We consider speeding up stochastic gradient descent (SGD) by parallelizing it across multiple workers. We assume the same data set is shared among N workers, who can take SGD steps and coordinate with a central server. While it is possible to obtain a linear reduction in the variance by averaging all the stochastic gradients at every step, this requires a lot of communication between the workers and the server, which can dramatically reduce the gains from parallelism.The Local SGD method, proposed and analyzed in the earlier literature, suggests machines should make many local steps between such communications. While the initial analysis of Local SGD showed it needs \Omega ( \sqrt{T}) communications for T local gradient steps in order for the error to scale proportionately to 1/(NT), this has been successively improved in a string of papers, with the state of the art requiring \Omega \left( N \left( \mbox{ poly} (\log T) \right) \right) communications. In this paper, we suggest a Local SGD scheme that communicates less overall by communicating less frequently as the number of iterations grows.

Actor-critic method achieves state-of-the-art performance in many domains including robotics, game playing, and control systems (LeCun et al. (2015); Mnih et al. (2016); Silver et al. (2017)). Temporal Difference (TD) Learning may be thought of as a component of actor critic, and better bounds for TD Learning are usually ingredients of actor-critic analyses. We consider the problem of policy evaluation in reinforcement learning: given a Markov Decision Process (MDP) and a policy, we need to estimate the value of each state (expected discounted sum of all future rewards) under this policy. Policy evaluation is important because it is effectively a subroutine of many other algorithms such as policy iteration and actor-critic. The main challenges for policy evaluation are that we usually do not know the underlying MDP directly and can only interact with it, and that the number of states is typically too large forcing us to maintain a low-dimensional approximation of the true vector of state values.

Synchronous mini-batch SGD is state-of-the-art for large-scale distributed machine learning. However, in practice, its convergence is bottlenecked by slow communication rounds between worker nodes. A natural solution to reduce communication is to use the \emph{`local-SGD'} model in which the workers train their model independently and synchronize every once in a while. This algorithm improves the computation-communication trade-off but its convergence is not understood very well. We propose a non-asymptotic error analysis, which enables comparison to \emph{one-shot averaging} i.e., a single communication round among independent workers, and \emph{mini-batch averaging} i.e., communicating at every step. We also provide adaptive lower bounds on the communication frequency for large step-sizes ($ t^{-\alpha} $, $ \alpha\in (1/2 , 1 ) $) and show that \emph{Local-SGD} reduces communication by a factor of $O\Big(\frac{\sqrt{T}}{P^{3/2}}\Big)$, with $T$ the total number of gradients and $P$ machines.

Consider a number of workers running SGD independently on the same pool of data and averaging the models every once in a while -- a common but not well understood practice. We study model averaging as a variance-reducing mechanism and describe two ways in which the frequency of averaging affects convergence. For convex objectives, we show the benefit of frequent averaging depends on the gradient variance envelope. For non-convex objectives, we illustrate that this benefit depends on the presence of multiple globally optimal points. We complement our findings with multicore experiments on both synthetic and real data.