PAGODA: A Model for Autonomous Learning in Probabilistic Domains
Marie Ellen Desjardins, Marie Ellen Desjardins · 1992
Machine learning approaches have traditionally made strong simplifying assumptions: that a benevolent teacher is available to present and classify instances of a single concept to be learned; that no noise or uncertainty is present in the environment; that a complete and correct domain theory is available; or that a useful language is provided by the designer. Additionally, much existing machine learning research has been done in a piecemeal fashion, addressing subproblems without a uniform conceptual approach to designing intelligent systems. The resulting learning techniques often are only useful for narrowly defined problems, and are so dependent on the underlying assumptions that they do not generalize well---or at all---to complex domains. pagoda (Probabilistic Autonomous GOal-Directed Agent), the intelligent agent design presented in this thesis, avoids making any of the above assumptions. It incorporates solutions to the problems of deciding what to learn, selecting a learning ...