Multi-Factor Authentication and Secure Multi-Party Computation for Privacy-Preserving Federated Learning

Privacy-Preserving Federated Learning (PPFL) is an emerging secure distributed learning paradigm that aggregates locally trained gradients from users into a federated model.However, the server poses a potential risk to participants' privacy through inference attacks, and the quality of data provided by participants can be inconsistent.A key challenge in Federated Learning (FL) is the excessive involvement of low-quality data in the training process, which can render the model ineffective.This research proposes a combination of Multi-Factor Authentication and Secure Multi-Party Computation (MFMP) to ensure secure client model updates and prevent unauthorized data from being included in the global model.The local model, which utilizes a Convolutional Neural Network (CNN), effectively extracts image patterns and sequential data.Data is initially collected from the MNIST and CIFAR-10 datasets, with preprocessing using standard scaling to categorize the data.Encryption using the Advanced Encryption Standard (AES) provides symmetric encryption for efficient and secure image processing while minimizing computational overhead.The proposed method achieves better accuracy of 99.75%, 98.84%, and 92.33% on the MNIST, CIFAR-10, and Fashion-MNIST datasets, respectively, when compared with existing methods such as the Federated Learning framework Communication-efficient and Privacy-preserving (FLCP).