×

Paradigms in measure theoretic learning and in informant learning. (English) Zbl 0924.68172

Summary: We investigate many paradigms of identifications for classes of languages (namely: consistent learning, EX learning, learning with finitely many errors, behaviorally correct learning, and behaviorally correct learning with finitely many errors) in a measure-theoretic context, and we relate such paradigms to their analogues in learning on informants. Roughly speaking, the results say that most paradigms in measure-theoretic learning wrt some classes of distributions (called \(\delta\) canonical) are equivalent to the corresponding paradigms for identification on informants.

MSC:

68T05 Learning and adaptive systems in artificial intelligence
68Q45 Formal languages and automata
03D80 Applications of computability and recursion theory
PDFBibTeX XMLCite
Full Text: DOI