FADngs: Federated Learning for Anomaly Detection

Code for our paper on anomaly detection in federated learning settings, titled FADngs. This repository support self-supervised training of networks for federated anomaly detection.

1. Preparation

Environments

Start by installing all dependencies.

pip install -r requirements.txt

Datasets

We need Cifar-10 and Cifar-100 datasets for experiments. Please download the above datasets to ~/data, or automatically download when running our code.

2. Training

To train the models using method FADngs in the paper, run this command:

CUDA_VISIBLE_DEVICES=0,1 python federated_train_sed.py --arch <NETWORK> --training-mode simclr_CSI --shift_trans_type rotation 
--clusters 10 --ratio_pollution 0.01

Option:

• --exp-name: Experiment name. We will log results into a text file of this name.
• --training-mode: Choose from ("SimCLR", "SupCon", "SupCE"). This will choose the right network modules for the checkpoint.
• --arch: Choose from available architectures: "resnet18", "resnet34", "resnet50", "resnet101".
• --shift_trans_type: Type of shifting transformations. Choose from ("rotation", "cutperm").
• --clusters: number of clusters when performing k-means clustering.
• --ratio_pollution: ratio of anomaly samples in datasets