Federated Learning Poisoning Attack

Required Packages

pika
torch
torchvision
numpy
requests
tqdm
pyyaml

Set up a RabbitMQ server for message communication over the network environment. docker-compose.yaml file:

version: '3'

services:
  rabbitmq:
    image: rabbitmq:management
    container_name: rabbitmq
    ports:
      - "5672:5672"   # RabbitMQ main port
      - "15672:15672" # Management UI
    environment:
      RABBITMQ_DEFAULT_USER: user
      RABBITMQ_DEFAULT_PASS: password
    volumes:
      - rabbitmq_data:/var/lib/rabbitmq
volumes:
  rabbitmq_data:
    driver: local

Then run the RabbitMQ container

docker-compose up -d

Configuration

Application configuration is in the config.yaml file:

name: Federated Learning poisoning attack testbed
server:   # server configuration
  num-round: 2  # number of training rounds
  clients: 3    # number of FL clients
  mode: hyper   # server mode: `fedavg` or `hyper`
  model: RNNModel     # class name of DNN model
  data-name: ICU      # training data
  parameters:
    load: True     # allow to load parameters file
  validation: True  # allow to validate on server-side
  ### algorithm
  data-distribution:      # data distribution config
    num-data-range:       # minimum and maximum number of label's data
      - 0
      - 500
  genuine-rate: 0.5       # genuine clients rate and send to malicious clients 
  random-seed: 1

rabbit:   # RabbitMQ connection configuration
  address: 127.0.0.1    # address
  username: admin
  password: admin

log_path: .   # logging directory

learning:
  epoch: 5          # client local epochs
  learning-rate: 0.01
  hyper-lr: 0.0005      # learning rate of hyper model
  momentum: 0.5
  batch-size: 256

This configuration is use for server and all clients.

List of DNN model

For ICU

CNNModel
RNNModel
TransformerModel

How to Run

Alter your configuration, you need to run the server to listen and control the request from clients.

Server

python server.py

Client

python client.py

If using a specific device configuration for the training process, declare it with the --device argument when running the command line:

python client.py --device cpu

Client attack

Random

python client.py --attack True --attack_mode Random --attack_round 30 --attack_args 0.5

LIE

python client.py --attack True --attack_mode LIE --attack_round 1 --attack_args 0.74

Min-Max

python client.py --attack True --attack_mode Min-Max --attack_round 2

Min-Sum

python client.py --attack True --attack_mode Min-Sum --attack_round 2

Opt-Fang

python client.py --attack True --attack_mode Opt-Fang --attack_round 2

Parameter Files

On the server, the *.pth files are saved in the main execution directory of server.py after completing one training round.

If the *.pth file exists, the server will read the file and send the parameters to the clients. Otherwise, if the file does not exist, a new DNN model will be created with fresh parameters. Therefore, if you want to reset the training process, you should delete the *.pth files.