Privacy-Preserving Federated Learning Using Threshold Homomorphic Encryption

Federated Learning (FL) has occurred for example a privacy-preserving dispersed machine learning paradigm, allowing multiple participants to collaboratively train models without sharing raw information. However, FL remains susceptible to safety and privacy threats, including inference attacks and information exposure. Near address these challenges, this paper recommends the integration of Threshold Homomorphic Encryption (THE) to increase the privacy and safety of FL systems. THE allows encoded model updates to stand aggregated securely although ensuring that decryption demands collaboration from multiple gatherings, thereby preventing any only entity from accessing sensitive data. The proposed method is evaluated on the UNSW-NB15 dataset, representative its effectiveness in preservative model performance while significantly improving data confidentiality. Investigational results show that the THE-based FL framework alleviates privacy risks, reduces adversarial dangers, and ensures scalable protected computation. This paper underwrites to advancing privacy-aware spread learning and surfaces the way for safe AI applications in complex domains such as cybersecurity.