Research map / Label-efficient and robust learning / Zero-/few-shot and meta-learning
SGERA: Stein-Guided ECG-Report Alignment for ECG Representation Learning
Jian Chen, Yipeng Du, Wenhao Yuan, Shuai Wang, Jinfeng Xu, Zewei Liu, Running Zhao, Edith Cheuk-Han Ngai
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 |
|---|---|---|---|
| add ECG-Report Matching task with hard negative mining (ablation) | SGERA · ECG zero-shot (average) | AUC: 78.0 -> 79.5 | Improved |
| add ECG-Report Matching task with hard negative mining (ablation) | SGERA · ECG linear probing 1% (average) | AUC: 77.1 -> 79.0 | Improved |
| use hard negatives instead of random negatives in ERM (ablation) | SGERA · ECG zero-shot (average) | AUC: 79.1 -> 79.5 | Improved |
| replace CLIP InfoNCE loss with Stein-RBF instance loss (ablation) | SGERA · ECG zero-shot | AUC | Improved |
| add distribution-level Stein-Score alignment loss (ablation) | SGERA · ECG zero-shot | AUC | Improved |
| add KL regularization on latent space (ablation) | SGERA · ECG linear probing (1%) | AUC | Improved |
| use Stein-guided loss with all three components (rbf, KL, dist) (ablation) | SGERA · ECG zero-shot and linear probing 1% (average) | AUC: 77.5 -> 79.5 zero-shot; 76.10 -> 79.02 linear probe | Improved |
| Stein alignment vs baselines on rhythm and sub tasks | SGERA · PTBXL-Rhythm zero-shot | AUC: 88.6 (D-BETA) vs 83.3 | Hurt |
Closest papers by meaning
- From Token to Rhythm: A Multi-Scale Approach for ECG-Language Pretraining (ICML 2025 · similarity 0.87)
- Parallel-Learning of Invariant and Tempo-variant Attributes of Single-Lead Cardiac Signals: PLITA (AAAI 2025 · similarity 0.86)
- GEM: Empowering MLLM for Grounded ECG Understanding with Time Series and Images (NeurIPS 2025 · similarity 0.85)
- Boosting Masked ECG-Text Auto-Encoders as Discriminative Learners (ICML 2025 · similarity 0.84)
- Medformer: A Multi-Granularity Patching Transformer for Medical Time-Series Classification (NeurIPS 2024 · similarity 0.84)
- Reading Your Heart: Learning ECG Words and Sentences via Pre-training ECG Language Model (ICLR 2025 · similarity 0.84)
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.
