Table 1 Model architecture for MNIST (left) and CIFAR10 (right) networks
Layer type | Dimension | Layer type | Dimension |
|---|---|---|---|
Conv. + ReLU | 32 filters (3×3) | Conv. + Batch Norm. + ReLU | 64 filters (3×3) |
Conv. + ReLU | 32 filters (3×3) | Conv. + Batch Norm. + ReLU | 64 filters (3×3) |
Max Pooling | 2×2 | Max Pooling + Dropout (p=0.1) | 2×2 |
Conv. + ReLU | 64 filters (3×3) | Conv. + Batch Norm. + ReLU | 128 filters (3×3) |
Conv. + ReLU | 64 filters (3×3) | Conv. + Batch Norm. + ReLU | 128 filters (3×3) |
Max pooling | 2×2 | Max Pooling + Dropout (p=0.2) | 2×2 |
Fully connected + ReLU | 200 units | Conv. + Batch Norm. + ReLU | 256 filters (3×3) |
Fully connected + ReLU | 200 units | Conv. + Batch Norm. + ReLU | 256 filters (3×3) |
Softmax | 10 units | Max Pooling + Dropout (p=0.3) | 2×2 |
Conv. + Batch Norm. + ReLU | 512 filters (3×3) | ||
Max Pooling + Dropout (p=0.4) | 2×2 | ||
Fully Connected | 512 units | ||
Softmax | 10 units |