Federated Learning Based Neural Machine Translation

In this paper a work on Federated Learning based Machine Translation is performed using English-Assamese Language pair. Federated Learning is the concept of machine learning where data are distributed on different system called clients/nodes. Our objective was to performe a Machine Translation training using Federated Learning Architecture and observed the performance by varying the system parameters/hyperparameters. Also two different Federated Aggregation strategies have been used for comparison. It was observed that having a higher number of clients/nodes, there is a reduce in performance score.