LEAF: a FIPA compliant software toolkit for learning based MAS

  • Authors:

  • Steven Lynden;Omer F. Rana

  • Affiliations:

  • University of Wales, Cardiff, UK;University of Wales, Cardiff, UK

  • Venue:
  • Proceedings of the first international joint conference on Autonomous agents and multiagent systems: part 3
  • Year:
  • 2002

Quantified Score

Hi-index 0.00

Visualization

Abstract

This paper introduces LEAF, a FIPA compliant software toolkit for developing learning based multiagent systems. The FIPA-OS agent development toolkit is extended to include support for learning agents using techniques such as reinforcement learning, Q-learning and neural networks. A coordination mechanism is also provided that facilitates self organising/emergent behaviour, using an approach based on the utility assignment techniques of Collective Intelligence (COIN). A novel contribution of our work is that the notion of agent utility is extended to form two separate definitions of utility: the traditional functional utility of agents (defined in terms of how well the system is achieving its objectives), and a performance utility, which is based on performance engineering related issues.