Supervised and Unsupervised Learning for Temporal Data Analysis

The Hong Kong University of Science and Technology
Department of Computer Science and Engineering


PhD Thesis Defence


Title: "Supervised and Unsupervised Learning for Temporal Data Analysis"

By

Mr. Fengchao PENG


Abstract

Temporal data analysis has been widely applied in various areas, such as 
bio-informatics, outlier detection, and trajectory mining. These applications 
require a variety of machine learning methods, either supervised or 
unsupervised. In this thesis, we study several supervised and unsupervised 
learning methods suitable for our target applications. The first application is 
to monitor the battery level in wireless communication devices, where we 
develop a time series classification method to identify the working status of 
the devices. The classifier achieves a high accuracy, but incurs heavy 
annotation cost in preparing training data. To solve this problem, we propose 
an efficient and effective active learning method. Specifically, we adapt the 
idea of shapelet discovery and select the training data based on both the 
uncertainty and the utility of each data instance. This method outperforms the 
state-of-the-art active learning methods on time series data. The second 
application is to study the patterns in movement trajectories. In particular, 
we propose an unsupervised method to identify trajectory patterns that match 
well-known team strategies in professional basketball games. Our experimental 
results demonstrate the effectiveness of our proposed method in comparison with 
traditional methods.


Date:			Monday, 28 May 2018

Time:			4:00pm - 6:00pm

Venue:			Room 2132B
 			Lift 19

Chairman:		Prof. Wenjing Ye (MAE)

Committee Members:	Prof. Lionel Ni (Supervisor)
 			Prof. Qiong Luo (Supervisor)
 			Prof. Lei Chen
 			Prof. Yangqiu Song
 			Prof. Ping Gao (CBE)
 			Prof. Jiannong Cao (Comp., PolyU)


**** ALL are Welcome ****