Flower Attacks: This is an extension of the Flower framework for Federated Learning. It provides an extenion module that allows the user to perform various attacks on the federated learning process.
Project description
Flower Attacks
This repository is an extension of the Flower Framework that makes possible creating, running, testing and simulating various adversary threats within the FLower Framework. The structure is made very similar to the strategy module to be easy and smooth for someone working with Flower to implement.
Features
-
Federated Learning Attacks: Implement and simulate various types of attacks on federated learning processes to assess their resilience and security
-
Integration with Flower: Seamlessly integrates with the Flower framework, allowing for easy experimentation and extension.
-
Extensible Design: Designed to be easily extended with new types of attacks or modifications to existing ones.
Installation
To install flwr_attacks
, you can use pip:
pip install flwr_attacks
Usage
After installation, you can use flwr_attacks
as part of your federated learning experiments. Here is a basic example of how to integrate it with your Flower-based federated learning setup:
Configuration for the attack (assuming cfg is an existing configuration object)
from flwr_attacks import MinMax, AttackServer, generate_cids
adversary_cids, benign_cids = generate_cids(NUM_CLIENTS, adversary_fraction=0.4)
all_cids = adversary_cids + benign_cids
Initialize the MinMax attack with your configuration
attack = MinMaxAttack(
adversary_fraction=0.2, # 20% of clients are adversaries
activation_round=5, # Activate attack at round 5
adversary_clients=adversary_cids, # by default the attack will be able to access only the adversary clients. Use the argument adversary_accessed_cids to add specific access.
)
strategy = ...
# Create the AttackServer with the specified attack and strategy
attack_server = AttackServer(
strategy=strategy,
attack=attack,
)
Use the server as in a typical Flower server
Use simulation
history = fl.simulation.start_simulation(
client_fn=client_fn,
clients_ids=all_cids,
config=fl.server.ServerConfig(num_rounds=cfg.num_rounds),
server=attack_server,
)
or start the server
fl.server.start_server(
server=attack_server,
)
Contributing
Contributions to flwr_attacks
are welcome! If you have a new attack implementation, improvements or bug fixes, open an issue or a pull request.
License
flwr_attacks
is released under the Flower's Apache-2.0 License. See the LICENSE file for more details.
Contact
For any questions or feedback, please contact Nassos Bountioukos Spinaris at nassosbountioukos@gmail.com.
Acknowledgments
Special thanks to the Flower framework team for providing a solid foundation for federated learning experiments.
Project details
Download files
Download the file for your platform. If you're not sure which to choose, learn more about installing packages.
Source Distribution
Built Distribution
Hashes for flwr_attacks-0.1.4-py3-none-any.whl
Algorithm | Hash digest | |
---|---|---|
SHA256 | 1399e71bfa7cf21a8a8f27c21e12447c168e5972d0e66ec31159be940dd6b510 |
|
MD5 | 1d3415a3f9f9b153c811a0edc54634eb |
|
BLAKE2b-256 | 7dc26f6eed620e2ee2c78c20646289ff75b32e81491be69d6afe74af02071168 |