UTOPIA: Unlearnable Tabular Data via Decoupled Shortcut Embedding
Jiaming He, Fuming Luo, Hongwei Li, Wenbo Jiang, Wenshu Fan, Zhenbo Shi, Xudong Jiang, Yi Yu · 2026
Read UTOPIA: Unlearnable Tabular Data via Decoupled Shortcut Embedding by Jiaming He, Fuming Luo, Hongwei Li, Wenbo Jiang, Wenshu Fan, Zhenbo Shi, Xudong Jiang, Yi Yu (2026) 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-2602-07358.txt.
Description
Unlearnable cases protect images from unauthorized training by making them unlearnable. Tabular data in finance and healthcare resists the same tricks because features mix types and learning concentrates on few dimensions. Jiaming He and seven co-authors extend the protection to tables with UTOPIA.
Under a spectral dominance setting, they show certified unlearnability is possible when a poison spectrum overwhelms the clean meaning spectrum. UTOPIA then decouples tuning into two channels: high-salience features for meaning obfuscation and redundant low-salience features for a strongly correlated shortcut, preserving tabular constraints. Data owners get a principled way to share tables that models cannot profitably learn from.
Catalog record
| CREATOR | Jiaming He, Fuming Luo, Hongwei Li, Wenbo Jiang, Wenshu Fan, Zhenbo Shi, Xudong Jiang, Yi Yu |
|---|---|
| YEAR | 2026 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 53 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2602-07358