FSE: Fast-Slow Experts Model for Continual NER Learning
A new model called Fast-Slow Experts (FSE) has been proposed for Continual Learning in Named Entity Recognition (CLNER). The model addresses catastrophic forgetting and insufficient sharing of information across tasks. FSE uses a shared fast expert to learn token-level links and filter unlikely spans, while task-specific slow experts classify remaining candidates. It also introduces a length-decay negative sampling strategy to handle span imbalance. Experiments on OntoNotes and FewNERD datasets show state-of-the-art performance.
Key facts
- FSE is a Fast-Slow Experts enhanced span-based NER model for CLNER.
- The shared fast expert learns token-level links to filter out unlikely spans.
- Task-specific slow experts perform span classification on remaining candidates.
- A length-decay negative sampling strategy mitigates span imbalance.
- Experiments conducted on OntoNotes and FewNERD synthetic datasets.
- FSE achieves state-of-the-art performance in CLNER scenarios.
- The model promotes knowledge sharing across tasks.
- It reduces learning burden at each task to maintain plasticity.
Entities
—