Severity-Aware Curriculum Learning with Multi-Model Response Selection for Medical Text Generation

In three linesMulti-model framework with severity-aware curriculum learning for medical text generation. Three-stage progressive training (mild → moderate → critical cases) across 5 LLMs, relevance-based response selection at inference. MAQA dataset evaluation: 86.71% baseline, 90.30% after fine-tuning (BERTScore).

Summary generated by Claude — human-verified