search

yuchaoli / KSE

Exploiting Kernel Sparsity and Entropy for Interpretable CNN Compression
47 stars 21 forks source link

How to use KSE in fc? #5

Open Alexanzhuo opened 3 years ago

Alexanzhuo commented 3 years ago

Thank you for providing this work. I wonder that how to use kse in nn.Linear? In your paper, you argue that “ It is generic, and can be used to compress fully-connected layers by treating them as 1 × 1 convolutional layers.” In your implementation, it seems that you only use KSE for conv_2d?

yuchaoli commented 3 years ago

Yes, because of the nn.Linear can be directly transformed into 1 × 1 convolutional layers, so we do not implement the KSE in the FC layer.