MS student, University of Science and Technology of China
1 paper at NeurIPS 2025
Contrastive decoding fails to mitigate object hallucinations in MLLMs—apparent improvements stem from misleading factors rather than genuine effectiveness.
