Research Article
Inversion Attacks against CNN Models Based on Timing Attack
| Influence of convolution layers | Meaning | Attribute |
| | The size of input pictures | Inherent parameters of the convolution layer | | The size of the convolution kernel | Inherent parameters of the convolution layer | | Stride convolutions | Inherent parameters of the convolution layer | | Padding | Inherent parameters of the convolution layer | | The size of the feature map | Acquired by formula (1) | | Input channel | Inherent parameters of the convolution layer | | Output channel | Inherent parameters of the convolution layer | | Execution time of convolution | Provided with profiling file | | Bias | Inherent parameters of the convolution layer | | The memory of model weight | Provided with profiling file | | The memory of model parameters | Provided with profiling file | | Parameter to be solved | Inherent parameters of the model carrier | | Weight coefficient | Inherent parameters of the model carrier | | Parameter coefficient | Inherent parameters of the model carrier | | Calculation of convolution | Acquired by formula (5) |
|
|