Development of an API for Integration with Cyber Threat Intelligence Platforms through Federated Learning and Gradient Sparsification
Sharmila Devi Mandalapu, Deevi Radha Rani · 2025
Extending CTI platforms with Federated Learning and Gradient Sparsification, as well as integrating a GRU-SVM model into a cybersecurity application via an API also solves the increasing challenge of secure, efficient threat detection in evolving security architectures. Standard centralized models share highly important information and lack communication efficiency, which makes them ill-suited for the given CTI platforms’ decentralized structure. The Federated Learning technique adopted here in this API ensures that multiple CTI clients can work on the same model training without passing the raw data meaning data privacy and regulatory compliancy is preserved. Gradient Sparsification enhances the optimization exercise by cutting down the amount of data conveyed in the process of updating a model, by passing only the most crucial gradient information. Approach that is compared with other methods will significantly reduce the communication overhead and increases training speed while keeping high accuracy of the target model. The reason for selecting the GRU-SVM model is that it is ideal for processing sequential data and has a high accuracy and robust classification function in detecting the changing patterns in cyber threats. Evaluations show that the API performs with high accuracy of communication, and can accommodate many clients in a distributed environment, which proves this approach is cost efficient for real-time threat intelligence information sharing. This API combines GRU-SVM with Federated Learning and Gradient Sparsification and it will help organizations improve the security and the efficiency of the CTI systems and tools thus improving the overall response to new threats while keeping data privacy intact.