TAG: Gradient Attack on Transformer-based Language Models

Keyphrases

• Attack Algorithm 25%
• Cosine Similarity 25%
• Data Privacy 25%
• Dictionary Size 25%
• Distributed Learning 25%
• Gradient Leakage Attack 50%
• Gradient-based Attack 100%
• Ground-truth Labels 25%
• Large Animal Model 25%
• Leakage Mechanism 25%
• Local Training 25%
• Privacy Enhancement 25%
• Private Training 75%
• ROUGE-2 25%
• Token 25%
• Token Embedding 25%
• Training Data 100%
• Training Process 25%
• Transformer 25%
• Transformer-based Language Models 100%
• Weight Distribution 25%

Computer Science

• Cosine Similarity 25%
• Data Privacy 25%
• Dictionary Size 25%
• Distributed Learning 25%
• Experimental Result 25%
• Language Modeling 100%
• Training Data 100%
• Training Process 25%