Enhancing Fairness in Federated Learning: A Contribution-Based Differentiated Model Approach

Algorithm 2

DQN-based contribution scores update for FL.

	Input: current state
	Output: contribution scores set
(1)	Initialize the global model parameters for each cluster, experience replay memory
(2)	Initialize action-value function with random weights
(3)	Initialize target action-value function with weights
(4)	for to do
(5)	With probability select random actions
(6)	Otherwise select actions
(7)	Execute actions , update contribution score for each client.
(8)	Obtain clustering results by using Algorithm 1 based on contribution score
(9)	Run the FL algorithms independently on each cluster
(10)	Observe reward and next state
(11)	Store transition in
(12)	Sample random mini-batch of transitions from
(13)	Update weights by minimizing the loss
(14)	At every certain step, update the target network weights:
(15)	end for