A Nonparametric Training Algorithm for Decentralized Binary Hypothesis Testing Networks

We present a distributed nonparametric minimum-error training algorithm for networks of linear threshold classifiers performing decentralised binary hypothesis testing (detection). The training algorithm consists of communicating stochastic approximation algorithms. Knowledge of the network topology is required by the algorithm. We suggest that models of the variety in this study provide a paradigm for the study of adaptation in human decision making organizations.