Approximate Structured Diffusion for Sequence Labelling

In three linesNew approach combining diffusion and CRF for sequence labelling in NLP. Method conditions a CRF on the full label sequence (noisy), bypassing span limitations of standard CRFs. Results: 16.5% error reduction on POS-tagging.

Summary generated by Claude — human-verified