tflearn on mnist datasets
-
LeNet (Accuracy: 0.997) (ref)
- conv (5x5, 6 feature maps), max_pooling (2x2) original l2 pooling
- conv (6x6, 16 feature maps), max_pooling (2x2)
- conv (6x6, 120 feature maps)
- fully-connected, 84
- full-connected, 10, softmax
-
Default (Accuracy : 0.9921)
-
full-connected, 20 , ReLU
-
full-connected, 10, softmax
Leiming Yu [homepage] [github]
- AlexNet (paper)