Fig. 1

Illustration of our scenario for training and inference steps. In training step, our data filtering method reduces the size of raw data dramatically. Then, we use 10-fold cross validation to train shallow neural network (SNN) model using filtered data. As a result, we get the optimized parameters for both filtering and SNN model. In inference step, the protocol works as in the scenario using HE introduced in Background section. Since we use filtering method in preprocessing, only some chosen genes are needed to be encrypted in ciphertext. Then, we run HE-friendly SNN algorithm to evaluate our model. As a result, the client can receive the evaluated value using their private data without revealing any information