A Novel Information Measure for Predictive Learning in a Social System Setting

Di Prodi, P., Porr, B. and Worgotter, F. (2010) A Novel Information Measure for Predictive Learning in a Social System Setting. In: International Conference on Simulation of Adaptive Behavior, Paris, France, 25-28 August 2010, pp. 511-522.

Abstract

We introduce a new theoretical framework, based on Shannon's communication theory and on Ashby's law of requisite variety, suitable for artificial agents using predictive learning. The framework quantifies the performance constraints of a predictive adaptive controller as a function of its learning stage. In addition, we formulate a practical measure, based on information flow, that can be applied to adaptive controllers which use hebbian learning, input correlation learning (ICO/ISO) and temporal difference learning. The framework is also useful in quantifying the social division of tasks in a social group of honest, cooperative food foraging, communicating agents. Simulations are in accordance with Luhmann, who suggested that adaptive agents self-organise by reducing the amount of sensory information or, equivalently, reducing the complexity of the perceived environment from the agents perspective.

• Item Type: Conference Proceedings
• Status: Published
• Refereed: Yes
• Glasgow Author(s) Enlighten ID: Porr, Dr Bernd
• Authors: Di Prodi, P., Porr, B., and Worgotter, F.
• College/School: College of Science and Engineering > School of Engineering > Biomedical Engineering