t-BMPNet: Trainable Bitwise Multilayer Perceptron Neural Network over Fully Homomorphic Encryption Scheme

Table 3

Comparison of t-BMPNet with other approaches—BFV and CKKS—in terms of execution time (minute), applicability to complex neural networks, and whether it is trainable under the encrypted domain.

Execution time comparison of MLPNN [2, 3, 6, 7]
Method	Type	Scheme	Parameter	Feedforward		Complex network?	Trainable?	Related works
				Square	Sigmoid
t-BMPNet	Fully	TFHE		4.94	9.48	Yes	Yes	—
				19.33	39.32
				75.91	147.5
Other approaches	Leveled	BFV		0.15	—	No (requires FHE scheme)	No	[2, 25]
		CKKS		0.57	0.97			[26–28]