Exploiting locality of interaction in networked distributed POMDPs

Yoonheui Kim, Ranjit Nair, Pradeep Varakantham, Milind Tambe, Makoto Yokoo

Research output: Chapter in Book/Report/Conference proceedingConference contribution

14 Citations (Scopus)

Abstract

In many real-world multiagent applications such as distributed sensor nets, a network of agents is formed based on each agent's limited interactions with a small number of neighbors. While distributed POMDPs capture the real-world uncertainty in multiagent domains, they fail to exploit such locality of interaction. Distributed constraint optimization (DCOP) captures the locality of interaction but fails to capture planning under uncertainty. In previous work, we presented a model synthesized from distributed POMDPs and DCOPs, called Networked Distributed POMDPs (ND-POMDPs). Also, we presented LID-JESP (locally interacting distributed joint equilibrium-based search for policies: a distributed policy generation algorithm based on DBA (distributed breakout algorithm). In this paper, we present a stochastic variation of the LID-JESP that is based on DSA (distributed stochastic algorithm) that allows neighboring agents to change their policies in the same cycle. Through detailed experiments, we show how this can result in specdups without a large difference in solution quality. We also introduce a technique called hyper-link-based decomposition that allows us to exploit locality of interaction further, resulting in faster run times for both LID-JESP and its stochastic variant without any loss in solution quality.

Original languageEnglish
Title of host publicationDistributed Plan and Schedule Management - Papers from the AAAI Spring Symposium, Technical Report
Pages41-48
Number of pages8
Publication statusPublished - Aug 21 2006
Event2006 AAAI Spring Symposium - Stanford, CA, United States
Duration: Mar 27 2006Mar 29 2006

Publication series

NameAAAI Spring Symposium - Technical Report
VolumeSS-06-04

Other

Other2006 AAAI Spring Symposium
CountryUnited States
CityStanford, CA
Period3/27/063/29/06

Fingerprint

Parallel algorithms
Decomposition
Planning
Sensors
Experiments
Uncertainty

All Science Journal Classification (ASJC) codes

  • Engineering(all)

Cite this

Kim, Y., Nair, R., Varakantham, P., Tambe, M., & Yokoo, M. (2006). Exploiting locality of interaction in networked distributed POMDPs. In Distributed Plan and Schedule Management - Papers from the AAAI Spring Symposium, Technical Report (pp. 41-48). (AAAI Spring Symposium - Technical Report; Vol. SS-06-04).

Exploiting locality of interaction in networked distributed POMDPs. / Kim, Yoonheui; Nair, Ranjit; Varakantham, Pradeep; Tambe, Milind; Yokoo, Makoto.

Distributed Plan and Schedule Management - Papers from the AAAI Spring Symposium, Technical Report. 2006. p. 41-48 (AAAI Spring Symposium - Technical Report; Vol. SS-06-04).

Research output: Chapter in Book/Report/Conference proceedingConference contribution

Kim, Y, Nair, R, Varakantham, P, Tambe, M & Yokoo, M 2006, Exploiting locality of interaction in networked distributed POMDPs. in Distributed Plan and Schedule Management - Papers from the AAAI Spring Symposium, Technical Report. AAAI Spring Symposium - Technical Report, vol. SS-06-04, pp. 41-48, 2006 AAAI Spring Symposium, Stanford, CA, United States, 3/27/06.
Kim Y, Nair R, Varakantham P, Tambe M, Yokoo M. Exploiting locality of interaction in networked distributed POMDPs. In Distributed Plan and Schedule Management - Papers from the AAAI Spring Symposium, Technical Report. 2006. p. 41-48. (AAAI Spring Symposium - Technical Report).
Kim, Yoonheui ; Nair, Ranjit ; Varakantham, Pradeep ; Tambe, Milind ; Yokoo, Makoto. / Exploiting locality of interaction in networked distributed POMDPs. Distributed Plan and Schedule Management - Papers from the AAAI Spring Symposium, Technical Report. 2006. pp. 41-48 (AAAI Spring Symposium - Technical Report).
@inproceedings{237cf72d033b4259979f3d12a04dee7b,
title = "Exploiting locality of interaction in networked distributed POMDPs",
abstract = "In many real-world multiagent applications such as distributed sensor nets, a network of agents is formed based on each agent's limited interactions with a small number of neighbors. While distributed POMDPs capture the real-world uncertainty in multiagent domains, they fail to exploit such locality of interaction. Distributed constraint optimization (DCOP) captures the locality of interaction but fails to capture planning under uncertainty. In previous work, we presented a model synthesized from distributed POMDPs and DCOPs, called Networked Distributed POMDPs (ND-POMDPs). Also, we presented LID-JESP (locally interacting distributed joint equilibrium-based search for policies: a distributed policy generation algorithm based on DBA (distributed breakout algorithm). In this paper, we present a stochastic variation of the LID-JESP that is based on DSA (distributed stochastic algorithm) that allows neighboring agents to change their policies in the same cycle. Through detailed experiments, we show how this can result in specdups without a large difference in solution quality. We also introduce a technique called hyper-link-based decomposition that allows us to exploit locality of interaction further, resulting in faster run times for both LID-JESP and its stochastic variant without any loss in solution quality.",
author = "Yoonheui Kim and Ranjit Nair and Pradeep Varakantham and Milind Tambe and Makoto Yokoo",
year = "2006",
month = "8",
day = "21",
language = "English",
isbn = "1577352653",
series = "AAAI Spring Symposium - Technical Report",
pages = "41--48",
booktitle = "Distributed Plan and Schedule Management - Papers from the AAAI Spring Symposium, Technical Report",

}

TY - GEN

T1 - Exploiting locality of interaction in networked distributed POMDPs

AU - Kim, Yoonheui

AU - Nair, Ranjit

AU - Varakantham, Pradeep

AU - Tambe, Milind

AU - Yokoo, Makoto

PY - 2006/8/21

Y1 - 2006/8/21

N2 - In many real-world multiagent applications such as distributed sensor nets, a network of agents is formed based on each agent's limited interactions with a small number of neighbors. While distributed POMDPs capture the real-world uncertainty in multiagent domains, they fail to exploit such locality of interaction. Distributed constraint optimization (DCOP) captures the locality of interaction but fails to capture planning under uncertainty. In previous work, we presented a model synthesized from distributed POMDPs and DCOPs, called Networked Distributed POMDPs (ND-POMDPs). Also, we presented LID-JESP (locally interacting distributed joint equilibrium-based search for policies: a distributed policy generation algorithm based on DBA (distributed breakout algorithm). In this paper, we present a stochastic variation of the LID-JESP that is based on DSA (distributed stochastic algorithm) that allows neighboring agents to change their policies in the same cycle. Through detailed experiments, we show how this can result in specdups without a large difference in solution quality. We also introduce a technique called hyper-link-based decomposition that allows us to exploit locality of interaction further, resulting in faster run times for both LID-JESP and its stochastic variant without any loss in solution quality.

AB - In many real-world multiagent applications such as distributed sensor nets, a network of agents is formed based on each agent's limited interactions with a small number of neighbors. While distributed POMDPs capture the real-world uncertainty in multiagent domains, they fail to exploit such locality of interaction. Distributed constraint optimization (DCOP) captures the locality of interaction but fails to capture planning under uncertainty. In previous work, we presented a model synthesized from distributed POMDPs and DCOPs, called Networked Distributed POMDPs (ND-POMDPs). Also, we presented LID-JESP (locally interacting distributed joint equilibrium-based search for policies: a distributed policy generation algorithm based on DBA (distributed breakout algorithm). In this paper, we present a stochastic variation of the LID-JESP that is based on DSA (distributed stochastic algorithm) that allows neighboring agents to change their policies in the same cycle. Through detailed experiments, we show how this can result in specdups without a large difference in solution quality. We also introduce a technique called hyper-link-based decomposition that allows us to exploit locality of interaction further, resulting in faster run times for both LID-JESP and its stochastic variant without any loss in solution quality.

UR - http://www.scopus.com/inward/record.url?scp=33747180593&partnerID=8YFLogxK

UR - http://www.scopus.com/inward/citedby.url?scp=33747180593&partnerID=8YFLogxK

M3 - Conference contribution

AN - SCOPUS:33747180593

SN - 1577352653

SN - 9781577352655

T3 - AAAI Spring Symposium - Technical Report

SP - 41

EP - 48

BT - Distributed Plan and Schedule Management - Papers from the AAAI Spring Symposium, Technical Report

ER -