| Original language | English |
|---|---|
| Pages (from-to) | 223-241 |
| Journal | International Journal of Approximate Reasoning |
| Volume | 36 |
| Issue number | 3 |
| DOIs | |
| Publication status | Published (in print/issue) - 1 Jul 2004 |
Bibliographical note
Other Details------------------------------------
Version space is a machine learning paradigm developed in the 1990s, but its expressive power is limited, and thus its applicability restricted. This paper presents a version space in a much more expressive hypothesis space, where a hypothesis is a set of hypertuples that together is maximal and consistent with given data. The set of all such hypotheses is a generalised version space. Evaluations show that the generalised version space compares favourably with the state-of-the-art classifiers. The hypertuple concept has become a general knowledge representation scheme, and has been used in some later publications by the authors.