Research map / Label-efficient and robust learning / Zero-/few-shot and meta-learning
Multi-Label Few-Shot ICD Coding as Autoregressive Generation with Prompt
Zhichao Yang, Sunjae Kwon, Zonghai Yao, Hong Yu
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (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 |
|---|---|---|---|
| Use autoregressive generation with a prompt and SOAP pretraining rather than standard fine-tuning | ICD coding · MIMIC-III-few | macro F1: 4.3 -> 30.2 | Improved |
| Use autoregressive generation with a prompt and SOAP pretraining rather than standard fine-tuning | ICD coding · MIMIC-III-full | macro F1: 10.4 -> 13.4 | Improved |
| Add SOAP assessment-and-plan generation pretraining instead of PubMed pretraining (ablation) | GP · MIMIC-III-few | macro F1: 7.3 -> 30.2 | Improved |
| Add SOAP assessment-and-plan generation pretraining instead of PubMed pretraining (ablation) | GP · MIMIC-III-few, unmentioned diagnoses | F1: 7.0 -> 17.1 | Improved |
| Add SOAP assessment-and-plan generation pretraining instead of PubMed pretraining (ablation) | GP · MIMIC-III-few, mentioned diagnoses | F1: 17.3 -> 21.3 | Improved |
| Use SOAP assessment-and-plan generation pretraining rather than Wikipedia and Book pretraining (ablation) | GP · MIMIC-III-few | macro F1: 2.9 -> 30.2 | Improved |
| Rerank MSMN and GPsoap candidate predictions using a prompted cross-attention model | MSMN + GPsoap · MIMIC-III-full | macro F1: 10.4 -> 14.6 | Improved |
| Rerank MSMN and GPsoap candidate predictions instead of concatenating them (ablation) | MSMN + GPsoap · MIMIC-III-full | macro F1: 14.0 -> 14.6 | Improved |
| Rerank MSMN and AGMHT candidate predictions using a prompted cross-attention model | MSMN + AGMHT · MIMIC-III-full | micro F1: 58.1 -> 59.7 | Improved |
| Replace SOAP pretraining with PubMed pretraining (ablation) | GP · MIMIC-III-full | micro F1: 44.2 -> 49.8 | Hurt |
| Use SOAP-pretrained generation rather than AGMHT | ICD coding · MIMIC-III-few | micro precision: 49.4 -> 38.5 | Hurt |
Closest papers by meaning
- PromptMRG: Diagnosis-Driven Prompts for Medical Report Generation (AAAI 2024 · similarity 0.83)
- SMMILE: An expert-driven benchmark for multimodal medical in-context learning (NeurIPS 2025 · similarity 0.82)
- CUPCase: Clinically Uncommon Patient Cases and Diagnoses Dataset (AAAI 2025 · similarity 0.81)
- CMR-RD: Long-Tailed Adaptive VLM for Explainable CMR Diagnosis (CVPR 2026 · similarity 0.81)
- Novel Class Discovery in Chest X-rays via Paired Images and Text (AAAI 2024 · similarity 0.81)
- Multimodal Medical Code Tokenizer (ICML 2025 · similarity 0.81)
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.
