MLLMs Hallucinate when Information Distribution Drifts in Synergy Heads
HEAL injects dynamic information calibration factors into the value vectors of synergy heads and actively regulates visual-language dependencies, steering the output distribution towards factual evidence, offering a simple and interpretable pathway to enhance model trustworthiness.
Meng'en Qin, Jun-Ye Chen, Ju-Cheng Liu et al.
· 0 citations