Search papers, labs, and topics across Lattice.
Beijing University of Posts and Telecommunications
5
0
9
SelPE achieves high-fidelity structured text synthesis under strict privacy constraints, outperforming traditional methods in low-data environments.
Visual Evidence Pre-Alignment boosts MLLM performance by ensuring models effectively utilize fine-grained visual evidence, rather than just relying on coarse captions.
Vision-language models can learn to correct their own systematic errors by explicitly modeling confusion patterns between similar categories, leading to a 50% reduction in misclassifications.
MLLMs encode conflicting knowledge signals as linearly separable features in mid-to-late layers, revealing a distinct processing stage for conflict resolution.
LLM factual errors aren't just about missing knowledge; sometimes the model *knows* the truth but deliberately says otherwise.
