Recursive Euclidean Distance-based Robust Aggregation Technique for Federated Learning

Introduction

This is a PyTorch implementation of our paper. * This repository used code from federated learning and Attack-Resistant-Federated-Learning

Citing Recursive Euclidean Distance Based Robust Aggregation Technique For Federated Learnin

`` @article{recursiveEuclideanDistanc, title={Recursive Euclidean Distance Based Robust Aggregation Technique For Federated Learning}, author={Charuka Herath, Yogachandran Rahulamathavan, Xiaolan Liu}, journal={https://arxiv.org/abs/2303.11337}, year={2023} }

@article{FheFL, title={FheFL: Fully Homomorphic Encryption Friendly Privacy-Preserving Federated Learning with Byzantine Users}, author={Yogachandran Rahulamathavn, Charuka Herath, Xiaolan Liu, Sangarapillai Lambotharan, Carsten Maple}, journal={https://arxiv.org/abs/2306.05112}, year={2023} }

## Requirements: Software

1. Pytorch from [the offical repository](https://pytorch.org/).
2. Install tensorboardX.

pip install tensorboardX

## Preparation for Training & Testing
1. The code will automatically download MNIST dataset.

## Usage
### Label-flipping attack experiments
Label Flipping attack on MNIST

python main_nn.py --model smallcnn --epochs 100 --gpu 0 --iid 0 --fix_total --frac 0.1 --num_attackers 4 --attacker_ep 10 --num_users 100 --attack_label 1 --agg euclidean_distance --noise 1

Change `--agg` tag to select aggregation algorithm and change `--num_attackers` to specify the number of attackers.

### Run based on Dynamic data distribution setting

python main_nn.py --model smallcnn --epochs 100 --gpu 0 --iid 0 --is_dynamic 1 --fix_total --frac 0.1 --num_attackers 0 --attacker_ep 0 --num_users 100 --attack_label 1 --agg average --is_dynamic 1

### LSFE approach with noise added secure aggregation
#### Label flipping attack --> 10 attackers and 5 attacker epochs

python main_nn.py --model smallcnn --epochs 1 --gpu 0 --iid 0 --fix_total --frac 0.1 --num_attackers 10 --attacker_ep 5 --num_users 100 --attack_label 1 --agg euclidean_distance --noise 1

#### Backdoor poisoning attack

python main_nn.py --model smallcnn --epochs 100 --gpu 0 --iid 0 --fix_total --frac 0.1 --num_attackers 10 --attacker_ep 5 --num_users 100 --attack_label 1 --agg euclidean_distance --noise 1 --is_backdoor True

### LSFE approach with added noise

python main_nn.py --model smallcnn --epochs 1 --gpu 0 --iid 0 --fix_total --frac 0.1 --num_attackers 10 --attacker_ep 5 --num_users 100 --attack_label 1 --agg lsfe --noise 0

## LSFE scripts

### mnist

python main_nn.py --model smallcnn --epochs 200 --gpu 0 --iid 1 --fix_total --frac 0.1 --num_attackers 20 --attacker_ep 5 --num_users 100 --attack_label 1 --agg euclidean_distance --noise 0 --dataset mnist

python main_nn.py --model smallcnn --epochs 200 --gpu 0 --iid 1 --fix_total --frac 0.1 --num_attackers 20 --attacker_ep 5 --num_users 100 --attack_label 1 --agg average --noise 0 --dataset mnist

python main_nn.py --model smallcnn --epochs 200 --gpu 0 --iid 1 --fix_total --frac 0.1 --num_attackers 20 --attacker_ep 5 --num_users 100 --attack_label 1 --agg median --noise 0 --dataset mnist

python main_nn.py --model smallcnn --epochs 200 --gpu 0 --iid 1 --fix_total --frac 0.1 --num_attackers 20 --attacker_ep 5 --num_users 100 --attack_label 1 --agg trimmed_mean --noise 0 --dataset mnist

python main_nn.py --model smallcnn --epochs 200 --gpu 0 --iid 1 --fix_total --frac 0.1 --num_attackers 20 --attacker_ep 5 --num_users 100 --attack_label 1 --agg krum --noise 0 --dataset mnist

### CIFAR-10

python main_nn.py --model smallcnn --epochs 200 --gpu 0 --iid 1 --fix_total --frac 0.1 --num_attackers 20 --attacker_ep 5 --num_users 100 --attack_label 3 --agg euclidean_distance --noise 0 --dataset cifar

python main_nn.py --model smallcnn --epochs 200 --gpu 0 --iid 1 --fix_total --frac 0.1 --num_attackers 20 --attacker_ep 5 --num_users 100 --attack_label 3 --agg average --noise 0 --dataset cifar

python main_nn.py --model smallcnn --epochs 200 --gpu 0 --iid 1 --fix_total --frac 0.1 --num_attackers 20 --attacker_ep 5 --num_users 100 --attack_label 3 --agg median --noise 0 --dataset cifar

python main_nn.py --model smallcnn --epochs 200 --gpu 0 --iid 1 --fix_total --frac 0.1 --num_attackers 20 --attacker_ep 5 --num_users 100 --attack_label 3 --agg trimmed_mean --noise 0 --dataset cifar

python main_nn.py --model smallcnn --epochs 200 --gpu 0 --iid 1 --fix_total --frac 0.1 --num_attackers 20 --attacker_ep 5 --num_users 100 --attack_label 3 --agg krum --noise 0 --dataset cifar

### CIFAR-100

python main_nn.py --model smallcnn --epochs 200 --gpu 0 --iid 1 --fix_total --frac 0.1 --num_attackers 5 --attacker_ep 5 --num_users 100 --attack_label 3 --agg euclidean_distance --noise 0 --dataset cifar-100

python main_nn.py --model smallcnn --epochs 200 --gpu 0 --iid 1 --fix_total --frac 0.1 --num_attackers 5 --attacker_ep 5 --num_users 100 --attack_label 3 --agg average --noise 0 --dataset cifar-100

python main_nn.py --model smallcnn --epochs 200 --gpu 0 --iid 1 --fix_total --frac 0.1 --num_attackers 5 --attacker_ep 5 --num_users 100 --attack_label 3 --agg median --noise 0 --dataset cifar-100

python main_nn.py --model smallcnn --epochs 200 --gpu 0 --iid 1 --fix_total --frac 0.1 --num_attackers 5 --attacker_ep 5 --num_users 100 --attack_label 3 --agg trimmed_mean --noise 0 --dataset cifar-100

python main_nn.py --model smallcnn --epochs 200 --gpu 0 --iid 1 --fix_total --frac 0.1 --num_attackers 5 --attacker_ep 5 --num_users 100 --attack_label 3 --agg krum --noise 0 --dataset cifar-100