Parallel Stochastic Gradient Descent with Sound Combiners