DeepIP

DeepIP implements passport-based verification to protect deep neural network (DNN) intellectual property by embedding digital passports during training to enable ownership verification and detection of forged passports via performance-dependent validation.

Key Features:

• Passport-Based Verification: Embeds digital passports within DNN models during design and training to serve as secure identifiers for ownership verification.
• Robustness Against Modifications: Maintains verification capability even when the network undergoes modifications, preserving passport integrity despite alterations.
• Resilience to Ambiguity Attacks: Specifically designed to resist ambiguity attacks that introduce counterfeit watermarks to undermine ownership verification.
• Performance-Dependent Verification: Combines signature checks with performance validation such that forged passports cause significant degradation of model task performance.

Scientific Applications:

• Bioinformatics DNN protection: Protects proprietary deep learning models used in bioinformatics by enabling ownership verification and deterring unauthorized reuse.

Methodology:

Embeds digital passports during DNN training and verifies ownership by combining signature checks with performance-dependent validation, with experimental validation reported.