CMA uses diffusion-generated counterfactuals and two-player Shapley values to measure whether an image or the text drives a multimodal LLM's prediction, hitting 98% on synthetic biased benchmarks.
Title resolution pending
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.CV 1years
2026 1verdicts
CONDITIONAL 1representative citing papers
citing papers explorer
-
Which Modality Decides? Counterfactual Modality Attribution for Multimodal LLMs
CMA uses diffusion-generated counterfactuals and two-player Shapley values to measure whether an image or the text drives a multimodal LLM's prediction, hitting 98% on synthetic biased benchmarks.