LeNet Structure Used:

1. Convolution 1: (5, 5, 32)
2. Bacth normalization 1
3. Max pool 1: kernel size (2, 2), stride (2, 2)
4. Convolution 2: (5, 5, 64)
5. Bacth normalization 2
4. Max pool 2: kernel size (2, 2), stride (2, 2)
5. Flatten: (3136,)
6. FC 1: (3136 -> 1024)
7. Drop out 1
8. FC 2: (1024 -> 84)
9. FC 3: (84 -> 10)

Training Details:

1. Learning rate: 0.001
2. Drop out keep prob: 0.6
3. Batch size: 128
4. Number of epochs: 40

Training Results:

Test accuracy: 0.9945