Fig. 9From: Instance segmentation on distributed deep learning big data clusterParameter synchronization is the process of aggregating the gradients computed by each worker node during the “model forward-backward” job and using them to update the model weights. This process occurs during the “parameter synchronization” jobBack to article page