An Evolutionary Framework for Modelling Unknown Behaviours of Other Agents

Multiagent decision making becomes complicated when a subject agent interacts with other agents who could be either collaborative or competitive in their common environment. Generally the subject agent does not know what exact behaviours other agents will execute in the coming interactions. This occ...

Celý popis

Uloženo v:
Podrobná bibliografie
Vydáno v:IEEE transactions on emerging topics in computational intelligence Ročník 7; číslo 4; s. 1 - 14
Hlavní autoři: Pan, Yinghui, Ma, Biyang, Zeng, Yifeng, Tang, Jing, Zeng, Buxin, Ming, Zhong
Médium: Journal Article
Jazyk:angličtina
Vydáno: Piscataway IEEE 01.08.2023
The Institute of Electrical and Electronics Engineers, Inc. (IEEE)
Témata:
Collaboration
Computational modeling
Decision making
decision model
Evolutionary behaviour
genetic algorithm
Genetic algorithms
Heuristic algorithms
intelligent agents
Mathematical models
Modelling
Multi-agent systems
Multiagent systems
Planning
Predictive models
ISSN:2471-285X, 2471-285X
On-line přístup:Získat plný text
Tagy: Přidat tag
Žádné tagy, Buďte první, kdo vytvoří štítek k tomuto záznamu!
Popis
Shrnutí:Multiagent decision making becomes complicated when a subject agent interacts with other agents who could be either collaborative or competitive in their common environment. Generally the subject agent does not know what exact behaviours other agents will execute in the coming interactions. This occurs even when the agents are collaborative and cannot fully disclose their behaviours due to their privacy concerns. With the purpose of improving prediction of the true behaviours of other agents, conventional approaches often employ either an exhaust list or a guided search to expand the set of possible behaviours of other agents. In this article, we investigate an evolutionary approach guided by prior knowledge about other agents and develop a genetic algorithm based framework to modelling unknown behaviours, which may contain the true behaviours of other agents. We develop a new framework with a multi-population genetic algorithm in a general multiagent decision model - interactive dynamic influence diagrams, which represents how multiple agents optimise their interactive decisions from the viewpoint of individual agents. The new framework also provides an alternative way to solving the decision model. We demonstrate the performance of the proposed methods in two problem domains and provide empirical results in support.
Bibliografie:ObjectType-Article-1
SourceType-Scholarly Journals-1
ObjectType-Feature-2
content type line 14
ISSN:2471-285X
2471-285X
DOI:10.1109/TETCI.2022.3229308