Algorithmic Issues in Monitoring and Mining Network Data Streams

Muthu Muthukrishnan
Rutgers University

Telecommunication networks generate massive streams of traffic logs. Monitoring, auditing and mining such streams is an integral part of network management. Research over the past few years in this area has lead to certain key insights. It is difficult to find a needle in the haystack within the constraints faced in the network application. Fortunately, several interesting network traffic phenomena have the ``few good terms'' property, i.e., summarizing a few trends in the traffic---deviants, wavelet terms, histograms, correlations---is both feasible and insightful. Finally, data quality problems in network datafeeds present many new challenges and a principled approach to confronting data quality problems is an implicit concern for data stream mining. In this talk, I will survey the application, discuss these insights and present novel algorithmic solutions that have been developed.


S. (Muthu) Muthukrishnan graduated from the Courant Institute of Mathematical Sciences in 1994 with thesis work on two person probabilistic games and pattern matching, and continued with research in many different topics and multiple organizations: (Computational Biology, DIMACS), (Load Balancing, Univ. of Warwick @ UK), (Databases and Scheduling, Bell Labs), (Wireless Systems and IP Networks, AT&T Research), and (Wireless Technology for Social Networks, Rutgers Univ). More tuples should follow!

His work on building nation-wide location aware services for AT&T Wireless was covered in MSNBC and CBS, and he resented it at the National Academy of Engineering meeting in 2002. He has been on PCs of conferences and organized workshops and special programs in theoretical computer science, discrete mathematics, networking and databases. His personal agenda is to be a Scientist, Engineer and Mathematician, all in one.

Return to Program