[edit]
Exponential Convergence of Testing Error for Stochastic Gradient Methods
Proceedings of the 31st Conference On Learning Theory, PMLR 75:250-296, 2018.
Abstract
We consider binary classification problems with positive definite kernels and square loss, and study the convergence rates of stochastic gradient methods. We show that while the excess testing \emph{loss} (squared loss) converges slowly to zero as the number of observations (and thus iterations) goes to infinity, the testing \emph{error} (classification error) converges exponentially fast if low-noise conditions are assumed. To achieve these rates of convergence we show sharper high-probability bounds with respect to the number of observations for stochastic gradient descent.