Satish Bukkapatnam - 5/25/04
Daniel J. Epstein Industrial and Systems Engineering
University of Southern California

Nonlinear Intrusion Monitoring: A Dynamic Systems Approach

Complexity in TCP/IP networks and several other modern engineering systems is known to emerge from nonlinear stochastic dynamics of network traffic. Large-scale attacks and other pathologies, in general, cause qualitative variations in this dynamics. Features that can capture complexity are therefore imperative for tracking these variations, and thus effectively detecting/predicting various pathologies. Sensor-based modeling research provides a unique approach to realize this imperative. It augments the statistical and qualitative reasoning foundations of the current Intrusion Monitoring technologies with nonlinear dynamic principles. Features extracted using this approach are expected to minimize false-positives and false-negatives, which have been among the main shortcomings of the current technologies. This talk will specifically present an overview of sensor-based modeling research. Two new methods for model synthesis and feature extraction will be introduced. Initial results will be presented to reveal the feasibility of the approach.

Bio
Satish Bukkapatnam is an Assistant Professor of Industrial and Systems Engineering at the Viterbi School and he directs Manufacturing and Controls Laboratory (http://www-nmcl.usc.edu) of USC's Epstein Department. His research interests lie in sensor-based modeling for monitoring complex systems that include manufacturing machines and processes as well as other infrastructure and lifelines systems. His research has yielded over 22 papers (accepted or published) in journals.