A Fairness-Enhanced Federated Learning Scheduling Mechanism for UAV-Assisted Emergency Communication

. 2024 Feb 29;24(5):1599. doi: 10.3390/s24051599

Algorithm 1 A MAB-based fair scheduling algorithm for reducing energy consumption.

Initialize:
global model $w^{0}$ ; M terrestrial devices indexed by m, each with a local data volume of $D_{m}^{t}$ in round t;
Output:
global model $w^{T}$
1:
$The UAV :$
2:
Initialize global model $w^{0}$
3:
for each round $t = 1$ to T do
4:
$S_{t}$ ← ClientSelection(t, N), $|S_{t}| = N$
5:
Receive the trained local models $w_{m}^{t}$ from the terrestrial device in the set $S_{t}$
6:
Aggregate the models uploaded according to Equation (2)
7:
Send $w^{t}$ to M terrestrial devices
8:
end for
9:
$The terrestrial device U_{m} :$
10:
Receive the global model $w^{t}$ from the UAV
11:
for each local round $l = 1$ to T do
12:
Train the local model $w_{m}^{t}$ according to Equation (1)
13:
end for
14:
Send $w_{m}^{t}$ to the UAV
15:
$ClientSelection (t, K, N) :$
16:
Initialize the relevant parameters to ensure that the reward distribution is unknown
17:
Update the reward function according to Equation (15) and also update the UCB score according to Equation (16)
18:
Sort the UCB scores of individual terrestrial devices in descending order
19:
Select the first N terrestrial devices to form $S_{t}$ , i.e., $S_{t} = \underset{S_{t}}{arg max} \sum U C B_{m, t}$
20:
Return $S_{t}$ to the UAV