model-watermarking

This is the code of ICLR 2022 Oral paper 'Non-Transferable Learning: A New Approach for Model Ownership Verification and Applicability Authorization'.

[DAC'24] EmMark: Robust Watermarks for IP Protection of Embedded Quantized Large Language Models

Code and paper source for dynamics-level watermarking of flow matching models with random codes.

Structure-Aware Fingerprints (SAF) for black-box ownership verification of GNNs — motif- and role-based fingerprints robust to model extraction attacks, with controlled ablations.