Policy learning in continuous-time Markov decision processes
using Gaussian Processes

J 2017

Policy learning in continuous-time Markov decision processes using Gaussian Processes

BRÁZDIL, Tomáš, Ezio BARTOCCI, Dimitrios MILIOS, Guido SANGUINETTI, Luca BORTOLUSSI et. al.

Základní údaje

Originální název

Policy learning in continuous-time Markov decision processes using Gaussian Processes

Autoři

BRÁZDIL, Tomáš (203 Česká republika, garant, domácí), Ezio BARTOCCI (380 Itálie), Dimitrios MILIOS (300 Řecko), Guido SANGUINETTI (380 Itálie) a Luca BORTOLUSSI (380 Itálie)

Vydání

Performance Evaluation, 2017, 0166-5316

Další údaje

Jazyk

angličtina

Typ výsledku

Článek v odborném periodiku

Obor

10201 Computer sciences, information science, bioinformatics

Stát vydavatele

Nizozemské království

Utajení

není předmětem státního či obchodního tajemství

Odkazy

URL

Impakt faktor

Impact factor: 1.786

Kód RIV

RIV/00216224:14330/17:00107689

Organizační jednotka

Fakulta informatiky

DOI

http://dx.doi.org/10.1016/j.peva.2017.08.007

UT WoS

000413797400005

Klíčová slova anglicky

continuous-time Markov decision processes; reachability; gradient descent

Příznaky

Mezinárodní význam, Recenzováno

Změněno: 28. 4. 2020 07:51, RNDr. Pavel Šmerk, Ph.D.

Anotace

V originále

Continuous-time Markov decision processes provide a very powerful mathematical framework to solve policy-making problems in a wide range of applications, ranging from the control of populations to cyber–physical systems. The key problem to solve for these models is to efficiently compute an optimal policy to control the system in order to maximise the probability of satisfying a set of temporal logic specifications. Here we introduce a novel method based on statistical model checking and an unbiased estimation of a functional gradient in the space of possible policies. Our approach presents several advantages over the classical methods based on discretisation techniques, as it does not assume the a-priori knowledge of a model that can be replaced by a black-box, and does not suffer from state-space explosion. The use of a stochastic moment-based gradient ascent algorithm to guide our search considerably improves the efficiency of learning policies and accelerates the convergence using the momentum term. We demonstrate the strong performance of our approach on two examples of non-linear population models: an epidemiology model with no permanent recovery and a queuing system with non-deterministic choice.

Návaznosti

GA15-17564S, projekt VaV

Název: Teorie her jako prostředek pro formální analýzu a verifikaci počítačových systémů

Investor: Grantová agentura ČR, Teorie her jako prostředek pro formální analýzu a verifikaci počítačových systémů

Citovat

BRÁZDIL, Tomáš, Ezio BARTOCCI, Dimitrios MILIOS, Guido SANGUINETTI a Luca BORTOLUSSI. Policy learning in continuous-time Markov decision processes using Gaussian Processes. Performance Evaluation. 2017, roč. 116, č. 1, s. 84-100. ISSN 0166-5316. Dostupné z: https://dx.doi.org/10.1016/j.peva.2017.08.007.

@article{1564396,
   author = {Brázdil, Tomáš and Bartocci, Ezio and Milios, Dimitrios and Sanguinetti, Guido and Bortolussi, Luca},
   article_number = {1},
   doi = {http://dx.doi.org/10.1016/j.peva.2017.08.007},
   keywords = {continuous-time Markov decision processes; reachability; gradient descent},
   language = {eng},
   issn = {0166-5316},
   journal = {Performance Evaluation},
   title = {Policy learning in continuous-time Markov decision processes using Gaussian Processes},
   url = {http://dx.doi.org/10.1016/j.peva.2017.08.007},
   volume = {116},
   year = {2017}
}

TY  - JOUR
ID  - 1564396
AU  - Brázdil, Tomáš - Bartocci, Ezio - Milios, Dimitrios - Sanguinetti, Guido - Bortolussi, Luca
PY  - 2017
TI  - Policy learning in continuous-time Markov decision processes using Gaussian Processes
JF  - Performance Evaluation
VL  - 116
IS  - 1
SP  - 84-100
EP  - 84-100
SN  - 01665316
KW  - continuous-time Markov decision processes
KW  - reachability
KW  - gradient descent
UR  - http://dx.doi.org/10.1016/j.peva.2017.08.007
L2  - http://dx.doi.org/10.1016/j.peva.2017.08.007
N2  - Continuous-time Markov decision processes provide a very powerful mathematical framework to solve policy-making problems in a wide range of applications, ranging from the control of populations to cyber–physical systems. The key problem to solve for these models is to efficiently compute an optimal policy to control the system in order to maximise the probability of satisfying a set of temporal logic specifications. Here we introduce a novel method based on statistical model checking and an unbiased estimation of a functional gradient in the space of possible policies. Our approach presents several advantages over the classical methods based on discretisation techniques, as it does not assume the a-priori knowledge of a model that can be replaced by a black-box, and does not suffer from state-space explosion. The use of a stochastic moment-based gradient ascent algorithm to guide our search considerably improves the efficiency of learning policies and accelerates the convergence using the momentum term. We demonstrate the strong performance of our approach on two examples of non-linear population models: an epidemiology model with no permanent recovery and a queuing system with non-deterministic choice.
ER  -

BRÁZDIL, Tomáš, Ezio BARTOCCI, Dimitrios MILIOS, Guido SANGUINETTI a Luca BORTOLUSSI. Policy learning in continuous-time Markov decision processes using Gaussian Processes. \textit{Performance Evaluation}. 2017, roč.~116, č.~1, s.~84-100. ISSN~0166-5316. Dostupné z: https://dx.doi.org/10.1016/j.peva.2017.08.007.

Podrobný výpis o publikaci