Monitoring attention entropy and image-output correlation during decoding, then applying targeted contrastive corrections, reduces hallucination in multimodal LLMs without retraining.
hub
Title resolution pending
1 Pith paper cite this work, alongside 4,020 external citations. Polarity classification is still indexing.
1
Pith paper citing it
4,020
external citations · OpenAlex
hub tools
fields
cs.AI 1years
2026 1verdicts
CONDITIONAL 1representative citing papers
citing papers explorer
-
DICA: Dual-Indicator Guided Contrastive Alignment in Multimodal Large Language Models
Monitoring attention entropy and image-output correlation during decoding, then applying targeted contrastive corrections, reduces hallucination in multimodal LLMs without retraining.