4pm Monday 19th May 2007
Room 2511, JCMB, King's Buildings
Abstract
A silent step in a dynamic system is a step that is considered
unobservable. In this talk I introduce silent steps to Markov reward
chains as steps that have infinite rates and unknown probabilities. I
define and compare two methods of aggregation that abstract away from
these steps while preserving performance measures. The first method is
based on lumping, the second on the elimination of stochastic
discontinuity. The results have a direct application since they can
serve as a correctness criterion for the bisimulation based reductions
of non-determinism in Markovian process algebras.