Byzantine-Robust Federated Fire Detection with a Rotating Coordinator
Federated learning is applied to indoor fire detection using edge cameras, addressing privacy concerns around sensitive footage that cannot be centralized A curated indoor fire-detection dataset was assembled from eight public sources to support this research Model updates are compressed up to 10x with only a small loss in balanced accuracy, making the system edge-deployable and bandwidth-efficient A semi-decentralized Byzantine-robust FL method combines history-aware aggregation with a rotating
Analysis
TL;DR
- Federated learning is applied to indoor fire detection using edge cameras, addressing privacy concerns around sensitive footage that cannot be centralized
- A curated indoor fire-detection dataset was assembled from eight public sources to support this research
- Model updates are compressed up to 10x with only a small loss in balanced accuracy, making the system edge-deployable and bandwidth-efficient
- A semi-decentralized Byzantine-robust FL method combines history-aware aggregation with a rotating coordinator, eliminating the single point of failure inherent in fixed-server architectures
- The rotating-coordinator approach matches the fixed-server counterpart in accuracy and detection speed, validated on a held-out test set and a six-node physical cloud deployment
Why It Matters
This work addresses three critical practical barriers in deploying federated learning for safety-critical applications: bandwidth constraints, security against Byzantine attacks, and the fragility of centralized aggregation servers. For AI practitioners building edge-based detection systems, it demonstrates that robustness and efficiency can coexist without sacrificing model performance.
Technical Details
- Dataset: A curated indoor fire-detection dataset assembled from eight public sources, providing a benchmark for federated fire detection research
- Model Compression: An edge-deployable detector with model updates compressed up to 10x, incurring only a small loss in balanced accuracy — critical for limited uplink bandwidth environments
- Byzantine-Robust FL Method: A semi-decentralized approach combining history-aware aggregation with a rotating coordinator; this design evicts stealthy attacks that per-round filters miss and removes the fixed-server single point of failure
- Evaluation: Tested on a held-out test set where the rotating-coordinator method matched the fixed-server counterpart in both accuracy and detection speed; feasibility confirmed via a physically distributed six-node cloud deployment
Industry Insight
- Safety-critical FL deployments (fire detection, industrial monitoring) should prioritize rotating-coordinator architectures over fixed aggregation servers to eliminate single points of failure and improve resilience against coordinated Byzantine attacks
- Model compression strategies that preserve balanced accuracy under high compression ratios (10x+) are viable for edge-deployed systems, enabling practical bandwidth-constrained deployments
- The assembly of multi-source curated datasets from existing public data offers a reproducible blueprint for domain-specific FL research where centralized data collection is infeasible or privacy-prohibited
Disclaimer: The above content is generated by AI and is for reference only.