Skip to yearly menu bar Skip to main content


Asynchronous SGD Beats Minibatch SGD Under Arbitrary Delays

Konstantin Mishchenko · Francis Bach · Mathieu Even · Blake Woodworth

Hall J (level 1) #331

Keywords: [ Optimization ] [ stochastic ] [ Parallel ] [ Convex ] [ Asynchronous ] [ Non-Convex ]


The existing analysis of asynchronous stochastic gradient descent (SGD) degrades dramatically when any delay is large, giving the impression that performance depends primarily on the delay. On the contrary, we prove much better guarantees for the same asynchronous SGD algorithm regardless of the delays in the gradients, depending instead just on the number of parallel devices used to implement the algorithm. Our guarantees are strictly better than the existing analyses, and we also argue that asynchronous SGD outperforms synchronous minibatch SGD in the settings we consider. For our analysis, we introduce a novel recursion based on ``virtual iterates'' and delay-adaptive stepsizes, which allow us to derive state-of-the-art guarantees for both convex and non-convex objectives.

Chat is not available.