Hong-Han Shuai

Contradicting the assumption that MLLMs require specialized forensic fine-tuning to spot subtle image edits, simply contrasting cross-attention against a Gaussian-blurred baseline isolates genuine manipulation artifacts and cuts explanatory hallucination by up to 33.4% training-free.

Existing hallucination detection methods falter in the nuanced landscape of scientific peer reviews, revealing a critical gap in ensuring review reliability.

RADIANCE boosts the success rate of synthesizing rare concepts in text-to-image models by effectively rebalancing denoising trajectories without additional training.

N-ary hypergraphs can boost RAG performance by encoding richer relational facts, enabling more accurate and interpretable multi-hop reasoning for both open and closed-domain question answering.