Researchers Indurthi Srikanth and Ganesh Reddy Karri have introduced a software-defined networking framework that combines federated deep learning with cryptographic protocols to detect multi-vector distributed denial-of-service attacks and safeguard over-the-air firmware updates in Internet of Things environments. Evaluated across diverse Internet of Things datasets, the system attained an attack classification accuracy of up to 98.8 percent while preserving privacy at local edge gateways.
The research addresses emerging vulnerabilities brought about by the exponential development of Internet of Things devices. As connected hardware proliferates, networks face greater exposure to serious cyber threats that jeopardize overall network stability as well as the delivery of over-the-air firmware updates. Conventional security mechanisms deployed across these diverse settings frequently struggle with limited scalability, heightened privacy concerns, and insufficient real-time detection capabilities when managing heterogeneous environments.
To resolve these interrelated challenges, the authors developed a novel software-defined networking security framework centered on the Multi-expert dual stream heterogeneous Federated learning-based Convolutional Transformer model, abbreviated as MFCT. This architectural model is designed to perform collaborative attack classification directly across distributed edge gateways. By utilizing a federated learning paradigm, the design enables localized model training and classification while preserving data privacy at the edge gateways without transmitting sensitive raw device records.
Data confidentiality and secure collaborative aggregation are handled through Homomorphic Adaptive Elliptic Curve Cryptography, designated as HA-ECC. This cryptographic implementation secures cloud storage and enables encrypted model aggregation by permitting mathematical computations directly on encrypted data, thereby safeguarding sensitive transmission records. During evaluation, the HA-ECC component achieved encryption times between 1.2 and 3.0 seconds and decryption times between 1.2 and 1.8 seconds, demonstrating practical computational efficiency.
To streamline training operations and enhance computational efficiency, feature selection within the framework is optimized using the Adaptive Projection-iTerative-Methods-based Optimizer, or APTMO. In addition, the system incorporates Local Interpretable Model-Agnostic Explanations, known as LIME, to foster trust and system transparency. By generating human-understandable insights into the deep learning model’s predictive classifications, the explainability framework allows administrators to interpret how potential multi-vector attacks are identified.
Empirical evaluations across diverse Internet of Things datasets confirmed the framework’s reliable operational capacity. Alongside its peak classification accuracy of up to 98.8 percent, the proposed MFCT framework recorded an overall threat detection rate of around 98 percent. The security architecture also achieved an over-the-air firmware update integrity level of approximately 97.8 percent, confirming that remote firmware distributions remain protected against unauthorized interference during transit.
Published in the Journal of Cloud Computing, the study illustrates how uniting federated learning, adaptive cryptography, and explainable artificial intelligence can address persistent vulnerabilities in distributed network infrastructures. The authors conclude that their integrated architecture delivers robust and real-time multi-vector attack detection, secure firmware updates, and computational efficiency, while effectively maintaining scalability, interpretability, and privacy preservation across heterogeneous Internet of Things environments.
Subject of Research: Federated deep learning framework for IoT DDoS detection and secure firmware updates
Article Title: Secure federated multi-vector DDoS detection with explainable SDN-IoT framework: fortified OTA updates
Article References: Srikanth, I., & Karri, G. R. (2026). Secure federated multi-vector DDoS detection with explainable SDN-IoT framework: fortified OTA updates. Journal of Cloud Computing. https://doi.org/10.1186/s13677-026-00994-3
Image Credits: AI Generated
DOI: 10.1186/s13677-026-00994-3
Keywords: Internet of Things, federated learning, software-defined networking, DDoS detection, over-the-air updates, firmware security, edge gateways, Convolutional Transformer, Homomorphic Adaptive Elliptic Curve Cryptography, APTMO optimizer, LIME explainability, cyber threat detection, model aggregation, data privacy, network stability, cloud security
Cite Scienmag News
Blake Davidson. (November 18, 2026). Federated Deep Learning Framework Secures Connected Devices Against Multi-Vector Cyber Attacks. Scienmag. https://scienmag.com/federated-deep-learning-framework-secures-connected-devices-against-multi-vector-cyber-attacks/
Blake Davidson. "Federated Deep Learning Framework Secures Connected Devices Against Multi-Vector Cyber Attacks." Scienmag, 18 November 2026, https://scienmag.com/federated-deep-learning-framework-secures-connected-devices-against-multi-vector-cyber-attacks/. Accessed 18 November 2026.
Blake Davidson. "Federated Deep Learning Framework Secures Connected Devices Against Multi-Vector Cyber Attacks." Scienmag. November 18, 2026. https://scienmag.com/federated-deep-learning-framework-secures-connected-devices-against-multi-vector-cyber-attacks/








