Lite-BD: A Lightweight Black-box Backdoor Defense via Reviving Multi-Stage Image Transformations
Abdullah Arafat Miah 1 , Yu Bi 1 · 2026
Read Lite-BD: A Lightweight Black-box Backdoor Defense via Reviving Multi-Stage Image Transformations by Abdullah Arafat Miah 1 , Yu Bi 1 (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-2602-07197.txt.
Description
Backdoor attacks poison training data so a model misbehaves on triggered inputs, and black-box defenses are the useful option for hosted models. Abdullah Arafat Miah and one co-author study image transformations as defenses and find down-upscaling the most effective at breaking spatial triggers.
Their Lite-BD defense runs in two stages: a super-resolution down-upscaling step that neutralizes spatial triggers, followed by query-based frequency filtering that removes triggers hidden in specific bands. Tests against strong attacks show it holding up at low cost. Security builders get a lightweight, justified defense that does not need access to model internals.
Catalog record
| CREATOR | Abdullah Arafat Miah 1 , Yu Bi 1 |
|---|---|
| YEAR | 2026 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 29 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2602-07197