Table 4 Table of experiment results for inference step over encrypted state. For each threshold for the model size, the number of chosen genes for CN and variants data are as in Table 3. Encoding Duplication means the number of duplications for X in encoding step, refers the parameter m. For each case, we state the microAUC value and accuracy on the right columns. These results come from the our trained model with test dataset
From: Secure tumor classification by shallow neural network using homomorphic encryption
Threshold of the model size | Encoding Duplication | Computation Time (sec) Accuracy | |||||||
|---|---|---|---|---|---|---|---|---|---|
Encryption | Const. Mult | RotSum | A. Softmax | Decryption | Total | ||||
512 | 1 | 6.7 | 16.8 | 61.0 | 130.5 | 0.1 | 215.1 | 0.9866 | 84.60 |
2 | 13.4 | 17.8 | 29.6 | 131.0 | 0.1 | 191.9 | |||
4 | 26.5 | 17.5 | 11.9 | 132.9 | 0.1 | 188.9 | |||
8 | 52.6 | 23.8 | 6.5 | 133.0 | 0.1 | 216.0 | |||
1024 | 1 | 13.5 | 32.3 | 62.0 | 132.9 | 0.1 | 240.8 | 0.9882 | 85.15 |
2 | 26.7 | 36.0 | 31.4 | 131.2 | 0.1 | 225.4 | |||
4 | 53.4 | 35.8 | 11.9 | 130.7 | 0.1 | 231.9 | |||
8 | 105.4 | 47.9 | 6.9 | 134.0 | 0.1 | 294.3 | |||
2048 | 1 | 26.7 | 64.9 | 63.2 | 132.1 | 0.1 | 287.0 | 0.9857 | 86.47 |
2 | 54.0 | 70.5 | 29.3 | 131.4 | 0.1 | 285.3 | |||
4 | 106.8 | 72.2 | 13.2 | 131.6 | 0.1 | 323.9 | |||
8 | 211.5 | 98.1 | 8.3 | 133.9 | 0.1 | 451.9 | |||
4096 | 1 | 53.8 | 129.4 | 63.6 | 130.7 | 0.1 | 377.8 | 0.9862 | 86.25 |
2 | 107.7 | 143.1 | 31.3 | 135.1 | 0.1 | 417.6 | |||
4 | 213.9 | 152.6 | 15.2 | 140.5 | 0.1 | 522.3 | |||