First of all, thank you for your hard work in providing the community with such a good framework!
I've been digging through many binary network papers & repos because I want to find an efficient implementation in PyTorch. I've taken a look at the standard binary convolution provided here: https://gist.github.com/daquexian/7db1e7f1e0a92ab13ac1ad028233a9eb#file-binary_conv-py-L32.
I was wondering why the SignWeight does not have the same backward as SignSTE.
To my understanding, the same SignSTE should be applied to both the activations and the weights for backprop.
Any clarification or further discussion on this would be greatly helpful.
Thank you in advance!
First of all, thank you for your hard work in providing the community with such a good framework! I've been digging through many binary network papers & repos because I want to find an efficient implementation in PyTorch. I've taken a look at the standard binary convolution provided here: https://gist.github.com/daquexian/7db1e7f1e0a92ab13ac1ad028233a9eb#file-binary_conv-py-L32. I was wondering why the SignWeight does not have the same backward as SignSTE. To my understanding, the same SignSTE should be applied to both the activations and the weights for backprop. Any clarification or further discussion on this would be greatly helpful. Thank you in advance!