Evolutionary agent learning

Authors: Buczak, A. L.¹; Cooper, D. G.²; Hofmann, M. O.²

Source: International Journal of General Systems, Volume 35, Number 2, April 2006 , pp. 231-254(24)

Buy & download fulltext article:

Price: $56.94 plus tax (Refund Policy)

Abstract:

Evolutionary Platform for Agent Learning (EPAL) is a novel methodology for agent learning, based on an extension to genetic programming that Lockheed Martin Advanced Technology Laboratories (LM ATL) developed. EPAL allows software agents to learn how to better operate under external conditions, so they do not repeat their mistakes. It creates both subtle and drastic changes to agent behavior. Subtle changes arise from learning task parameters; drastic changes emerge from learning improved workflows containing new programming constructs and tasks. Although EPAL was constructed for learning by LM ATL developed extendible mobile agent architecture agents, it is a suitable learning methodology for any software agent whose behavior can be represented as a workflow and further decomposed into building blocks, such as operators, tasks, and parameters. Our agents learned to solve a real-world problem that is similar to problems encountered in Navy Fleet Battle Experiment-Juliet. Results for both parameter learning and workflow learning are very encouraging.

Keywords: Intelligent software agents; Machine learning; Genetic programming; Agent learning; Complex systems

Document Type: Research article

DOI: http://dx.doi.org/10.1080/03081070500422653

Affiliations: 1: Sarnoff Corporation, 201 Washington Road, Princeton, NJ, 08543-5300, USA 2: Lockheed Martin Advanced Technology Laboratories, 3 Executive Campus, Cherry Hill, NJ, 08002, USA

Publication date: 2006-04-01

More about this publication?