a learning algorithm can be thought of as searching through the space of hypotheses for a hypothesis function that works well on the training set, and also on new examples that it hasn’t seen yet to ...