Analysis patterns for learning agents

Abstract

In this paper, we focus on requirements analysis of agent based software development and introduce a set of patterns for analysis of (requirements for) learning agents. This paper also proposes a process for producing, evaluating and improving patterns that combines theoretical and practical approaches. Following this process, we first present the first version of our analysis pattern for learning agents. Then applicability of the pattern is investigated through the development of three case studies. The results of evaluating this pattern (via comparing it with a conventional approach) are presented and they are used to improve the pattern and introduce a new version. In order to decrease the complexity of the pattern and increasing its reusability, we also decompose the second version of the pattern into a set of patterns which introduces the third version.