The paper introduces a new event-driven state space-based analysis algorithm for hidden non-Markovian models (HnMMs). HnMMs have been developed recently to enable the analysis of hidden discrete stochastic systems based on their observable output, e.g. to determine the unobserved causes of observed defects. There are currently two known approaches for analyzing HnMMs: Proxel-based analysis is generally applicable, but very time consuming and therefore infeasible for most realistic models; the modified Forward solver is very fast, but restricted to models of Markov regenerative type, which is a harsh restriction. The approach presented here bridges the gap between these two algorithms. It adapts the constant time steps of the Proxel algorithm to the time intervals between two output symbols and on the other hand encodes history into the modified Forward solver, thereby eliminating the need for the models to be of Markov regenerative type. However, the event driven Proxel algorithm requires every transition to produce observable output. Performance experiments show that the algorithm can generate a speed-up of up to factor 50 compared to the general Proxel solver for this restricted class of models. This paper extends the range of HnMMs that can be analyzed feasibly. It is another step toward practical feasibility of HnMM analysis, making them more useful for practitioners in the industry.
[1]
Graham Horton,et al.
SOLVING HIDDEN NON-MARKOVIAN MODELS : HOW TO COMPUTE CONDITIONAL STATE CHANGE PROBABILITIES
,
2009
.
[2]
Lawrence R. Rabiner,et al.
A tutorial on hidden Markov models and selected applications in speech recognition
,
1989,
Proc. IEEE.
[3]
Improving the Efficiency of the Proxel Method by using Variable Time Steps
,
2008
.
[4]
Graham Horton,et al.
Using hidden non-Markovian Models to reconstruct system behavior in partially-observable systems
,
2010,
SimuTools.
[5]
Graham Horton,et al.
Improving the Efficiency of the Proxel Method by Using Individual Time Steps
,
2009,
ASMTA.
[6]
Graham Horton.
A NEW PARADIGM FOR THE NUMERICAL SIMULATION OF STOCHASTIC PETRI NETS WITH GENERAL FIRING TIMES
,
2002
.
[7]
Felix Salfner,et al.
Modeling Event-driven Time Series with Generalized Hidden Semi-Markov Models
,
2006
.
[8]
Graham Horton,et al.
The Effect of Rare Events on the Evaluation and Decoding of Hidden non-Markovian Models
,
2008
.
[9]
Sanja Lazarova-Molnar.
The proxel-based method
,
2005
.