THEMIS: Towards Practical Intellectual Property Protection for Post-Deployment On-Device Deep Learning Models
Yujin Huang Zhi Zhang Qingchuan Zhao Xingliang Yuan Chunyang Chen · 2025
Read THEMIS: Towards Practical Intellectual Property Protection for Post-Deployment On-Device Deep Learning Models by Yujin Huang Zhi Zhang Qingchuan Zhao Xingliang Yuan Chunyang Chen (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2503-23748.txt.
Description
Yujin Huang, Zhi Zhang, Qingchuan Zhao, Xingliang Yuan, and Chunyang Chen present THEMIS in this 2025 paper, a guard for AI models that live on phones. On-device models run offline and keep user data private, a clear win. The cost is exposure: the model file sits on the user's device, where it can be stolen or copied.
Hardware vaults called trusted execution environments can lock models away, but they are hard to scale and awkward with third-party models. Watermarking, the usual fallback, assumes skills many app developers lack. THEMIS aims for a practical middle path that works after deployment, on real devices. The paper weighs the trade-offs among protection strength, speed, and ease of use, as new graphics chips with built-in vaults begin to change what is possible.
Catalog record
| CREATOR | Yujin Huang Zhi Zhang Qingchuan Zhao Xingliang Yuan Chunyang Chen |
|---|---|
| YEAR | 2025 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 85 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2503-23748