Agentic data science with the Data Agent Kit
cloud.google.com·7h ago
AI Research
arXiv CS.AI·Aug 10·5 min read
TOFD: Target-Oriented Feature Decoupling against Poisoning Attacks in Split Federated Learning
TL;DR
Target-Oriented Feature Decoupling (TOFD) is a new framework designed to enhance the security of Split Federated Learning against poisoning attacks, achieving superior robustness with minimal computational overhead.
✦ Why It Matters
Implement TOFD to enhance the security of your federated learning systems against poisoning attacks today.
Key Takeaways
- Adopt TOFD for improved security in federated learning environments.
- Utilize Margin Perturbation to identify potential attack targets effectively.
- Implement adaptive sample purification to filter out poisoned data.
- Leverage adversarial guidance for robust optimization against attacks.
How It Works
TOFD operates in three stages: Target Inference identifies attack targets by refining safe zones using Margin Perturbation, Sample Purification filters out poisoned data based on calibrated thresholds, and Decoupling Optimization uses an adversarial guidance model to mitigate the impact of detected attacks during the training process.