Detecting Malicious Activity in Wireless Sensor Networks using Topic Modeling with Latent Dirichlet Allocation

Abstract:

This work introduces the concept topic modeling utilizing latent Dirichlet allocation (LDA) as a particular approach for detecting malicious activity in wireless sensor networks (WSN). A brief summary of the theory and derivation of the underlying mathematics for LDA is provided for general applications in topic modeling. One insightful real-world application is demonstrated to showcase the extensibility and utility of the LDA algorithm for use in detecting vulnerabilities in computer systems.

Biography:

Joseph Natarian (Member, IEEE) received the B.Sc. degree in electrical engineering and computer science from the Wright State University, Dayton, Ohio, in 2008.  He is currently pursuing the M.Sc in electrical engineering from the University of Dayton, Dayton, Ohio. Since 2007 he has been supporting the Air Force Research Laboratory (AFRL) in Dayton, Ohio. From 2007 to 2008, he worked for General Dynamics, Advanced Information Systems, where he supported multiple research projects in the Collaborative Interfaces Branch of the Warfighter Interface Division within AFRL. In 2008 he joined the Civil Service as a member of the in-house research team in the Distributed Collaborative Sensor System Technology Branch of the Autonomic Trusted Sensing for Persistent Intelligence Technology Office within AFRL, as an research engineer. As a research egineer, he has served as a subject matter expert and technical advisor for multiple Defense Adnaved Research Project Agency (DARPA) research programs from the Information Innovation Office (I2O) . His research interests include trust in complex systems, tools to evaluate system security, and techniques for identifying and traversing threat vectors of complex systems.  Currently Mr. Natarian is a program manager in the advanced programs division of the Sensors Directorate at AFRL.