* Proceedings of the 20 th International Conference on Artificial
Intelligence and Statistics (AISTATS) 2017. JMLR: W&CP volume 54 * This version updates the large-scale LSTM experiments, along with
other minor changes. In earlier versions, an inconsistency in our
implementation of FedSGD caused us to report much lower learning rates for
the large-scale LSTM. We reran these experiments, and also found that fewer
local epochs offers better performance, leading to slightly better results
for FedAvg than previously reported Access Paper or Ask Questions
