Research Article
Robust Fully Distributed Minibatch Gradient Descent with Privacy Preservation
| Parameter setups | Time consumption (seconds) | Results | ā | Number of features () | | Max tree size () | Bits per feature () | Key size () | Blocks | Message size to parent | Encrypt/decrypt a block | Send plain- text model | Encrypt shares | One aggregation round | Overall time of mini-batch | Prob. of good tree |
| 4 | | 4 | 19 | 10 | 1024 | 1 | 8192 | 0.041 | 0.103 | 0.123 | 0.143 | 1.847 | 0.999 | 2048 | 1 | 16384 | 0.300 | 0.103 | 0.900 | 0.404 | 4.451 | 0.997 | 6 | 67 | 14 | 1024 | 2 | 16384 | 0.041 | 0.103 | 0.246 | 0.186 | 2.850 | 0.997 | 2048 | 1 | 16384 | 0.300 | 0.103 | 0.900 | 0.404 | 5.466 | 0.996 | | 4 | 19 | 10 | 1024 | 99 | 811008 | 0.041 | 0.420 | 12.177 | 4.362 | 45.649 | 0.969 | 2048 | 50 | 819200 | 0.300 | 0.420 | 45.000 | 15.305 | 155.074 | 0.904 | 6 | 67 | 14 | 1024 | 137 | 1122304 | 0.041 | 0.420 | 16.851 | 5.998 | 74.609 | 0.951 | 2048 | 69 | 1130496 | 0.300 | 0.420 | 62.100 | 21.083 | 255.624 | 0.850 |
|
|