Effective Backdoor Mitigation in Vision-Language Models Depends on the Pre-training Objective

Open Webpage

Sahil Verma, Gantavya Bhatt, Avi Schwarzschild, Soumye Singhal, Arnav Mohanty Das, Chirag Shah, John P. Dickerson, Pin-Yu Chen, Jeff Bilmes

Published: 01 Jan 2025, Last Modified: 01 Oct 2025Trans. Mach. Learn. Res. 2025EveryoneRevisionsBibTeXCC BY-SA 4.0