Sandra Carrasco

Thank you for your answer @pedropgusmao . Yes, I've performed several experiments using FedAvg and never had this issue before. Here you can access the output log: https://drive.google.com/file/d/1TWg7EgCHFbDMID7SaW78B1vNnfemWyjZ/view?usp=sharing Thanks!

Hi @pedropgusmao , I printed `sum([np.isnan(w).sum() for w in weights ])` before the evaluation to detect if the parameters being evaluated have NaN, but apparently they haven't. This is the...

@danieljanes the weights aren't NaN before or after the aggregation, but after the forward pass (in the second evaluation round) the results are NaN. We thought it might be related...

Sure, here you have a minimal example using Flower 0.18, with only 50 images: https://github.com/sandracl72/flower_fedopt_debug.git You have to run it with `--nowandb` arg. I've left it in case you want...

Perfect , thanks for the quick reply ! In the meanwhile, could you provide the config you used for the decoder ablation described in the paper? Thank you!

Thank you ! It would be great if you could also provide some explanation of all the evaluation metrics you log. Thanks.

Please, could you tell us when, aproximatelly, could be the code realeased? Thank you.