Primal-dual version spaces

Several learning strategies, based on the version space learning strategy, have been proposed to incrementally learn disjunctive concepts from examples. All of them must however store past training instances for the learning process to be successful. If the amount of training data is large, it then causes a heavy load. We propose a new learning strategy, the “primal-dual version spaces” learning strategy, which can incrementally learn disjunctive concepts well without keeping track of the past training instances at all