Private Federated Learning for APT Detection in Internet of Drones

The Internet of Drones (IoD) is a decentralized network that connects drones to controlled airspace. The connection of drones in these networks is through the Internet of Things. Hence, these networks are vulnerable to all the security and privacy threats that affect IoT networks. In addition, as the application of these networks is highly sensitive in many cases, there are greater potential security threats. The components of these networks work together to identify new and advanced threats. One of the ways to identify new and advanced threats in these networks is distributed machine learning where the data is sent to a central server to learn the general model. This model violates the privacy of network components. It also has a very high level of communication. On the other hand, the central server as the only point of failure may have many problems. In this case, federated learning helps distributed and decentralized networks to share their local model instead of sending their local and secret data. Since the shared models may also disclose some information, we propose a secure and privacy-preserving protocol based on homomorphic encryption. The protocol proposed was for federal learning model and detection of new and advanced threats in the Internet of Drones.