable feedback of the root cause of current perceived network problem to the end users. With this client-observed network data providing user-perceived performance insights, we also expect mobile operators can adopt similar approach for more comprehensive network management solutions. The context-inference framework developed through this project is also used to improve other aspects of the smartphone application experience. We developed AppJoy for personalized application discovery to address the problem of finding interesting applications in a fast-growing app store. Unlike other download-based approaches, AppJoy uses a unique usage-driven recommendation algorithm. We also developed Nihao for predictive application launcher on Android, so the user can quickly find the application they intend to use at any given moment. The results of these studies have been disseminated to public through anonymized dataset and software release. The PI used the software and dataset from this project to develop the materials for a new course "91.650, Topics in Wireless Networks and Mobile Computing" for Computer Science graduate students at UMass Lowell (14 enrollment for Spring 2011), and the dataset from this project was also used for an online course "94.564, Secure Mobile Networks" offered annually through UMass Lowell's Division of Continuing Studies and Corporate Education (28 enrollment for Fall 2013). Through this project, we had the opportunity to directly involving 4 PhD students, 1 master student, and 2 undergraduate students through Research Experiences for Undergraduates (REU) program. This project helped provide to graduate students and undergraduate students the following research skills and experience: (1) How to investigate a problem and formulate mathematical problems; (2) How to devise algorithms and carry out experiments; (3) How to write research papers and make presentations; and 4) how to develop and publish Android applications. The results include 7 peer-reviewed papers, 1 PhD dissertation, 1 Master thesis, 1 dateset for mobile application usage, and 2 Android applications available for public download.
