Research map / Label-efficient and robust learning / Domain adaptation and generalization
Transfer Learning for Sequence Labeling Using Source Model and Target Data
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| Transfer source model parameters and fine-tune the target model instead of random initialization | BLSTM and BLSTM+CRF progressive NER · CoNLL 2003 and I-CAB 2006 | F1: 0.60 to 3.50 points; 7.32 points | Helped |
| Fine-tune all transferred model weights rather than freezing embeddings and BLSTM (ablation) | Transferred sequence-labeling model · CoNLL 2003 and I-CAB 2006; new entity categories | F1 | Helped |
| Freeze source embeddings and BLSTM while updating only the output layer (ablation) | Transferred sequence-labeling model · CoNLL 2003 and I-CAB 2006; new entity categories | F1 | Hurt |
| Also freeze transferred output-layer weights (ablation) | Transferred sequence-labeling model · CoNLL 2003 and I-CAB 2006; new entity categories | F1 | Hurt |
| Add a BLSTM neural adapter connected to frozen source-model activations (ablation) | All-weights-fine-tuned transferred model · CoNLL 2003 and I-CAB 2006; original entity categories | F1 | Helped |
| Add a BLSTM neural adapter connected to frozen source-model activations (ablation) | All-weights-fine-tuned transferred model · I-CAB 2006; new entity category | F1: 6.54 and 3.71 points | Helped |
| Add the neural adapter (ablation) | BLSTM+CRF with all weights fine-tuned · CoNLL 2003 test, LOC target | F1: 90.39 -> 90.99 | Helped |
| Add the neural adapter (ablation) | BLSTM+CRF with all weights fine-tuned · CoNLL 2003 test, PER target | F1: 90.23 -> 90.36 | Helped |
| Add the neural adapter (ablation) | BLSTM+CRF with all weights fine-tuned · CoNLL 2003 test, ORG target | F1: 89.28 -> 90.16 | Helped |
| Add the neural adapter (ablation) | BLSTM+CRF with all weights fine-tuned · CoNLL 2003 test, MISC target | F1: 90.30 -> 90.34 | Helped |
| Add the neural adapter (ablation) | BLSTM+CRF with all weights fine-tuned · I-CAB 2006 test, LOC target | F1: 75.49 -> 76.87 | Helped |
| Add the neural adapter (ablation) | BLSTM+CRF with all weights fine-tuned · I-CAB 2006 test, PER target | F1: 70.74 -> 72.82 | Helped |
Closest papers by meaning
- Empower Sequence Labeling with Task-Aware Neural Language Model (AAAI 2018 · similarity 0.89)
- Transfer Learning for Sequence Tagging with Hierarchical Recurrent Networks (ICLR 2017 · similarity 0.89)
- Transfer Learning for Sequences via Learning to Collocate (ICLR 2019 · similarity 0.88)
- Low Resource Sequence Tagging with Weak Labels (AAAI 2020 · similarity 0.86)
- Three Heads Are Better than One: Improving Cross-Domain NER with Progressive Decomposed Network (AAAI 2024 · similarity 0.85)
- GuideNER: Annotation Guidelines Are Better than Examples for In-Context Named Entity Recognition (AAAI 2025 · similarity 0.83)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.