Search papers, labs, and topics across Lattice.
LMU Munich & Munich Center for Machine Learning (MCML)
4
0
7
70
BitResEdit achieves unprecedented text alignment in image editing while preserving background details, outperforming prior methods by a significant margin.
LLM-labeled data can match human-labeled data in aggregate performance for hostility detection, but be warned: the errors are systematically different, especially in nuanced cases.
Despite advancements in OCR, current models fail to generalize beyond a small set of scripts, often hallucinating characters or producing noise when faced with unfamiliar writing systems.
Cross-lingual alignment can actually *hurt* transfer learning performance because aligning embeddings doesn't necessarily help with the downstream task.
