We consider version spaces in a hyp othesis space of oriented hyp erplanes defining decision functions. Version spaces are represented by the training data [4, 11].
Our implementation of the unanimous-voting rule tests whether version spaces
are empty in the hyp othesis space. Testing is realized via supp ort vector machines
(SVM) [12]. Therefore, our approach is a combination of the two learning schemes
and it is called version space supp ort vector machines (VSSVM).
We conducted exp eriments on datasets from the UCI ML rep ository [1]. Our
results are promising: 100% accuracy and an acceptable coverage.
The remainder of the pap er is organized as follows. Section 2 formalizes the
classification task. Version spaces and SVM are briefly sketched in section 3, and
in section 4 we intro duce VSSVM. Our initial exp eriments with VSSVM are given
in section 5. Section 6 concludes the pap er.