Revealing Task-Relevant Model Memorization for Source-Protected Unsupervised Domain Adaptation

Keyphrases

• Unsupervised Domain Adaptation 100%
• Source Model 100%
• Memorization 100%
• Task-irrelevant 100%
• Generative Models 66%
• Privacy Protection 50%
• Source-free 50%
• Target Domain 33%
• Source Data 33%
• Domain Adaptation Method 33%
• Training Data 16%
• Training Model 16%
• Data Privacy 16%
• Transferability 16%
• Source Domain 16%
• Black Box 16%
• Domain Data 16%
• Domain Adaptation 16%
• Information Security 16%
• Model Performance 16%
• Cross-dataset Recognition 16%
• Generated Data 16%
• Desensitization 16%
• New Subjects 16%
• Privacy Disclosure 16%
• Unintended Memorization 16%
• Domain Model 16%
• Pseudo Data 16%
• Model Privacy 16%
• Access Model 16%
• Metric Scores 16%
• Inverse Process 16%
• Real Sources 16%

Computer Science

• Unsupervised Domain Adaptation 100%
• Generative Model 100%
• Training Data 25%
• Privacy Protection 25%
• Domain Adaptation 25%
• Information Security 25%
• Performance Model 25%
• Source Domain Data 25%
• Disclosure Privacy 25%