Awesome

FoolsGold

A sybil-resilient distributed learning protocol that penalizes Sybils based on their gradient similarity.

FoolsGold is also described in two papers:

1. Peer-reviewed conference paper pdf:

"The Limitations of Federated Learning in Sybil Settings." 
Clement Fung, Chris J.M. Yoon, Ivan Beschastnikh.
23rd International Symposium on Research in Attacks, Intrusions and Defenses (RAID) 2020.

Bibtex:

@InProceedings{Fung2020,
  title     = {{The Limitations of Federated Learning in Sybil Settings}},
  author    = {Clement Fung and Chris J. M. Yoon and Ivan Beschastnikh},
  year      = {2020},
  series    = {RAID},
  booktitle = {Symposium on Research in Attacks, Intrusion, and Defenses},
}

2. Arxiv paper

Running a minimal MNIST example

Get the MNIST data.

Download and gunzip files from http://yann.lecun.com/exdb/mnist/
Download zipfile from https://git-disl.github.io/GTDLBench/datasets/mnist_datasets/
Move all unzipped contents to ML/data/mnist.

Clean the filenames (replace dot with dash) and prep the data:

cd ML/data/mnist
mv train-labels.idx1-ubyte train-labels-idx1-ubyte
mv train-images.idx3-ubyte train-images-idx3-ubyte
mv t10k-labels.idx1-ubyte t10k-labels-idx1-ubyte
mv t10k-images.idx3-ubyte t10k-images-idx3-ubyte
python parse_mnist.py

Create poisoned MNIST 1-7 data

From main directory navigate to the ML directory: cd ML/
Run: python code/misslabel_dataset.py mnist 1 7

Run FoolsGold

From main directory navigate to the ML directory: cd ML/
And run the following command for a 5 sybil, 1-7 attack on mnist.

python code/ML_main.py mnist 1000 5_1_7