A Dual-Purpose Framework for Backdoor Defense and Backdoor Amplification in Diffusion Models
Vu Tuan Truong, Long Bao Le · 2025
Read A Dual-Purpose Framework for Backdoor Defense and Backdoor Amplification in Diffusion Models by Vu Tuan Truong, Long Bao Le (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-2502-19047.txt.
Description
Truong and Le built this two-way tool in twenty twenty-five for guarding image generators. Diffusion models can be backdoored: a hidden trigger makes them output a set harmful image. Owners need to spot such traps, yet the search method itself is powerful. Their framework, PureDiffusion, works on both sides of that line.
Two new loss functions guide a search that reconstructs the hidden trigger. One uses shifts the trigger causes across the model's time steps. The other uses a consistency quirk that fires when the backdoor wakes. Recovered triggers then flag attacks with near-perfect accuracy in tests. Turned around, the same search can sharpen a planted trigger. The authors report far faster, stronger attacks that way.
Catalog record
| CREATOR | Vu Tuan Truong, Long Bao Le |
|---|---|
| YEAR | 2025 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 70 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2502-19047