Federated learning uses physical laws to secure critical infrastructure anomaly detection

Researchers have developed a novel federated learning approach that uses physical invariants, such as conservation of mass, to secure anomaly detection models in critical water infrastructure. This method allows operators to train shared intrusion detection models without revealing sensitive operational data. By automatically mining these invariants from clean operational data, the system can verify the integrity of model updates and distinguish between honest contributions and those from data poisoning attacks. The approach has been tested on water testbeds like SWaT and WADI, demonstrating its effectiveness in rejecting fabricated telemetry and improving detection rates for various attacks, with compliance verifiable through zero-knowledge proofs. AI

IMPACT Enhances security for critical infrastructure by enabling collaborative model training without compromising data privacy.

RANK_REASON Academic paper detailing a new method for federated learning in a specific application domain. [lever_c_demoted from research: ic=1 ai=1.0]

Read on arXiv cs.LG →

• critical water infrastructure
• FedAvg
• Federated learning
• FoolsGold
• Jeff Nijsse
• Physics-Attested Federated Learning
• WADI
• zk-SNARKs

• paper
• safety
• infra

AI-generated summary · Google Gemini · from 1 sources. How we write summaries →