Vision-language models answer from whichever modality is encoded more saliently in their final-layer representations, and specific attention heads can be manipulated to shift that preference.
On the robustness of large multimodal models against image adversarial attacks
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.CL 1years
2025 1verdicts
CONDITIONAL 1representative citing papers
citing papers explorer
-
How Do Vision-Language Models Process Conflicting Information Across Modalities?
Vision-language models answer from whichever modality is encoded more saliently in their final-layer representations, and specific attention heads can be manipulated to shift that preference.