Search papers, labs, and topics across Lattice.
Beijing University of Posts and Telecommunications
2
0
4
By surgically removing "hallucination patterns" from a model's hidden state, HulluEdit offers a reference-free, single-pass method to dramatically reduce object hallucinations in LVLMs without sacrificing visual grounding.
Achieve more interpretable and adaptive multimodal recommendation by explicitly controlling *what* and *how* to fuse modalities with a mixture-of-experts architecture guided by interaction data and dynamically weighted by routing entropy.
