TsingZ0 / PFLlib

37 traditional FL (tFL) or personalized FL (pFL) algorithms, 3 scenarios, and 20 datasets.
GNU General Public License v2.0
1.35k stars 283 forks source link

fedavg隐私在训练过程中没有消耗问题 #175

Closed xhjiang1998 closed 5 months ago

xhjiang1998 commented 5 months ago

python main.py -data mnist -m cnn -algo FedAvg -gr 100 -did 0 -go cnn -dp true -nc 10 -ls 5 client本地迭代5次,观察dp情况。发现每个客户端打印的epsilon都不变,理论上应该随着训练而逐渐损耗隐私预算的啊 image

TsingZ0 commented 5 months ago

我们这边实现DP是通过opacus库实现的,具体特性可能跟FL中所探索的DP不太一样,具体请参考opacus的说明文档。