Fig. 4From: Secure tumor classification by shallow neural network using homomorphic encryptionIllustration of the size of each model trained on the entire training set with given pair of (dcn,kvar). To optimize the computational cost on the privacy-preserving inference step based on homomorphic encryption, we seek the best parameters (dcn and kvar) with the model sizes of less or equal to 2B for each B=9,10,11, and 12Back to article page