
News
Colloquia
Multitask Learning
Rich Caruana
Cornell University
Tuesday, March 31, 2005
Walker 6113 4:00 p.m. to 5:00 p.m.
Refreshments at AE 401 3:30 p.m.
If it is hard to learn one problem, certainly it is harder to learn
100 problems. This is not necessarily true; what you learn for one
problem can make learning another problem easier. Multitask Learning
is a machine learning method that learns each problem better by also
learning from the training signals of *other* related problems. It
does this by learning all of the problems in parallel while using a
shared representation; what is learned for each problem helps other
problems be learned better.
In the talk I demonstrate multitask learning on a halfdozen problems.
Two of these are real problems in medical decision making for which
multitask learning currently outperforms all other methods. I explain
how multitask learning works, and show that there are many ways to use
it on real problems, some of which are rather surprising. I'll also
present suggestions for how to get the most out of multitask learning
in artificial neural nets, present an algorithm for multitask learning
with casebased methods like knearest neighbor and kernel regression,
and sketch an algorithm for multitask learning in decision trees.
Last updated: March 21, 2005

