Found 1 document(s)
Page 1 of 1
-
Multimodal Large Language Models (MLLMs) have achieved impressive performance in generating text conditioned on inputs from different modalities. In this work, we introduce backdoor infection, a cross-modal attack framew...