Federated Learning with Secure Aggregation for Privacy-Preserving Deep Learning in IoT Environments
Vinay Kumar Kasula, Mounica Yenugula, Bhargavi Konda, Akhila Reddy Yadulla, Chaitanya Tumma, Sarath Babu Rakki · 2025
To effectively handle the massive data generated by large-scale IoT applications, deep learning has been widely applied in IoT environments. However, deep models face security threats such as inference attacks and model reverse-engineering during the training process, which can lead to the leakage of original data fed into the model. Federated Learning (FL) with Secure Aggregation offers a robust solution to this issue by enabling distributed model training without sharing raw data. This method ensures privacy by securely aggregating model updates from IoT devices, mitigating risks of data exposure during transmission. To further enhance model performance, a regularization term is introduced to address overfitting and facilitate the learning of significant features. Experimental results demonstrate that the proposed approach effectively enhances the generalization capability of the model. As the number of iterations increases, the accuracy difference between the FL-trained model and the model trained on centralized data remains below 0.5%. Therefore, the proposed method not only protects user privacy but also ensures model usability, achieving a balance between privacy and utility.