Advances, Systems and Applications
From: FedEem: a fairness-based asynchronous federated learning mechanism
Symbol | Description |
---|---|
T, t | Server update frequency, server update index |
\(\mathcal {S}^t\) | Server updates the selected subset of client in time \(\textit{t}\) |
\(Q, q, \epsilon\) | Local step count per round, round index |
\(w^t\) | The model after t updates |
\(g_i\left( w ; \zeta _i\right) :=g_i(w)\) | Random gradient |
\(\eta _g \eta _l\) | Global and local learning rate |
K | The number of clients selected for a single aggregation |
\(\sigma _g^2, \sigma _{\ell }^2\) | Global and Local gradient variance |
\(\tau _i(t)\) | Outdatedness of client \(\textit{i}\)’s model after \(\textit{t}\) rounds of global updates |
\(\tau _{\max }\) | Obsolete upper bound |