Learning can be done by managing a set of general models and a set of specific models kept in a version space. A version space is nothing but a representation that is used to get relevant information from a set of learning examples.
A version space description consists of two trees which are complement to each other: one represents general model and the other represents specific model. Both positive and negative examples are used to get the two set of models converge on one just-right model. That is, positive examples generalize specific models and negative examples specialize general models. Ultimately, a correct model that matches only the observed positive examples is obtained. Query By Committee (QBC) is an algorithm which implements this technique in order to acquire knowledge.