Adaptive distributed resource allocation using cooperative information-sharing
Adaptive distributed resource allocation using cooperative information-sharing
In this thesis, we develop a multi-agent information-sharing protocol to efficiently solve sequential RA problems. This protocol lets the agents cooperate with one another, by sharing information, to build reliable estimates of the system states which, in turn, generate effective adaptive behaviour. We use Q-learning, an algorithm for generating robust estimates without any prior knowledge of the environment dynamics, as the tool for estimate learning using the shared information. In so doing, our research significantly extends the state-of-the-art in cooperative MAS technology for sequential RA problems.
First, our information-sharing protocol — the post task-completion (PTC) protocol — is novel in its principle of application and is also the most effective information-sharing method for generating high-quality estimates in dynamic distributed sequential RA problems. We establish its merits using theoretical analyses. Second, we design communication heuristics based on PTC that can be used in a real sequential RA application: call routing in mesh networks. Using empirical analyses, we show that these heuristics outperform a host of benchmark algorithms used in this application domain. Third, we improve the adaptiveness of PTC in highly dynamic environments by extending the communication heuristic. Empirical analyses demonstrate that this extended protocol achieves superior responsiveness to state changes caused by network failures and successfully diagnoses failures. Finally, we extend the effectiveness of the extended PTC heuristic by using selective information-sharing. We show that sharing information using a notion of information redundancy significantly reduces the communication overhead of PTC but without sacrificing its performance advantages.
In achieving the above-mentioned contributions, we successfully establish the validity of our claim about cooperative information-sharing being able to generate effective state estimation and adaptive behaviour required for efficiently solving dynamic distributed sequential RA problems. In so doing, we also contribute towards extending the current technology for distributed network call routing by providing a highly efficient communication protocol that generates high quality solutions for the routing problem.
University of Southampton
Dutta, Partha Sarathi
fadbfe53-0016-41f2-a751-71852a10dffb
2005
Dutta, Partha Sarathi
fadbfe53-0016-41f2-a751-71852a10dffb
Dutta, Partha Sarathi
(2005)
Adaptive distributed resource allocation using cooperative information-sharing.
University of Southampton, Doctoral Thesis.
Record type:
Thesis
(Doctoral)
Abstract
In this thesis, we develop a multi-agent information-sharing protocol to efficiently solve sequential RA problems. This protocol lets the agents cooperate with one another, by sharing information, to build reliable estimates of the system states which, in turn, generate effective adaptive behaviour. We use Q-learning, an algorithm for generating robust estimates without any prior knowledge of the environment dynamics, as the tool for estimate learning using the shared information. In so doing, our research significantly extends the state-of-the-art in cooperative MAS technology for sequential RA problems.
First, our information-sharing protocol — the post task-completion (PTC) protocol — is novel in its principle of application and is also the most effective information-sharing method for generating high-quality estimates in dynamic distributed sequential RA problems. We establish its merits using theoretical analyses. Second, we design communication heuristics based on PTC that can be used in a real sequential RA application: call routing in mesh networks. Using empirical analyses, we show that these heuristics outperform a host of benchmark algorithms used in this application domain. Third, we improve the adaptiveness of PTC in highly dynamic environments by extending the communication heuristic. Empirical analyses demonstrate that this extended protocol achieves superior responsiveness to state changes caused by network failures and successfully diagnoses failures. Finally, we extend the effectiveness of the extended PTC heuristic by using selective information-sharing. We show that sharing information using a notion of information redundancy significantly reduces the communication overhead of PTC but without sacrificing its performance advantages.
In achieving the above-mentioned contributions, we successfully establish the validity of our claim about cooperative information-sharing being able to generate effective state estimation and adaptive behaviour required for efficiently solving dynamic distributed sequential RA problems. In so doing, we also contribute towards extending the current technology for distributed network call routing by providing a highly efficient communication protocol that generates high quality solutions for the routing problem.
Text
992371.pdf
- Version of Record
More information
Published date: 2005
Identifiers
Local EPrints ID: 465758
URI: http://eprints.soton.ac.uk/id/eprint/465758
PURE UUID: a3c7b7f2-3a45-4700-a8f7-4d3e5b8feac1
Catalogue record
Date deposited: 05 Jul 2022 02:53
Last modified: 16 Mar 2024 20:21
Export record
Contributors
Author:
Partha Sarathi Dutta
Download statistics
Downloads from ePrints over the past year. Other digital versions may also be available to download e.g. from the publisher's website.
View more statistics