Related Experiment Videos
DBST-FL: Dynamic Behavioural and Semantic Trust for Robust Federated Learning in Industrial IoT
Ammar Alazab1, Abin Kumbalapalliyil Tom1, Tony Jan1
1Centre for Artificial Intelligence Research and Optimization (AIRO), Torrens University Australia (TUA), 46-52 Mountain Street, Ultimo, NSW 2007, Australia.
Sensors (Basel, Switzerland)
|August 13, 2026
Summary
Federated learning in Industrial IoT is enhanced by DBST-FL, a novel framework combining behavioural and semantic trust to defend against sophisticated attacks. This approach ensures robust collaborative model training without compromising data privacy.
Area of Science:
- Cybersecurity
- Machine Learning
- Industrial Internet of Things (IIoT)
Background:
- Federated learning (FL) enables collaborative model training in IIoT without raw data sharing.
- Existing FL defenses struggle against advanced poisoning and backdoor attacks due to single-dimensional trust assessments.
Purpose of the Study:
- To propose DBST-FL, a dynamic framework for robust federated learning in IIoT.
- To enhance defence mechanisms against sophisticated adversarial attacks in FL.
Main Methods:
- DBST-FL integrates behavioural trust (gradient alignment, consistency) and semantic trust (utility, stress validation).
- A non-compensatory multiplicative fusion mechanism combines trust scores.
- A trust-aware aggregation strategy is employed to mitigate malicious clients.
Main Results:
- DBST-FL demonstrates competitive performance against robust aggregation baselines.
- The framework shows superior resilience against adversarial and backdoor attacks.
- Achieves linear per-round computational complexity for scalability in IIoT.
Conclusions:
- Integrating behavioural and semantic trust provides an effective and scalable defence against advanced threats in FL.
- DBST-FL enhances the robustness of federated learning in IIoT environments.
