jiang-zl/LightSecAggg

Source code for MLSys 2022 submission "LightSecAgg: a Lightweight and Versatile Design for Secure Aggregation in Federated Learning"

LightSecAgg: a Lightweight and Versatile Design for Secure Aggregation in Federated Learning

(Under Review)

Implementation

We implement our LightSecAgg algorithm by reusing part of the source code from FedML.ai. To review the core algorithmic implementation, please check fedml_api/distributed/lightsecagg.

Running Script

cd ./fedml_experiments/distributed/lightsecagg

# MNIST
sh run_lightsecagg_distributed_pytorch.sh 4 4 lr hetero 10 1 64 0.1 mnist "./../../../data/MNIST" sgd 0

# CIFAR-10
sh run_lightsecagg_distributed_pytorch.sh 8 8 resnet56 homo 100 20 64 0.001 cifar10 ./../../../data/cifar10 adam 0
nohup sh run_lightsecagg_distributed_pytorch.sh 8 8 resnet56 homo 100 20 64 0.001 cifar10 ./../../../data/cifar10 adam 0 > ./lightsecagg_lr001.log 2>&1 &