A Sufficient Condition for Learning Unbounded Unions of Languages with Refinement Operators.
Tomohiko Okayama, Ryo Yoshinaka, Keisuke Otaki, Akihiro Yamamoto · International Symposium on Artificial Intelligence and Mathematics · 2014
This paper presents a natural sufficient condition on a class of languages under which all the unions of any number of languages from the class are learnable from positive examples (data) in the Gold-style. Learning unions of languages models information extraction from mixed data from different sources. The Gold-style learning has provided many fruitful results on learning unions of bounded number of languages, while few positive results on learning unions of unbounded number of languages has been known. In this research, we focus on a condition of the class of languages on which refinement operators are defined. Refinement operators are fundamental tools to transform a hypotheses, which represents a language, into a set of hypotheses, which represent a subsets of the language.