Bounds for Synchronizing Markov Decision Processes - Archive ouverte HAL Accéder directement au contenu
Communication Dans Un Congrès Année : 2022

Bounds for Synchronizing Markov Decision Processes

Résumé

We consider Markov decision processes with synchronizing objectives, which require that a probability mass of $1-\varepsilon$ accumulates in a designated set of target states, either once, always, infinitely often, or always from some point on, where $\varepsilon = 0$ for sure synchronizing, and $\varepsilon \to 0$ for almost-sure and limit-sure synchronizing. We introduce two new qualitative modes of synchronizing, where the probability mass should be either positive, or bounded away from 0. They can be viewed as dual synchronizing objectives. We present algorithms and tight complexity results for the problem of deciding if a Markov decision process is positive, or bounded synchronizing, and we provide explicit bounds on $\varepsilon$ in all synchronizing modes. In particular, we show that deciding positive and bounded synchronizing always from some point on, is coNP-complete.

Dates et versions

hal-04342391 , version 1 (13-12-2023)

Identifiants

Citer

Laurent Doyen, Marie van den Bogaard. Bounds for Synchronizing Markov Decision Processes. CSR 2022, Jun 2022, Saint-Petersbourg / Virtual, Russia. pp.133-151, ⟨10.1007/978-3-031-09574-0_9⟩. ⟨hal-04342391⟩
24 Consultations
0 Téléchargements

Altmetric

Partager

Gmail Facebook X LinkedIn More