Usage Models

Work on a more aggressive reliability estimator can be found here.

If you are interested in approximating the number of tests to achieve a particular reliability, Click Here....

If you have been granted access to work on the new edition of the report, please Click Here....

The original technical report on computations, Computations for Markov Chain Usage Models, has become a bit dated. In particular, it omits the following.

  • The new reliability model published in Modern Statistical and Mathematical Methods in Reliability and implemented in the JUMBL and elsewhere.
  • The fast analysis methods described in patent 7,219,049 for near-linear analysis of large models.

The other key point is that modern usage models tend to be analyzed as stochastic grammars, and not so much as Markov chains. That is, the statistics for the labels on arcs are more significant than the statistics for the states or the arcs, in many applications. This needs to be properly discussed in the text.

The Scilab code needs to be updated, and placed in a more comprehensive library package. This requires a Scilab expert, or someone wishing to become a Scilab expert, to help with this. Most significantly, it would be good to have a Scilab-hosted parser for TML, so that models written in TML could be directly analyzed in Scilab.

The exposition about these topics needs to be improved, and a set of exercises should be included to make the report into a proper textbook on statistical testing using the techniques described.

More generally, there are other items of interest:

  • The use, analysis, and representation of continuous variables in otherwise discrete models.
  • The effect of stateful transitions on the analysis.