Machine Learning in an Agent: A Generic Model and an Intelligent Agent based on Inductive Decision Learning

摘要

In this study, our aim is to present a model for learning Agent-based systems composed essentially of a learning and a reasoning components acting respectively on and from a critical element of this model, its Cognitive knowledge base. We have built this model to overcome the domains dependency of existing learning Agent-based systems. Indeed, the proposed model is a generic pattern in the sense that it covers all machine learning approaches without specifying any learning method but deferring it to this pattern s instantiation. Also, this model is general by being independent of any application but allowing its instantiation by different domain. Then, as a prototypical application, an Intelligent E-mails Handler, developed as an instantiation of our generic and general model, will be presented. An inductive learning approach with the induction decision tree algorithm, found suitable for learning in the E-mails domain, have been chosen. The inductive learning method used in our experimentation is appropriate to the E-mails Handler objectives; indeed it incrementally improves this system's autonomy, intelligence, personalization and performance.