Differential Vector Erasure: Unified Training-Free Concept Erasure for Flow Matching Models
Zhiqi Zhang, Xinhao Zhong, Yi Sun, Shuoyang Sun, Bin Chen, Shu-Tao Xia, Xuan Wang · 2026
Read Differential Vector Erasure: Unified Training-Free Concept Erasure for Flow Matching Models by Zhiqi Zhang, Xinhao Zhong, Yi Sun, Shuoyang Sun, Bin Chen, Shu-Tao Xia, Xuan Wang (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-2602-01089.txt.
Description
Zhiqi Zhang and co-authors scrub chosen ideas from flow picture makers, no retraining. Past scrub plans aim at diffusion nets and lean on costly tuning, while flow nets run on new math. Their hunch is that ideas live in the aim of the speed field steering the flow. A gap vector between a target idea and a safe anchor thus marks what to cut. At run time, the plan projects the speed field onto that gap, wiping the idea while sparing the rest.
Catalog record
| CREATOR | Zhiqi Zhang, Xinhao Zhong, Yi Sun, Shuoyang Sun, Bin Chen, Shu-Tao Xia, Xuan Wang |
|---|---|
| YEAR | 2026 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 51840 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2602.01089