Reasoning and learning under uncertainty using dynamic probabilistic models for real time problem determination
Abstract
This thesis investigates the use Dynamic Bayesian Networks for the purpose of real time diagnosis and problem determination in distributed systems. Basically, the increasing complexity and importance of distributed computer networks have given rise to a steadily high demand for advanced network fault management that allow real-time fault localization and accurate problem diagnosis. Due to their ability to handle uncertainty and represent cause and effect relationships, Bayesian Belief Networks (BBN) is one of the state of the art approaches that can be used as a framework for fault diagnosis in distributed computer systems. However, current approaches to diagnosis using Bayesian Networks assume a static model of the system which does not account for failure/repair dynamics (i.e., the system state is assumed to be static during diagnosis process). In highly dynamic systems, this is not the case. There is a need to expand a static Bayesian Network model into a dynamic Bayesian Network model, in order to model situations where the node states change over time. Another limitation of current approaches is that dependency information about component states and diagnostic test outcomes is assumed to be given prior to diagnosis, while in the reality it is either initially unknown, or changing due to addition/deletion of components, dynamic routing, and other non-stationarities in the network state and behavior. Our work addresses these challenges and advance previous work by considering the issue of learning (or updating) a Bayesian network given data in order to make models adaptive. We discuss a problem determination approach that adapts to system dynamics via learning parameter of a Dynamic Bayesian Network. In summary, our work improves state-of-the-art approaches to real-time problem determination by handling uncertainties that result from missing information and dynamic environment. These features can be incorporated in systems management technologies to increase reliability and service level.
Subject Area
Computer science
Recommended Citation
Hernandez, Karina J, "Reasoning and learning under uncertainty using dynamic probabilistic models for real time problem determination" (2004). ETD Collection for Pace University. AAI3145958.
https://digitalcommons.pace.edu/dissertations/AAI3145958
Remote User: Click Here to Login (must have Pace University remote login ID and password. Once logged in, click on the View More link above)