Dirac-Based Reduction Techniques for Quantitative Analysis of Discrete-Time Markov Models - Topics in Theoretical Computer Science
Conference Papers Year : 2020

Dirac-Based Reduction Techniques for Quantitative Analysis of Discrete-Time Markov Models

Mohammadsadegh Mohagheghi
  • Function : Author
  • PersonId : 1093226
Behrang Chaboki
  • Function : Author
  • PersonId : 1093227

Abstract

Iterative methods are widely used in probabilistic model checking to compute quantitative reachability values. Redundant computation is a significant drawback of iterative methods that affects the performance of probabilistic model checking. In this paper we propose a new approach to avoid redundant computations for reachability analysis of discrete-time Markov processes. Redundant computations can be avoided by considering transitions with Dirac distributions. If two states of a Markov chain are connected with a Dirac transition, the iterative method can postpone the update of the source state until the value of the destination state is computed. Using this approach, we propose two heuristics to improve the performance of iterative methods for computing unbounded reachability probabilities in discrete-time Markov chains and Markov decision processes. The proposed heuristics can be lifted to the computations of expected rewards and have been implemented in PRISM model checker. Several standard case studies have been used and experimental results show that our heuristics outperform most well-known previous iterative methods.
Fichier principal
Vignette du fichier
495613_1_En_1_Chapter.pdf (388.52 Ko) Télécharger le fichier
Origin Files produced by the author(s)

Dates and versions

hal-03165383 , version 1 (10-03-2021)

Licence

Identifiers

Cite

Mohammadsadegh Mohagheghi, Behrang Chaboki. Dirac-Based Reduction Techniques for Quantitative Analysis of Discrete-Time Markov Models. 3rd International Conference on Topics in Theoretical Computer Science (TTCS), Jul 2020, Tehran, Iran. pp.1-16, ⟨10.1007/978-3-030-57852-7_1⟩. ⟨hal-03165383⟩
42 View
31 Download

Altmetric

Share

More