sairin1202 / bic Goto Github PK

Hello,
Thank you for your code, I find it interesting.
However, I noticed that you hardcoded it to be able to run on CIFAR with 5 states (20 classes per state). For this, you created 5 bias removal layers , one for each state. How can I run it on other configurations (50 states in total for example), with other datasets contain more than 100 classes?

Thank you.

Thank you for your contribution. I would like to ask why there is a big error between the experimental results and the paper. I don't know right now.

Hi, I was wondering if you could provide the Python and PyTorch versions that this code was tested with?

Hello,
There are two (stage1()) functions defined in the Trainer class. Which one of them is correct? (the main difference is in the "retain_graph=True" during backpropagation).

Thank you

In the original paper, bias correction layer consists of only two parameters "alpha" and "beta", which are only applied to the logits corresponding to the last trained group of classes. Here it seems like you have multiple such layers for each group of classes.