DeepIPR: Deep Neural Network Ownership Verification With Passports

Area of Science:

• Computer Science
• Artificial Intelligence
• Machine Learning

Background:

• Deep Neural Networks (DNNs) require significant investment.
• Protecting DNN intellectual property from illegal copying and redistribution is essential.
• Existing DNN watermarking methods are vulnerable to ambiguity attacks.

Purpose of the Study:

• To investigate the vulnerabilities of DNN watermarking methods to ambiguity attacks.
• To propose novel, robust, and resilient DNN ownership verification schemes.
• To ensure the integrity and ownership of developed DNN models.

Main Methods:

• Investigated existing watermark-based DNN ownership verification techniques.
• Developed novel passport-based DNN ownership verification schemes.
• Designed DNN models to deter forged passports by degrading inference performance.

Main Results:

• Ambiguity attacks pose significant threats to current DNN watermarking methods.
• Proposed passport-based schemes demonstrate robustness against network modifications.
• The novel schemes are resilient to ambiguity attacks, effectively verifying DNN ownership.
• Experimental results validate the effectiveness of the passport-based approach.

Conclusions:

• Existing DNN watermarking techniques are insufficient against sophisticated attacks.
• Passport-based DNN ownership verification offers a more secure solution.
• The proposed method ensures DNN integrity and protects intellectual property effectively.