Mitigating Manifold Departure: Uncertainty-Aware Subspace Rectification for Trustworthy MLLM Decoding

MLLMs frequently hallucinate objects inconsistent with visual inputs due to over-reliance on language priors, leading to a phenomenon termed "Manifold Departure". Manifold-Guided Adaptive Projection (MGAP) is proposed as a training-free decoding method to mitigate hallucinations while preserving representation structure.