Researchers have detailed new adversarial attacks targeting federated learning setups for Generative Adversarial Networks (GANs). These attacks involve malicious clients manipulating data by flipping labels or oversampling poisoned samples during local training. The objective is to skew the global generator's output, causing it to map target labels to incorrect classes. The study quantifies the impact using Kullback-Leibler divergence, showing that while semantic damage increases linearly with poisoning strength, deviations from the true distribution grow quadratically, making detection challenging. AI
IMPACT Highlights vulnerabilities in federated learning for GANs, potentially impacting secure model training and data privacy.
RANK_REASON Academic paper detailing novel adversarial attacks on federated learning models.
Read on Hugging Face Daily Papers →
- CIFAR-10
- federated learning
- FEMNIST
- Gans
- Kullback--Leibler divergence
- label flipping
- MNIST database
- oversampling
- Hugging Face
AI-generated summary · Google Gemini · from 2 sources. How we write summaries →
