A Computational Framework for Convergent Agents

摘要

As a computational approach, a framework is proposed for computing the limits of formal theory sequences. It defines a class of agents, called convergent agents. The approach provides a method to generate a new theory by the limit of some sequence of theories, and also has potential applications to many scientific and engineering problems. As applications of the framework, some convergent agents are discussed briefly, e.g., GUINA, which can learn new versions from the current versions of a theory and some external samples, and the learned versions converge to the truth one wants to know.