Random

The simplest selection strategy is random selection. It was first introduced along with federated learning itself in 2016. It is still the most frequently used strategy due to its simplicity and good results. The core parameter for this strategy is c, which is the percentage of available clients to select for a particular learning round. Common values for c range between 10% and 20%, depending on the total number of available clients.

It is based on the paper

McMahan, H. Brendan, Eider Moore, Daniel Ramage, Seth Hampson, and Blaise Agüera y. Arcas. 2016. 
“Communication-Efficient Learning of Deep Networks from Decentralized Data.” 
arXiv [cs.LG]. arXiv. http://arxiv.org/abs/1602.05629.

The algorithm may be selected by choosing random as the algorithm in the config file.

It requires the following parameters:

Key	Description	Example Value
c	Describes the share of all clients to participate in the round	0.2