Andrea Cossu

Hi @ashok-arjun ! Feel free to contribute to this issue by adding relevant papers (you can join our Zotero account) or by linking them here in this thread :smile:

Thank you for the warning :grin: Currently, we are in the process of entirely restructuring our notebooks. Hopefully, there will be a working version soon. Anyway, for a quick solution...

@ggraffieti are you still working on it? Otherwise @gupta1912 can give this a try.

@gupta1912 I can assign you to this issue if you still want to do it. I cal also provide some help if you encounter any issue.

This is related to the issue about the reproduction of AR1* #5 , which has been introduced in the same paper.

I changed that part of code with the following one, but the final accuracy is still lower than expected. ```python if isinstance(self.prev_model, MultiTaskModule): # output from previous output heads. with...

Oh right, the change above applies only to `MultiTaskModule`, so I guess that is still a useful fix. There may be another problem for single-headed models, too.

Hi @travela, what do you mean exactly? Are you referring to the results of a strategy present in this repository?

I see, thank you! I think the best way to do it is to choose a reference paper and implement that exact experimental setup. If the results match then it...

Thanks @travela ! I have a few questions: 1. Shouldn't the reference performance be 0.79, given Table 1 in the paper for MNIST dataset? 2. The paper says they use...