Stop Gambling! It Just Takes Too Long

Publikation: Beitrag in Buch/Konferenzbericht/Sammelband/GutachtenBeitrag in Buch/Sammelband/GutachtenBeigetragenBegutachtung

Abstract

The classical stochastic shortest path problem (SSPP) on integer-weighted Markov decision processes asks for a scheduler that maximizes the expected accumulated weight before reaching a goal state while ensuring that the goal is reached almost surely. In this article, we introduce the positively almost-surely terminating stochastic shortest path problem (PAST-SSPP), a variant of the classical SSPP, where the additional requirement is made that the expected number of steps for reaching the goal has to be finite. We show that PAST-SSPP can be solved in polynomial time. To this end, we provide an extension of the well-known spider construction as introduced by Christel Baier et al. used for the solution of the classical SSPP.

Details

OriginalspracheEnglisch
TitelPrinciples of Formal Quantitative Analysis
Redakteure/-innenNathalie Bertrand, Clemens Dubslaff, Sascha Klüppelholz
Herausgeber (Verlag)Springer, Cham
Seiten139-157
Seitenumfang19
ISBN (elektronisch)978-3-031-97439-7
ISBN (Print)978-3-031-97438-0
PublikationsstatusElektronische Veröffentlichung vor Drucklegung - 30 Aug. 2025
Peer-Review-StatusJa

Publikationsreihe

ReiheLecture notes in computer science
BandLNCS 15760
ISSN0302-9743

Externe IDs

ORCID /0000-0003-4829-0476/work/194824700
ORCID /0000-0003-1724-2586/work/194825125

Schlagworte

Schlagwörter

  • almost-sure termination, Markov decision processes, optional stopping theorem, Stochastic shortest path problem