Successful principles of inductive inference, such as maximum likelihood estimation, minimum description length or structural risk minimization, are deeply rooted in probability and statistics. Yet, not all inference in machine learning is of inductive nature. For example, in problems such as multi-task learning or transfer learning, the goal is to take advantage of what has been learned in one domain while learning in another domain. For example, learning to drive a truck is easier for someone who already knows how to drive a car. The corresponding process of knowledge transfer, i.e., of transferring knowledge from one domain to another, perhaps only partly and not necessarily one-to-one, is largely similarity-basedor analogical. Obviously, formal reasoning of that kind can nicely be supported by fuzzy inference techniques.