CYBRIA - Pioneering Federated Learning for Privacy-Aware Cybersecurity with Brilliance
Pratik Thantharate, T Anurag
Abstract
Centralized machine learning approaches for cyber-security raise significant privacy and security concerns due to raw data aggregation from distributed sources. This paper presents Cybria, a federated learning framework for collaborative cyber threat detection without compromising confidential data. The decentralized approach trains models on local data distributed across clients and shares only intermediate model updates to generate an integrated global model. We develop a federated learning architecture tailored for privacy-preserving intrusion detection. Comparative evaluations on the Bot-IoT dataset demonstrate that Cybria's federated model achieves 89.6% accuracy compared to 81.4% for a centralized deep neural network. The ~10% improvement highlights the benefits of collective learning from decentralized data for cyber defense applications. However, real-world deployment faces challenges like statistical heterogeneity, systemic bias, and data poisoning attacks. Advances in secure aggregation, differential privacy, and adversarial defenses are crucial to robust large-scale adoption. With thoughtful human-centric design, federated intelligence paves the path for an ecosystem approach to security where organizations collectively build threat awareness without centralizing data.