With the rapid development of machine learning and growing concerns about data privacy, federated learning has become an increasingly prominent focus. However, challenges such as attacks on model parameters and the lack of incentive mechanisms hinder the effectiveness of federated learning. Therefore, we propose a Privacy Protected Blockchain-based Federated Learning Model (PPBFL) to enhance the security of federated learning and promote the active participation of nodes in model training. Blockchain ensures that model parameters stored in the InterPlanetary File System (IPFS) remain unaltered. A novel adaptive differential privacy addition algorithm is simultaneously applied to local and global models, preserving the privacy of local models and preventing a decrease in the security of the global model due to the presence of numerous local models in federated learning. Additionally, we introduce a new mix transactions mechanism to better protect the identity privacy of local training clients. Security analysis and experimental results demonstrate that PPBFL outperforms baseline methods in both model performance and security.

翻译：随着机器学习的迅速发展和数据隐私问题的日益突出，联邦学习已成为一个日益重要的研究重点。然而，模型参数攻击和激励机制缺失等挑战阻碍了联邦学习的有效性。为此，我们提出了一种基于区块链的隐私保护联邦学习模型（PPBFL），以增强联邦学习的安全性并促进节点积极参与模型训练。区块链确保存储在星际文件系统（IPFS）中的模型参数不被篡改。我们同时将一种新颖的自适应差分隐私添加算法应用于局部模型和全局模型，既保护了局部模型的隐私，又防止了联邦学习中因大量局部模型的存在而导致全局模型安全性下降。此外，我们引入了一种新的混合交易机制，以更好地保护本地训练客户端的身份隐私。安全分析和实验结果表明，PPBFL在模型性能和安全性方面均优于基线方法。