Globally, number of mobile devices and connections reached 8 billion recently,
and is projected to surpass 1.5 devices per capita by 2021. These mobile devices are generating enormous amount of data, related to information of the users and the context. However, most of these fine details aren’t being utilised in current mobile systems, due to difficulties associated with exploiting them. For example, some key characteristics of mobile system such as fine-grained spatio-temporal dynamics and user behaviours are not simple patterns that could be easily understood.
In the meantime, data mining techniques and machine learning approaches have been advancing greatly in recent years. Transcending human understandings are achieved in various fields such as image and voice recognition or recommendation systems. Instead of leaving these valuable data sitting on the shelf, there is an urgent need of an intelligent mobile system that is capable of taking these insights into improving the performance itself without much human interactions.
My work bridges data science theories with mobile systems and networks. My research aims to discover the complex fine-grained spatial-temporal patterns, as well as human behaviours when users are “on the move”. We then establish mechanisms to apply these discovered insights through new system architecture, more efficient applications and improved algorithms.
The challenges not only lie in finding the optimal approach of understanding the data, but also integrating those insights into the system. So how did we overcome the challenges?
We first collect real-world fine-grained mobile data in large scale. We then examine and compare several mining or learning techniques that could potentially achieve good results. After that, we mathematically model the system with the improved understanding gained from data and tune different parameters to determine the “weak point” of system.
Here’s what we have done so far.
We exploit users’ spatio-temporal correlation of interests and their mobile viewing behaviours to improve their video streaming experience. Spatio-temporal distribution of mobile users is studied to better provide advertisements that users actually like. Furthermore, users’ behaviours of interacting with mobile device are characterised to improve mobile energy efficiency. Lastly, We attempt to discover people’s evacuation patterns through their mobile connections for better disaster preparation and response.
Mobile system is ubiquitous now, and we believe the services it provides could
always get better.