A Clonal Selection Optimization System for Multiparty Secure Computing

Algorithm 1

Parallelized model training process.

	Input:
	Original dataset , stride , number of nodes , local iterations
	Central server splits to participating nodes:
	Batch size of each participating node
	For all participating nodes parallel do
	Initialize the model parameters of each node
	For all data in dataset
	Generate multiple parameter weight update vectors according to
	Calculate the corresponding loss function value
	Select the optimal update vector
	Update the parameters of the node model
	Until number of iteration do
	Update the global parameter weights for the node model
	End for
	Aggregate the update information of all nodes
	Output
	Update central model parameters