Affiliations: [a] Applied Computational Intelligence Research Unit, The University of the West of Scotland, UK | [b] School of Electrical and Information Engineering, University of South Australia, Adelaide, Australia
Correspondence:
[*]
Corresponding author: School of Computing, the University of the West of Scotland, UK. Tel.: +44 141 848 3305; E-mail: [email protected]
Abstract: We investigate the problem of agent communication when such agents are cooperating rather than competing. We do this using the statistical technique of canonical correlation analysis. We consider two stochastic process methods for performing canonical correlation analysis (CCA). The first creates a Gaussian Process formulation of regression in which we use the current projection of one data set as the target for the other and then repeat in the opposite direction; this is useful in the very limited environment in which the agents have prior and precise knowledge of the nature of their interaction. We extend the problem by creating a problem which no single agent can solve, but for which together they can find a solution. We investigate methods by which we can automatically find the number and composition of the groups necessary to solve the problem. Finally we develop a Dirichlet process of Gaussian models in which the Gaussian models are determined by Probabilistic CCA [2]. The Dirichlet Process enables us to have groups of agents cooperating on a task without having to specify in advance how many groups of agents there are or how many agents join each group.
