search

huangtiansheng / fedslr

9 stars 1 forks source link

Related works #1

Closed vaseline555 closed 1 year ago

vaseline555 commented 1 year ago

Hi, I have run into this and found your work interesting!
BTW, I am curious on some missing citations (or baseline comparison) in your work, e.g. FedPara (ICLR 2021; which uses low rank decomposition for communication efficiency), FedRep (ICML 2021; which uses partial model exchange for updating global knowledge while maintaining local part intact), SuPerFed (KDD 2022; which mixes local and global knowledge like APFL but explictly induces mode connectivity, and is currently SOTA in PFL)
Is there any reason to exclude these works?

huangtiansheng commented 1 year ago

Apology for our delay response!

We did try implementing FedPara and FedRep, but the results are not satisfactory (potentially due to implementation issues, which might be misleading). Therefore we did not release the code of these two baselines.

For SuPerFed, we did not cite and compare because it is relatively new when FedSLR is published. I would try implementing it if I have time. Thank you for bring this issue to us!

vaseline555 commented 1 year ago

Nice. Thank you and hope you have a great day!