Sanjoy Dasgupta

UCSD

Title: Active learning of linear separators

Abstract:
In the well-studied setting of "supervised learning", the goal is
to learn a classifier from *labeled* data: for instance, given a
collection of credit card transactions which have been labeled
as legitimate or fraudulent, to learn a rule which will accurately
classify future transactions.

The problem with this model is that often, *unlabeled* data is easy
to come by but labels are expensive. For instance, if you're building
a speech recognizer, it's easy enough to get raw speech samples --
just walk around with a microphone -- but labeling even one of these
samples is a tedious process in which a human must examine the speech
signal and carefully segment it into phonemes.

In the field of "active learning", the goal is as always to construct
an accurate classifier, but the labels of the data points are initially
hidden and there is a charge for each label you want revealed. The hope
is that by intelligent adaptive querying, you can get away with
significantly fewer labels than you would need in the regular supervised
learning framework (that is, when all points automatically come labeled).

I'll give detailed background on active learning, and then show that for
a wide range of data distributions, it is possible to learn linear
separators using exponentially fewer labels than would be needed in
supervised learning.