REFIT: A Unified Watermark Removal Framework for Deep Learning Systems with Limited Data

Keyphrases

• Watermark Removal 100%
• Deep Learning System 100%
• Watermarking 60%
• Unlabeled Data 60%
• Training Data 40%
• Watermarking Scheme 40%
• Embedding Scheme 40%
• Watermark Embedding 40%
• Removal Attack 40%
• Elastic Weight Consolidation 40%
• Computationally Expensive 20%
• Copyright 20%
• Evaluation Model 20%
• Training Model 20%
• Adversary 20%
• Real Threats 20%
• Attack Scenario 20%
• Data Augmentation 20%
• Threat Model 20%
• Limited Training Data 20%
• Deep Neural Network 20%
• Watermarking Techniques 20%
• Training Deep Neural Networks 20%
• Copyright Infringement 20%
• Catastrophic Forgetting 20%
• Robust Watermark 20%
• Tuning Framework 20%

Computer Science

• Learning Systems 100%
• Training Data 100%
• Unlabeled Data 100%
• Deep Learning 100%
• Watermarking Scheme 66%
• Deep Neural Network 66%
• watermark embedding 66%
• Experimental Result 33%
• Evaluation Models 33%
• Watermarking 33%
• Attack Scenario 33%
• Threat Model 33%
• Data Augmentation 33%
• Watermarking Technique 33%
• Copyright Infringement 33%