Mark worked on the difficult problem of transfer learning. That is, improving the ability to learn a concept based on previous learning. Since all learning requires some form of bias, either in representation or search, the problem is whether such biases can be transferred between "similar" learning tasks