# THEMIS: Towards Practical Intellectual Property Protection for Post-Deployment On-Device Deep Learning Models

arXiv: https://arxiv.org/abs/2503.23748 · HTML: https://mlsanity.com/paper/2503.23748/

## Key facts

- Cited by 2 papers in the map (data through 14 Oct 2026).
- Placed in Watermarking generated content and models, one of 12 directions in LLM safety and security.
- Builds on 42 of its 67 references that are in the map.
- First posted on arXiv 31 Mar 2025.

## Placed in

- LLM safety and security / [Watermarking generated content and models](https://mlsanity.com/safety/watermarking-generated-content-and-models/)

## Builds on (most cited first, 30 of 42)

- [High-Resolution Image Synthesis with Latent Diffusion Models](https://mlsanity.com/paper/2112.10752/) (2021)
- [Distilling the Knowledge in a Neural Network](https://mlsanity.com/paper/1503.02531/) (2015)
- [Explaining and Harnessing Adversarial Examples](https://mlsanity.com/paper/1412.6572/) (2014)
- [Rethinking the Inception Architecture for Computer Vision](https://mlsanity.com/paper/1512.00567/) (2015)
- [MobileNetV2: Inverted Residuals and Linear Bottlenecks](https://mlsanity.com/paper/1801.04381/) (2018)
- [Fashion-MNIST: a Novel Image Dataset for Benchmarking Machine Learning Algorithms](https://mlsanity.com/paper/1708.07747/) (2017)
- [EfficientNetV2: Smaller Models and Faster Training](https://mlsanity.com/paper/2104.00298/) (2021)
- [Practical Black-Box Attacks against Machine Learning](https://mlsanity.com/paper/1602.02697/) (2016)
- [Image Segmentation Using Deep Learning: A Survey](https://mlsanity.com/paper/2001.05566/) (2020)
- [Stealing Machine Learning Models via Prediction APIs](https://mlsanity.com/paper/1609.02943/) (2016)
- [MobileBERT: a Compact Task-Agnostic BERT for Resource-Limited Devices](https://mlsanity.com/paper/2004.02984/) (2020)
- [DiffusionDB: A Large-scale Prompt Gallery Dataset for Text-to-Image Generative Models](https://mlsanity.com/paper/2210.14896/) (2022)
- [Turning Your Weakness Into a Strength: Watermarking Deep Neural Networks by Backdooring](https://mlsanity.com/paper/1802.04633/) (2018)
- [Knockoff Nets: Stealing Functionality of Black-Box Models](https://mlsanity.com/paper/1812.02766/) (2018)
- [Embedding Watermarks into Deep Neural Networks](https://mlsanity.com/paper/1701.04082/) (2017)
- [BadNL: Backdoor Attacks against NLP Models with Semantic-preserving Improvements](https://mlsanity.com/paper/2006.01043/) (2020)
- [PRADA: Protecting against DNN Model Stealing Attacks](https://mlsanity.com/paper/1805.02628/) (2018)
- [Entangled Watermarks as a Defense against Model Extraction](https://mlsanity.com/paper/2002.12200/) (2020)
- [A Survey of Privacy Attacks in Machine Learning](https://mlsanity.com/paper/2007.07646/) (2020)
- [Stealing Hyperparameters in Machine Learning](https://mlsanity.com/paper/1802.05351/) (2018)
- [Backdooring Convolutional Neural Networks via Targeted Weight Perturbations](https://mlsanity.com/paper/1812.03128/) (2018)
- [SoK: How Robust is Image Classification Deep Neural Network Watermarking? (Extended Version)](https://mlsanity.com/paper/2108.04974/) (2021)
- [How to Prove Your Model Belongs to You: A Blind-Watermark based Framework to Protect Intellectual Property of DNN](https://mlsanity.com/paper/1903.01743/) (2019)
- [Handcrafted Backdoors in Deep Neural Networks](https://mlsanity.com/paper/2106.04690/) (2021)
- [Cache Telepathy: Leveraging Shared Resource Attacks to Learn DNN Architectures](https://mlsanity.com/paper/1808.04761/) (2018)
- [Robust Watermarking of Neural Network with Exponential Weighting](https://mlsanity.com/paper/1901.06151/) (2019)
- [A First Look at Deep Learning Apps on Smartphones](https://mlsanity.com/paper/1812.05448/) (2018)
- [DeepPayload: Black-box Backdoor Attack on Deep Learning Models through Neural Payload Injection](https://mlsanity.com/paper/2101.06896/) (2021)
- [Model Extraction Warning in MLaaS Paradigm](https://mlsanity.com/paper/1711.07221/) (2017)
- [Mind Your Weight(s): A Large-scale Study on Insufficient Machine Learning Model Protection in Mobile Apps](https://mlsanity.com/paper/2002.07687/) (2020)

## Built on by (most cited first, 2 of 2)

- [Test-time GNN Model Evaluation on Dynamic Graphs](https://mlsanity.com/paper/2509.23816/) (2025)
- [SoK: Attack and Defense Landscape of Mobile On-device AI Systems](https://mlsanity.com/paper/2607.00362/) (2026)

## Abstract (arXiv)

On-device deep learning (DL) has rapidly gained adoption in mobile apps, offering the benefits of offline model inference and user privacy preservation over cloud-based approaches. However, it inevitably stores models on user devices, introducing new vulnerabilities, particularly model-stealing attacks and intellectual property infringement. While system-level protections like Trusted Execution Environments (TEEs) provide a robust solution, practical challenges remain in achieving scalable on-device DL model protection, including complexities in supporting third-party models and limited adoption in current mobile solutions. Advancements in TEE-enabled hardware, such as NVIDIA's GPU-based TEEs, may address these obstacles in the future. Currently, watermarking serves as a common defense against model theft but also faces challenges here as many mobile app developers lack corresponding machine learning expertise and the inherent read-only and inference-only nature of on-device DL models prevents third parties like app stores from implementing existing watermarking techniques in post-deployment models. To protect the intellectual property of on-device DL models, in this paper, we propose THEMIS, an automatic tool that lifts the read-only restriction of on-device DL models by reconstructing their writable counterparts and leverages the untrainable nature of on-device DL models to solve watermark parameters and protect the model owner's intellectual property. Extensive experimental results across various datasets and model structures show the superiority of THEMIS in terms of different metrics. Further, an empirical investigation of 403 real-world DL mobile apps from Google Play is performed with a success rate of 81.14%, showing the practicality of THEMIS.