ReCU

ReCU copied to clipboard

a little question about information entropy

1

i am wondering in your paper why use latent full precision weights to calculate information entropy rather than binarized weights? It seems make no sense considering latent weights.

XA23i

为什么在论文中不比较与ReActNet基于mobilenet-v1提出的新网络结构在应用ReCU后的效果呢?

1

PeiqinSun

evaluation error

1

During evaluation i get following errors Traceback (most recent call last): File "/home/ahmsoy00/Projects/DATE23_Invited_Variational_Bayes/classification/CIFAR-10/ReCU/main.py", line 319, in main() File "/home/ahmsoy00/Projects/DATE23_Invited_Variational_Bayes/classification/CIFAR-10/ReCU/main.py", line 118, in main val_loss, val_prec1, val_prec5 = validate(val_loader, model, criterion,...

SoyedTuhinAhmed

How does torch.clamp() reviving dead weights?

1

Appreciate your excellent work! I look up the code in **binarized_modules.py** and find that **torch.clamp()** is applied to constrain weights within ±Q_tau. In backwards, this approach stops "dead weights"(which >=Q_tau...

YKrisLiu

For ResNet in cifar experiment , the shortcut/downsampling is binarized to [-1,+1] via BinarizeConv2d. However in ImageNet experiment , the shortcut/downsampling remains the fullprecision approach via nn.Conv2d. Is this the...

kriskrisliu

ReActNet architecture

1

Hello developers, Thank you for sharing your code. Would you mind sharing also the full instructions to reproduce your results with the training set from ReActNet? In particular, did you...

CosimoRulli