Z. Zhou, P. Mertikopoulos, N. Bambos, P. W. Glynn, and Y. Ye. Under review.
The recent surge of breakthroughs in machine learning and artificial intelligence have sparked renewed interest in large-scale stochastic optimization problems that are universally considered hard. One of the most widely used methods for solving such problems is distributed asynchronous stochastic gradient descent (DASGD), a family of algorithms that results from parallelizing stochastic gradient descent on distributed computing architectures in a fully asychronous manner. However, a key obstacle in the efficient implementation of DASGD is the issue of delays: when a computing node contributes a gradient update, the global model parameter may have already been updated by other nodes several times over, thereby rendering this gradient information stale. These delays can quickly add up if the computational throughput of a node is saturated, so the convergence of DASGD may be compromised in the presence of large delays. Nevertheless, by carefully tuning the algorithm’s step-size, we show that convergence to the critical set is still achieved in mean square, even if the delays grow unbounded at a polynomial rate. We also establish finer results in a broad class of structured optimization problems (called variationally coherent), where we show that DASGD converges to a global optimum with probability 1 under the same delay assumptions. Together, these results contribute to the broad landscape of large-scale non-convex stochastic optimization by offering state-of-the-art theoretical guarantees and providing insights for algorithm design.