[Colloq] REMINDER: Thesis Defense, Huanmei Wu, Thursday, 10 AM

Rachel Kalweit rachelb at ccs.neu.edu
Wed Jun 29 15:39:14 EDT 2005


The College of Computer and Information Science Presents:
Ph.D. Defense for Huanmei Wu

Title: Structured Time Series Stream Data

Date: Thursday, June 30, 2005, 10:00am -12:00pm
Location: 366 WVH (CCIS Conference room)

Abstract:
Management of structured time series stream data is an important 
technique for databases, networks, operating systems, and theoretical 
research.  Analysis of structured time series stream data is widely used 
for many applications such as economic forecasting, stock market 
analysis, military intelligence and satellite data tracking. This 
dissertation introduces our research on modeling, analysis and 
prediction of structured time series stream data. Unlike previous work, 
this research considers the internal structure of time series stream data.

These tasks are difficult to achieve because of the constant changes,
sheer volume, and complex structure of stream data. Our objectives
include: (i) online piecewise linear modeling of structured time series
streams, (ii) analyzing and characterizing structured time series stream
data using statistical and probabilistic approaches, and (iii) 
developing real-time prediction methods based on the online modeling 
results of the corresponding internal structure.

We have developed an integrated structured stream data (SSD) framework 
to achieve these objectives. The framework is composed of (a) a 
piecewise linear model (PLM) which captures the internal structure of 
time series stream data, (b) a finite state machine which decompose the 
incoming data into line segments according to the underlying PLM,
(c) a set of analytical and probabilistic methods for data analysis,
including online subsequence matching which generates dynamic query
subsequences, defines new application-specific subsequence similarity
measures and performs similarity matching with consideration of the
internal structure of the data streams and (d) a model-based 
probabilistic prediction method for exact future data value prediction 
and data property forecasting.

This framework covers general application domains with structured time
series stream data. The techniques are very useful for real-time systems
where response time is critical. We have applied the framework to 
multiple problem domains, such as financial data analysis and tumor 
respiratory motion characterization (used in cancer radiation treatment).

Committee:
Prof. Betty Salzberg, NU (advisor)
Prof. David Kaeli, NU (co-advisor)
Prof. Javed Aslam, NU
Prof. Steve Jiang, MGH

Refreshments will be served.






More information about the Colloq mailing list