# Entangled Watermarks as a Defense against Model Extraction

arXiv: https://arxiv.org/abs/2002.12200 · HTML: https://mlsanity.com/paper/2002.12200/

## Key facts

- Cited by 149 papers in the map (data through 14 Oct 2026).
- First posted on arXiv 27 Feb 2020.

## Built on by (most cited first, 30 of 149)

- [A Recipe for Watermarking Diffusion Models](https://mlsanity.com/paper/2303.10137/) (2023)
- [Dataset Inference: Ownership Resolution in Machine Learning](https://mlsanity.com/paper/2104.10706/) (2021)
- [Unbiased Watermark for Large Language Models](https://mlsanity.com/paper/2310.10669/) (2023)
- [WAVES: Benchmarking the Robustness of Image Watermarks](https://mlsanity.com/paper/2401.08573/) (2024)
- [Protecting Language Generation Models via Invisible Watermarking](https://mlsanity.com/paper/2302.03162/) (2023)
- [Untargeted Backdoor Watermark: Towards Harmless and Stealthy Dataset Copyright Protection](https://mlsanity.com/paper/2210.00875/) (2022)
- [Proof-of-Learning: Definitions and Practice](https://mlsanity.com/paper/2103.05633/) (2021)
- [I Know What You Trained Last Summer: A Survey on Stealing Machine Learning Models and Defences](https://mlsanity.com/paper/2206.08451/) (2022)
- [SoK: How Robust is Image Classification Deep Neural Network Watermarking? (Extended Version)](https://mlsanity.com/paper/2108.04974/) (2021)
- [Are You Copying My Model? Protecting the Copyright of Large Language Models for EaaS via Backdoor Watermark](https://mlsanity.com/paper/2305.10036/) (2023)
- [Risk Taxonomy, Mitigation, and Assessment Benchmarks of Large Language Model Systems](https://mlsanity.com/paper/2401.05778/) (2024)
- [Black-box Dataset Ownership Verification via Backdoor Watermarking](https://mlsanity.com/paper/2209.06015/) (2022)
- [Fingerprinting Deep Neural Networks Globally via Universal Adversarial Perturbations](https://mlsanity.com/paper/2202.08602/) (2022)
- ["Real Attackers Don't Compute Gradients": Bridging the Gap Between Adversarial ML Research and Practice](https://mlsanity.com/paper/2212.14315/) (2022)
- [Image Watermarks are Removable Using Controllable Regeneration from Clean Noise](https://mlsanity.com/paper/2410.05470/) (2024)
- [Defending against Model Stealing via Verifying Embedded External Features](https://mlsanity.com/paper/2112.03476/) (2021)
- [Copy, Right? A Testing Framework for Copyright Protection of Deep Learning Models](https://mlsanity.com/paper/2112.05588/) (2021)
- [SSLGuard: A Watermarking Scheme for Self-supervised Learning Pre-trained Encoders](https://mlsanity.com/paper/2201.11692/) (2022)
- [Explanation as a Watermark: Towards Harmless and Multi-bit Model Ownership Verification via Watermarking Feature Attribution](https://mlsanity.com/paper/2405.04825/) (2024)
- [The Threat of Offensive AI to Organizations](https://mlsanity.com/paper/2106.15764/) (2021)
- [Watermarking Diffusion Model](https://mlsanity.com/paper/2305.12502/) (2023)
- [Dataset Inference for Self-Supervised Models](https://mlsanity.com/paper/2209.09024/) (2022)
- [Distillation-Resistant Watermarking for Model Protection in NLP](https://mlsanity.com/paper/2210.03312/) (2022)
- [Quantization Backdoors to Deep Learning Commercial Frameworks](https://mlsanity.com/paper/2108.09187/) (2021)
- [Watermarking Graph Neural Networks based on Backdoor Attacks](https://mlsanity.com/paper/2110.11024/) (2021)
- [Data Hiding with Deep Learning: A Survey Unifying Digital Watermarking and Steganography](https://mlsanity.com/paper/2107.09287/) (2021)
- [StolenEncoder: Stealing Pre-trained Encoders in Self-supervised Learning](https://mlsanity.com/paper/2201.05889/) (2022)
- [On the Difficulty of Defending Self-Supervised Learning against Model Extraction](https://mlsanity.com/paper/2205.07890/) (2022)
- [Increasing the Cost of Model Extraction with Calibrated Proof of Work](https://mlsanity.com/paper/2201.09243/) (2022)
- [MOVE: Effective and Harmless Ownership Verification via Embedded External Features](https://mlsanity.com/paper/2208.02820/) (2022)

## Abstract (arXiv)

Machine learning involves expensive data collection and training procedures. Model owners may be concerned that valuable intellectual property can be leaked if adversaries mount model extraction attacks. As it is difficult to defend against model extraction without sacrificing significant prediction accuracy, watermarking instead leverages unused model capacity to have the model overfit to outlier input-output pairs. Such pairs are watermarks, which are not sampled from the task distribution and are only known to the defender. The defender then demonstrates knowledge of the input-output pairs to claim ownership of the model at inference. The effectiveness of watermarks remains limited because they are distinct from the task distribution and can thus be easily removed through compression or other forms of knowledge transfer. We introduce Entangled Watermarking Embeddings (EWE). Our approach encourages the model to learn features for classifying data that is sampled from the task distribution and data that encodes watermarks. An adversary attempting to remove watermarks that are entangled with legitimate data is also forced to sacrifice performance on legitimate data. Experiments on MNIST, Fashion-MNIST, CIFAR-10, and Speech Commands validate that the defender can claim model ownership with 95% confidence with less than 100 queries to the stolen copy, at a modest cost below 0.81 percentage points on average in the defended model's performance.