Feature Engineering for Activity Recognition

PhD Thesis Proposal Defence


Title: "Feature Engineering for Activity Recognition"

by

Mr. Yin ZHU


Abstract:

In today's world, we have increasingly sophisticated means of recording the 
daily activity of humans as well as other moving objects in both physical and 
virtual worlds. These recorded activities include phone calls, uses of Apps on 
smartphones, and expression of opinions over social-network objects such as 
photos. These activities and actions give rise to a huge amount of data.
  Activity recognition aims to understand users’ actions and intentions based on 
models built from these data.
  Having accurate activity recognition allows us to recognition the semantic 
functions of places, to help users and managers identify spammers in an online 
social network, and to predict the future activity levels of social-network 
users.


In this PhD research, we view activity recognition as a form of classification. 
Good discriminative features are essential in building a high-performance 
classifier. Traditionally, feature engineering is performed by humans through 
manual labor. In this thesis, we propose a novel method to automatically or 
semi-automatically extract a large number of features from very high volumes of 
data.  I first propose a semi-automatic feature engineering method that 
explores the relationship between activity signals and context.  By uncovering 
the relationship between actions and their context, we can automatically 
construct new features. We apply this feature engineering method to two 
different applications, semantic place prediction and social activity level 
prediction. We call this method conditional-feature method. We then propose a 
second feature engineering method based on social information, by proposing a 
novel socially regularized feature learning. This method complements the 
above-mentioned conditional-feature method. We apply this socially regularized 
method to two applications: spammer detection in a social network and social 
activity level recognition. We experimentally verify the above 
feature-engineering methods and show that they are superior to existing 
features for activity recognition.


Date:			Tuesday, 1 April 2014

Time:                   2:00pm - 4:00pm

Venue:                  Room 5507
                         lifts 25/26

Committee Members:	Prof. Qiang Yang (Supervisor)
 			Dr. Ke Yi (Chairperson)
 			Dr. Lei Chen
 			Dr. Huamin Qu


**** ALL are Welcome ****