Learning in Presence of Ontology Mapping Errors

摘要: The widespread use of ontologies to associate semantics with data has resulted in a growing interest the problem learning predictive models from sources that different model same underlying domain (world interest). Learning such \emph{semantically disparate} involves mapping resolve semantic disparity among used. Often, practice, used may contain errors and as algorithms setting must be robust presence errors. We reduce semantically disparate variant nasty classification noise. This reduction allows us transfer theoretical results latter former.