This synthesis project proposes the use of techniques developed in the
field of artificial intelligence to predict behavior of clients in
networked environments with a client-server architecture. Successful 
prediction of user behavior allows the system to utilize its resources 
more efficiently. Predicting future access patterns of clients may
allow the system to fetch some of the data from the server in advance. 
This not only reduces the response time as seen by the client, but it may 
also help to reduce the burst load on the network.

In our project we have chosen to concentrate on World Wide Web (WWW)
based systems, as most of these systems are inherently stateless and 
asynchronous.  This makes the problem of predicting user behavior more 
difficult, while at the same time it promises greater improvements in
performance with successful predictions. As a starting point,
we focus our attention on web-based intelligent tutoring
systems. We believe that such systems will increasingly augment the
traditional classroom model of teaching and learning as the Internet
spreads wider and wider. The access times experienced by the users
while downloading material from these systems will be a critical factor
in their widespread acceptance. In addition, user modeling is an
integral part of many intelligent tutoring systems. We hope to
exploit these models to predict user access patterns.

To test our ideas, we have built a World Wide Web based intelligent tutoring 
system called MANIC (Multimedia Asynchronous Networked Individualized 
CourseWare). The implementation includes a student model server which creates
and maintains the models of individual students using this system. The
server uses these models to predict future access patterns of the student.
This information is used by a proxy to prefetch
documents from the web server in an effort to improve response times
experienced by the users of the system. All the major components
of the system have been implemented and a beta version is
currently being used to teach a 1-credit course in the UMass
Computer Science Department.