Privacy-Enhanced Data Fusion for Federated Learning Empowered Internet of Things
Algorithm 1
Local data fusion with differential privacy protection.
Input: Initial model parameter received from the FL fusion server, learning rate , local sensor dataset , gradient clipping , privacy budget , sensitivity , and Gaussian noise to be added satisfying
Output: Final model parameter
(1)
for do
(2)
Calculate the gradient for each batch
(3)
Clip the gradient by and calculate average gradient
