otechno/multi-model-fl repository overview

The code implements a robust Federated Learning (FL) method, called as multi-model based FL published by Somayeh Kianpisheh et al., in IEEE Global Communications Conference, 2024. For use of this code, please refer to the published paper as follows:

S. Kianpisheh et al., "Multi-Model based Federated Learning Against Model Poisoning Attack: A Deep Learning Based Model Selection for MEC systems," IEEE Global Communications Conf., 2024, pp. 1737-1742.

Abstract

Federated Learning (FL) enables training of a global model from distributed data. However, the singular-model based operation of FL is open with uploading poisoned models compatible with the global model structure and can be exploited as a vulnerability to conduct model poisoning attacks. A multi-model based FL enhances the opportunity of model poisoning attack mitigation. A master model is trained by a set of slave models. To enhance the opportunity of attack mitigation, the structure of client models dynamically change. The FL protocol consists of the steps as below: 0. Model Selection (it selects either master or slave model for a client)

1. Download Model
2. Local Training of Models
3. Attack Mitigation
4. Aggregation
5. Knowledge Transfer (transfer knowledge from slave model to master model)

Complementary Explanation

The code uses Gated Recurrent Unit with 32 and 28 neurons in hidden layer, as master and slave model, respectively. A random strategy for model selection, has been implemented. The published paper however, provides a deep learning based method.

Poisoning Attack

To simulate the attack, the method in the paper below, has been adopted: R. Al Mallah et al., “Untargeted poisoning attack detection in federated learning via behavior attestation,” IEEE Access, 2023.

Data

The data for train and test have been selected from open source CICDDoS 2019 dataset: I. Sharafaldin et al., “Developing realistic distributed denial of service attack dataset and taxonomy,” Conf. on Security Technology, 2019.

Run

1. Set the parameters in Config.py
2. Run main.py
3. Screen and Log.txt shows: a) log of the poisoning, model selection and mitigation within FL. b) The accuracy and the resilience to measure the performance of FL.