Title:   Autonomic Data Integration Systems
Speaker: AnHai Doan (from UIUC)

Abstract:
The world today is a vast information bazaar, with millions of sources
providing data in every imaginable format and mode of interaction. Data
integration systems hold the promise of acting as crucial middlemen in
this chaotic market, by interacting with data sources, translating, and
combining their data in order to obtain the information requested by
users. However, today such systems are still very hard to build and costly
to operate. They must be told in tedious detail how to interact with data
sources, and must be constantly modified to deal with changes at the
sources.

In this talk I will describe the AIDA project whose vision is autonomic
data integration systems: those that take only minutes to be deployed
(instead of weeks or months as is the case today), that require only
minimal human coaching to rapidly reach and maintain competence, and that
continuously improve over time, in terms of both performance and
capabilities. I will discuss some fundamental issues that arise -- such
as schema reconciliation, object matching and fusion, and source schema
construction.  I will also describe a conceptually novel approach to
building autonomic systems on the Web: that of mass collaboration.
I show how machine learning techniques can be extended to deal effectively
with these problems.

Finally, I discuss how the ideas developed in the context of this project
can be applied to extracting information from text, marking up data on the
Semantic Web, and constructing the "Capitalist" World-Wide Web.