Stochastic Systems Group  

The context of my talk is fault diagnosis and alarm handling in telecommunication networks. I take a model based approach, in which I derive the diagnosis algorithm automatically from a model of the system, at a certain level of abstraction. Main difficulties are the following:
I shall mainly discuss how to handle the first difficulty, and give hints for the other ones. Petri Nets will serve as a mathematical framework for distributed, discrete event, systems with local state and partially ordered time. PN will be made probabilistic in a suitable way. This will allow me to carry on the techniques of HMM (Hidden Markov Model) to the present case. I'll develop an appropriate version of the Viterbi algorithm for this, and I'll discuss its distributed implementation.
Finally, as modelling is not coming by itself out of the blue, I'll discuss a practical methodology for it, based on UML modelling of the network management system.
This is a joint research project with Alcatel, France Telecom CNET, and the company ILOG.
Problems with this site should be emailed to jonesb@mit.edu