Priced Learning

In iterative learning the memory of the learner can only be updated when the hypothesis changes; this results in only finitely many updates of memory during the overall learning history. Priced learning relaxes this constraint on the update of memory by imposing some price on the updates of the memory – depending on the current datum – and requiring that the overall sum of the costs incurred has to be finite. There are priced-learnable classes which are not iteratively learnable. The current work introduces the basic definitions and results for priced learning. This work also introduces various variants of priced learning.