DIMACS Monitoring Message Streams Working Group: Fall 2003

Working Group: Fall 2003

Date	Speaker	Topic
October 27, 4:30pm	Muthu Muthukrishnan, Rutgers CS	Text Streaming
POSTPONED	Diane Lambert, Bell Labs	TBA
December 1, 4:30pm	Rick Mammone, CAIP	Rutgers Warning and Indication System Engineering Laboratory-The WISE Lab
December 18, 2:00pm	Michael Littman, Rutgers CS	Sequential Decision Making Algorithms in Filtering

In the past few years we have developed new algorithmic methods
for dealing with fast streams of mostly numerical, categorical or 
hierarchical data. These were motivated by IP network traffic logs. 

Emerging applications now involve streams of text data. 

How to apply the extant work to these text streams? 
I will explore this question, as interactively as possible

We have been exploring the possibility of applying algorithms from
machine learning and sequential decision making to text filtering.
This is preliminary work and we welcome feedback from the the
"monitoring message stream" community.  To that end, I will provide
background on the Markov decision process (MDP) and partially
observable MDP models and will describe our current efforts in using
these models for filtering.  Our motivation is to automatically make
filtering decisions that trade off exploration and exploitation.

This is joint work with Peng Song, David Madigan, and Paul Kantor

The mission of the WISE lab is to develop and improve data processing
for a variety of situational awareness systems.  Data from various
sensors such as: nuclear, biological, chemical, surveillance sensors
as well as text messages generated by observers in the field are
transported to a central server.  The challenge is to identify the
various sources of the data. For example if the data pertains to a
person then the system should automatically identify the person.  If
the data comes from a material then system should identify the
material, for example: materials used for explosives, nuclear,
biological and chemical weapons.  In addition the system should
indicate the location of the identified person or material.  The WISE
lab has developed digital signal processing and pattern recognition
algorithms to translate the raw sensor data into these important
classes.  Applications of the algorithms in the area of biometrics,
chemometrics, and geometrics will be presented.  In addition a new
Short Messaging Service (SMS) warning system will be described. The
WISE Lab is interested in exploring the question of how text analysis
can help in the overall warning and indicators systems.  For more
information about the WISE Lab see: www.caip.rutgers.edu/wiselab.