* Faculty       * Staff       * Contact       * Institute Directory
* Research Groups      
* Undergraduate       * Graduate       * Institute Admissions: Undergraduate | Graduate      
* Events       * Institute Events      
* Lab Manual       * Institute Computing      
No Menu Selected

* Research

Ph.D. Theses

Network Data Modeling Via Grammatical Structures

By Sahin Cem Geyik
Advisor: Boleslaw K. Szymanski
June 4, 2012

Data modeling in computer networks means finding a compact and faithful representation of the data on which network applications work. This data can range from the information that is being transferred in a wireless network to mobility information of nodes in a mobile network, or to the observations of the entity behaviors recorded by the network, etc. In any kind of application, data modeling holds an important place if high efficiency is a requirement. In this thesis, we examine the use of grammatical structures, with a high emphasis on probabilistic context free grammars (PCFG), as the modeling framework for such data. Informally, PCFGs are regular context free grammars where production rules are assigned a probability value, representing how likely these rules are to be used when generating a sentence from this grammar. Utilization of PCFGs includes initially deriving the grammatical structure from the real world traces, and later applying the structure for the necessary purposes of the application. These purposes include saving disk space in space-constrained systems (such as sensor networks), classification and recognition of observed events, facilitating manual inspection (i.e. improving interpretability), etc. The subject matter presented in this thesis contains both the grammar construction process as well as different domains to which the grammatical structure can be applied.

The contributions of this PhD thesis can be summarized as follows:

  • An automated Probabilistic Context Free Grammar (PCFG) construction algorithm, which lowers the time complexity of the previous approaches. We also evaluate (both theoretically and empirically) how much of an advantage we provide over the previous approaches complexity-wise, and the comparison of the goodness of grammars constructed with multiple methods.
  • Application of the PCFG learning and later processing approach to different domains of Network Data Modeling. These domains include:
    • Event Recognition in Sensor Networks,
    • Mobility Modeling and Synthetic Trace Generation for Mobile Networks,
    • Behavior Modeling in Social Networks, and,
    • Learning of Service Composition Rules for Service Oriented Architecture in Sensor Networks.
  • We also contribute to metric-based service composition in sensor networks, as well as to application of switch options in pervasive sensor applications.

* Return to main PhD Theses page