research18h
Giskard : Byzantine Robust and Confidential Aggregation for Large-Scale Decentralized Learning
Researchers propose Giskard, a method for confidential and Byzantine-robust aggregation in decentralized learning. Giskard enables secure parameter sharing while filtering out malicious contributions. You can apply this to protect decentralized machine learning from attacks and data breaches.
Key takeaways
- Giskard provides confidentiality via cryptographic techniques.
- Giskard detects and filters Byzantine attacks.
- Decentralized learning can be secured against malicious actors.