Our definition of stability is thus not directly related to the concept of high bias which is associated with learners such as MLR whose learned model is less influenced by small changes in the data. However, high bias may be a necessary or at least useful condition for a stable learner – provided it is able to find the correct model in the first place.