HPSGD: Hierarchical Parallel SGD With Stale Gradients Featuring
This work addresses efficiency issues in distributed DNN training for machine learning practitioners, offering incremental improvements over existing methods.
The paper tackles the low cluster utilization in distributed deep neural network training due to time-consuming data synchronization by proposing HPSGD, a hierarchical parallel SGD strategy that overlaps synchronization with local training and uses an improved model updating method to handle stale gradients, achieving substantial speedup and better accuracy within fixed wall-time.
While distributed training significantly speeds up the training process of the deep neural network (DNN), the utilization of the cluster is relatively low due to the time-consuming data synchronizing between workers. To alleviate this problem, a novel Hierarchical Parallel SGD (HPSGD) strategy is proposed based on the observation that the data synchronization phase can be paralleled with the local training phase (i.e., Feed-forward and back-propagation). Furthermore, an improved model updating method is unitized to remedy the introduced stale gradients problem, which commits updates to the replica (i.e., a temporary model that has the same parameters as the global model) and then merges the average changes to the global model. Extensive experiments are conducted to demonstrate that the proposed HPSGD approach substantially boosts the distributed DNN training, reduces the disturbance of the stale gradients and achieves better accuracy in given fixed wall-time.