Accelerated Mini-Batch Stochastic Dual Coordinate Ascent
This work addresses the need for faster optimization methods in machine learning, particularly for regularized loss minimization, but it appears incremental as it extends an existing technique.
The paper tackles the problem of improving the convergence rate of stochastic dual coordinate ascent (SDCA) in the mini-batch setting, introducing an accelerated version that achieves a fast convergence rate and is implemented over a parallel computing system.
Stochastic dual coordinate ascent (SDCA) is an effective technique for solving regularized loss minimization problems in machine learning. This paper considers an extension of SDCA under the mini-batch setting that is often used in practice. Our main contribution is to introduce an accelerated mini-batch version of SDCA and prove a fast convergence rate for this method. We discuss an implementation of our method over a parallel computing system, and compare the results to both the vanilla stochastic dual coordinate ascent and to the accelerated deterministic gradient descent method of \cite{nesterov2007gradient}.