Multitask Learning is an inductive transfer method that improves learning on one problem by using the information contained in the training signals of other related problems. It does this by learning the problems in parallel while using a shared representation; what is learned for each problem can help other problems be learned better. In this talk I'll present results from using multitask learning with artificial neural nets on five problems. Three of these are medical decision- making problems dealing with community acquired pneumonia. In each of these problems multitask learning is currently the best performing machine learning method known.
I'll also explain how multitask learning works in backprop neural nets, present methods for making it work better, and show that there are many opportunites for applying multitask learning in real domains. I'll present an algorithm for multitask learning with case-based methods like k-nearest neighbor, and sketch an algorithm for multitask learning in decision trees. Multitask learning improves learning accuracy, can be applied to many different kinds of problems, and works with many different learning algorithms. It should prove useful in many real-world domains.