We present new probabilistic bounds on the generalization error of complex
classifiers that are "combinations" of simpler classifiers from a base
class of functions. The bounds in question are in terms of so called "margins"
of combined classifiers and they apply to voting methods of combining classifiers.
Although the bounds of this type originated in computer science literature,
their true nature is related to the inequalities developed in the theory
of empirical processes and in Probability in Banach spaces, especially,
to concentration inequalities of Talagrand and also to symmetrization and
comparison inequalities.
Seminar to be held in Room 107, 24 Hillhouse Avenue at 4:15 pm