This work addresses the instability in asynchronous data parallel optimization. It does so by introducing a novel distributed optimizer which is able to efficiently optimize a centralized model under communication constraints. The optimizer achieves this by pushing a normalized sequence of first-order gradients to a parameter server. This implies that the magnitude of a worker delta is smaller compared to an accumulated gradient, and provides a better direction towards a minimum compared to first-order gradients, which in turn also forces possible implicit momentum fluctuations to be more aligned since we make the assumption that all workers contribute towards a single minima. As a result, our approach mitigates the parameter staleness problem more effectively since staleness in asynchrony induces (implicit) momentum, and achieves a better convergence rate compared to other optimizers such as asynchronous textsceasgd and which we show empirically.
|Title of host publication||Proceedings of the 9th Asian Conference on Machine Learning|
|Editors||Min-Ling Zhang, Yung-Kyun Noh|
|Publisher||Proceedings of Machine Learning Research|
|Number of pages||16|
|Publication status||Published - Nov 2017|
|Series||Proceedings of Machine Learning Research|