DRET Enables Parameter-Efficient Biomedical Domain Adaptation

Distilled Rapid Embedding Transfer (DRET) injects domain knowledge from large models like BioBERT into lightweight models like DistilBERT without retraining on original corpora. The method uses a priority-based mechanism to hierarchically select authoritative embeddings for tasks like PICO classification.