Meta-Weight-Net_Code-Optimization

Meta-Weight-Net_Code-Optimization copied to clipboard

the necessity of "metaSGD" instead of direct "torch.nn.SGD"

2

Hi, thanks for your elegant code. I haven't figured out the necessity of "metaSGD" instead of direct "torch.nn.SGD", either. Is it possible that you provide more detailed descriptions? Thanks a...

KeysaYoung

RuntimeError: One of the differentiated Tensors appears to not have been used in the graph. Set allow_unused=True if this is the desired behavior.

2

Hi, thank you very much for your work. When I use my own dataset and model, It error in this Line: `pseudo_grads = torch.autograd. grad(pseudo_loss, pseudo_net.parameters(), create_graph=True)` The following error...

yzakzero

First of all, thank you very much for your work! I am using Resnet18 imported by pytorch (pretrain =true) and using my own dataset for image classification. Even with the...

yzakzero

question regarding performance of normal training on corrupted labels

20

Hi, thank you very much for the code. However, for noisy labels, I do not observe significant difference between MWN and normal training (without MWN) when the corruption prob is...

pipilurj

Is there possible to replace SGD with Adam

1

Hi, thank you very much for the code. And, I wonder why you rewrite the SGD as MetaSGD and is there any possible to use the Adam algorithm while training...

qwzlh

Thanks for your elegant code. I've got a question about the output weight. Though sigmoid function could output a non-zero vector. https://github.com/ShiYunyi/Meta-Weight-Net_Code-Optimization/blob/1d3cab0f4ad435e2569de97b21694771437321b3/model.py#L108 But the weight vector's elements could be all...

shanshuo