Fast Algorithms for Mining and Summarizing Co-evolving Sequences

Speaker:	Lei LI
		Computer Science Department
		Carnegie Mellon University

Title:		"Fast Algorithms for Mining and Summarizing
		 Co-evolving Sequences"

Date:		Thursday, 17 December 2009

Time:		11:00am - 12:00noon

Venue:		Room 3501 (near lifts 25/26), HKUST


Time series data arise in numerous applications, such as motion capture,
computer network monitoring, data center monitoring, environmental
monitoring and many more. Finding patterns in such collections of
sequences is crucial for leveraging them to solve real-world, domain
specific problems, for example, to build humanoid robots, to detect
pollution in drinking water, and to identify intrusion in computer

In this talk, we focus on fast algorithms on mining co-evolving time
series, with or without missing values. We will present two pieces of our
current work: time series mining and summarization with missing values,
and a parallel learning algorithm for the underlying model, Linear
Dynamical Systems (LDS). Algorithms proposed in these works allow us to
obtain meaningful patterns effectively and efficiently. Subsequently, they
allow us to perform various mining tasks including forecasting,
compression, and segmentation for co-evolving time series, even with
missing values. We also proposed a parallel learning algorithm for LDS to
fully utilize the power of multicore/multiprocessors, which will serve as
a corner stone of many applications and algorithms for time series. Our
algorithms scale linearly with respect to the length of sequences, and
outperform the competitors often by large factors.


Lei Li is a Ph.D. candidate of Computer Science Department at Carnegie
Mellon University. His research interests include machine learning and
data mining, particularly for time series, with applications in
environmental monitoring, motion capture, computer network, and bio-image
databases. He received a bachelor's degree in Computer Science (ACM class)
from Shanghai Jiao Tong University in 2006.