An emergent learning method capable of training a class of pattern classifiers in polynomial time and space

摘要

Emergent learning is a new learning method proposed in our previous work. In emergent learning, the solutions to a complex K-class classification problem emerged by simply combining the solutions of related smaller and simpler two-class subproblems, rather than directly solving the original complex K-class classification problem. In this paper we analyze the time and space complexity of training pattern classifiers with the emergent learning method. We demonstrate that the emergent learning method is capable of completely training a class of pattern classifers in both polynomial time and polynomial space.