AI Paper Atlas

Research map / Label-efficient and robust learning / Recognition and long-tail

Towards Reliable and Efficient Backdoor Trigger Inversion via Decoupling Benign Features

Xiong Xu, Kunzhe Huang, Yiming Li, Zhan Qin, Kui Ren

ICLR 2024 · Label-efficient and robust learning › Recognition and long-tail · approach: Dense features and foundation priors

Read the paper

AI-extracted summary. Written by AI from the paper's abstract only. Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Proposes a backdoor trigger inversion method that decouples benign features to make trigger reconstruction more reliable and efficient for backdoor defense.

Problem

Backdoor trigger inversion (BTI) for defending third-party models against implanted backdoors, which generates trigger patterns for benign samples on a backdoored model.

Approach

Decouples benign features from trigger features during trigger inversion so that generated triggers are more reliable and cheaper to compute than prior BTI approaches.

Closest papers by meaning

Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy