Structured updates (low-rank or masked) and sketched updates (quantized, rotated, subsampled) reduce uplink communication in federated learning by up to two orders of magnitude on convolutional and recurrent networks.
Brendan McMahan, Eider Moore, Daniel Ramage, Seth Hampson, and Blaise Aguera y Arcas
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.LG 1years
2016 1verdicts
CONDITIONAL 1representative citing papers
citing papers explorer
-
Federated Learning: Strategies for Improving Communication Efficiency
Structured updates (low-rank or masked) and sketched updates (quantized, rotated, subsampled) reduce uplink communication in federated learning by up to two orders of magnitude on convolutional and recurrent networks.