Statistical Multiresolution Analysis of Internet Traffic on Graphs: Good Idea or Wishful Thinking?

Eric Kolaczyk
Boston University
Mathematics and Statistic

Experience has shown that, for certain classes of signals and images, statistical processing techniques based on multiresolution analysis (MRA) can be quite powerful. With respect to the Internet, there are various contexts in which one might think to bring MRA techniques usefully to bear. For the case of Internet traffic data, because it is natural to view such data as indexed with respect to a network topology (i.e., nodes and links), it is tempting to hope that statistical techniques based on appropriate graph-based extensions of MRA might prove to be as powerful as in the more traditional settings of one- and two-dimensional data analysis. Recent work in this area suggests that such hopes may not be unfounded. However, more work is needed to better understand (a) the nature of MRA's on graphs, particularly as influenced by graph topology, (b) how such MRA's in turn interact with different classes of network-based signals, and (c) the manner in which this all plays out in the context of actual Internet traffic data. In this talk, I will discuss some preliminary work aimed at exploring these issues.

Presentation (PDF File)

Back to Workshop I: Multiscale Representation, Analysis and Modeling of Internet Data and Measurements