Personalized Learning using Multiple Kernel Models

This paper considers a personalized learning system with a large number of users each trying to learn a desired task. The user learning tasks have similarities as each of the users have common attributes. However, the learning tasks are each slightly different as the tasks are personalized to each learning system. There is also the concern that each learning system may not receive enough data to learn its desired task. To account for this a few canonical learners receive all the data from each personalized learning system. Each personalized learning system then takes a weighted sum of the canonical learners to realize their desired task. Here we will consider learning via kernel methods.