Smart device-based context and activity recognition
Updated March 2015
Interested to pursue a Master of Science by Research degree (validated by Lancaster University), or PhD degree at Sunway University, investigating the below research topic? Email me at sianlunl-at-sunway.edu.my (replace -at- with @).
With the rise of new technologies, particularly mobile devices, wearables and almost-ubiquitous connectivity, we are now swamp with exciting techniques that aim to help improve and automate many tasks in our everyday living. One particular area that looks into ways to enable machines to understand situations, needs and human better is context-awareness. Since the introduction of Ubiquitous Computing (or Pervasive Computing), researchers have been investigating various approaches that put computers into the background. As Mark Weiser proposed, the best technologies are those who stays in the background.
This is where acquisition of implicit information may contribute well - imagine the different devices around us, may it be computers in a room, or the smartphone you are carrying to even little sensors placed "everywhere" - they provide all sort of information in almost real time. By applying some clever mechanism, or what we technically call algorithms, we build systems that will make sense out of these seemingly "senseless" bulks of information. This is where context acquisition can play an important role. Starting from a few years back, this field of study has a new sexy name - big data. Though big data covers a much larger scope, but the gist of the approaches is still the same - how to make sense out of no(n)-sense.
One of my research interests focuses in acquiring contexts through unobtrusive manners. I am particularly interested to apply and adapt suitable machine learning/data mining techniques to interpret useful contexts from sensor data. One important requirement I place in such approaches is unobtrusiveness - I argue we should always place the users' needs and preferences first, and an obtrusive technique is something we should always try our best to avoid. This requirement led me to investigate suitable techniques that utilise data acquired from commercially-off-the-shelve (COTS) devices, such as smartphone, commercially available sensors and non-expensive devices.
Currently, I continue to work with colleagues, students and collaborators on smartphone-based activity recognition. An addition to this is the investigations on new applications based on COTS BCI devices. I have at the moment an Emotiv EEG headset and a Melon headband - both are capable of measuring EEG raw signals. I believe in the future we will have more of these devices and if we can come out with innovative and efficient techniques to interpret useful contexts out of these sensor devices, we can make computer systems more intuitive and useful by making them "understand" what contexts the users have and hence what adaptation and services would be required based on these contexts.