A Smart Privacy-Preserving Learning Method by Fake Gradients to Protect Users Items in Recommender Systems

Algorithm 1

Perturbed Matrix Factorization algorithm.

	Input: Random mechanism , learning rate , and redefined iteration number k
	Output: Item profile matrix V
	Randomly initialize for all i and j.
	fordo
	Initialize for all j in central server.
	fordo
	On user i: sample j uniformly
	from{}.
	ifthen
	end
	else
	Generate a fake gradient of .
	set
	end
	for all j.
	end
	For all j:
fordo
	Update on a local device by gradient descent.
	end
	end