Agent alliance formation using ART-networks as agent belief models

In today’s hyper-competitive business environ1 ments virtual organisations are becoming highly dynamic 2 and unpredictable. Individuals may want to work together 3 across organisation boundaries but do not have much prior 4 knowledge about potential partners. The semantic web and 5 its associated new standards appear very promising as can6 didates to support a new generation of virtual organisations. 7 Whilst knowledge can be represented in a machine inter8 pretable way, social-like behaviours can be expected in a vir9 tual organisation. In this paper ontology definition techniques 10 from the semantic web are applied to define a virtual state 11 space of a virtual organisation. Actors involved in an organi12 sation, from high level strategy making members to low level 13 physical devices, advertise their skills and local knowledge in 14 a community. A task initiator, with a virtual sensor to perceive 15 the advertised skills and with an adaptive belief model about 16 the community, seeks for the best matched partners for coope17 ration. The belief model is a fuzzy neural network based on 18 Adaptive Resonance Theory which takes the advertisements 19 of actors as its initial belief and learns actors’ actual capa20 bilities through interaction experience. Dynamic alliances 21 can then take place in an automated/semi-automated way 22 that exhibit adaptive ability, self-organisation, unsupervised 23 P. Jiang (B) Department of Computing, University of Bradford, Bradford BD7 1DP, UK e-mail: [email protected] Q. Mair Department of Computing, Glasgow Caledonian University, Glasgow G4 0BA, UK e-mail: [email protected] Z.-R. Feng Systems Engineering Institute, Xi’an Jiaotong University, Xi’an 710049, China e-mail: [email protected] learning and competition ability. The alliances thus exhibit 24 the inherent characteristics of realistic enterprises or human 25 societies. 26