-----------------------------------------------------------------------
                   B O S T O N   U N I V E R S I T Y
                      Computer Science Department
                          C O L L O Q U I U M
                      Wednesday, February 20, 11:00 AM 
                       (Coffee served at 10:45 AM)
                         Seminar Room / MCS 135



    Challenges in the Practical Application of Unsupervised Learning

                        Prof. Carla E. Brodley
           School of Electrical and Computer Engineering
                          Purdue University

                              Abstract

In this talk I will discuss the factors that impact the successful
application of unsupervised machine learning algorithms to real-world
problems.  In particular, the talk will focus on applications that
require finding the natural clusters (groups) of the data.  The first
application is content-based image retrieval of medical images.  The
goal is to create a system that given a query image retrieves visually
similar images.  We are investigating this technology within the domain
of high resolution CT images of the lungs.  A benefit of this system is
that given an image for which the diagnosis is unknown, the physician
can utilize the diagnostic and treatment information in the retrieved
images to help treat the query image patient. In a recent evaluation
trial our system doubled diagnostic accuracy. The machine learning task
in this domain is to find the features that best characterize each image
in order to build an automated retrieval system.  The second application
is to create automatic tools that will detect anomalous behavior in
computer network elements and determine whether it is the result of a
network denial of service (NDoS) attack. Our approach to anomaly
detection is to form a model of the normal behavior of a network element
and then monitor incoming (or outgoing) traffic for anomalies.  We have
placed monitoring hardware/software into both a Web and a file server at
Purdue and present preliminary results for several types of NDoS
attacks. The machine learning task that we address is how to form an
accurate, concise model that can detect anomalous behavior in real time.

Bio

Carla E. Brodley is an associate professor in the School of Electrical
and Computer Engineering at Purdue University.  She received her
bachelors degree from McGill University in 1985 and her PhD in computer
science from the University of Massachusetts at Amherst in 1994.  Her
research interests include machine learning and computer security.  In
2000 she co-chaired the KDD-CUP and in 2001 she served as program
co-chair for the 18th International Conference on Machine Learning.  She
is currently an associate editor for IEEE Transactions on Pattern
Analysis and Machine Intelligence and for the Journal of Artificial
Intelligence Research.

Host: Azer Bestavros
-------------------------------------------------------------------------
For colloquium info, including directions, see http://cs-www.bu.edu/colloquium
-------------------------------------------------------------------------