Research on Privacy Protection in Federated Learning Combining Distillation Defense and Blockchain

Changxu Wan,Ying Wang,Jianbo Xu,Junjie Wu,Tiantian Zhang,Yulong Wang

DOI: https://doi.org/10.3390/electronics13040679

IF: 2.9

2024-02-07

Electronics

Abstract:Traditional federated learning addresses the data security issues arising from the need to centralize client datasets on a central server for model training. However, this approach still poses privacy protection risks. For instance, central servers cannot verify privacy leaks resulting from poisoning attacks by malicious clients. Additionally, adversarial sample attacks can infer specific samples from the original data by testing the local models on client devices. This paper proposes a federated learning privacy protection method combining distillation defense technology with blockchain architecture. The method utilizes distillation defense technology to reduce the sensitivity of client devices participating in federated learning to perturbations and enhance their ability to resist adversarial sample attacks locally. This not only reduces communication overhead and improves learning efficiency but also enhances the model's generalization ability. Furthermore, the method leverages the "decentralized" nature of blockchain architecture as a trusted record-keeping mechanism to audit information interactions among clients and shared model parameters. This addresses privacy leakage issues resulting from poisoning attacks by some clients during the model construction process. Simulation experiment results demonstrate that the proposed method, compared with traditional federated learning, ensures model convergence, detects malicious clients, and improves the participation level of highly reputable clients. Moreover, by reducing the sensitivity of local clients to perturbations, it enhances their ability to effectively resist adversarial sample attacks.

engineering, electrical & electronic,computer science, information systems,physics, applied

What problem does this paper attempt to address?

The paper attempts to address the issue of how to protect participants' privacy in federated learning. Specifically, traditional federated learning, although reducing the risk of data storage by training models on local devices, still has some privacy protection risks. For example, the central server cannot verify poisoning attacks initiated by malicious clients, leading to privacy leaks, and adversarial sample attacks can infer specific samples in the original data by testing the local model. To address these issues, the study proposes a federated learning privacy protection method that combines distillation defense technology and blockchain architecture.

The main contributions of this method include:
1. Utilizing distillation defense technology to reduce the sensitivity of client devices participating in federated learning to perturbations, enhancing their ability to resist adversarial sample attacks, while reducing communication overhead, improving learning efficiency, and enhancing the generalization ability of the model.
2. Leveraging the "decentralized" nature of blockchain as a trusted recording mechanism to audit the information exchange and shared model parameters between clients, thereby addressing the privacy leakage issues caused by poisoning attacks during the model building process by some clients.

By combining these technologies, the study aims to provide a new solution to achieve a balance between data privacy protection and model performance in federated learning.