Research map / Label-efficient and robust learning / Recognition and long-tail
Temporal Label Smoothing for Early Event Prediction
Hugo Yèche, Alizée Pace, Gunnar Ratsch, Rita Kuznetsova
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 |
|---|---|---|---|
| Apply temporal label smoothing instead of multi-horizon prediction | Early event prediction · Circulatory Failure (HiRID) | Event Recall: 85.2 $\pm$ 1.7 -> 92.5 $\pm$ 0.5 | Improved |
| Apply temporal label smoothing instead of multi-horizon prediction | Early event prediction · Decompensation (MIMIC-III) | AUPRC: 34.9 $\pm$ 0.3 -> 35.5 $\pm$ 0.3 | Improved |
| Apply temporal label smoothing instead of multi-horizon prediction | Early event prediction · Circulatory Failure (HiRID) | Event Recall: 303 to 152 out of 2045 events | Improved |
| Use exponential temporal smoothing rather than q_step (ablation) | TLS · Circulatory Failure | AUPRC: 39.3 $\pm$ 0.2 -> $\mathbf{40.6}$ $\pm$ 0.3 | Improved |
| Use exponential temporal smoothing rather than q_shift (ablation) | TLS · Decompensation | AUPRC: 34.5 $\pm$ 0.4 -> $\mathbf{35.5}$ $\pm$ 0.3 | Improved |
| Use exponential temporal smoothing rather than q_linear (ablation) | TLS · Circulatory Failure | AUPRC: 39.4 $\pm$ 0.3 -> $\mathbf{40.6}$ $\pm$ 0.3 | Improved |
| Use exponential temporal smoothing rather than q_sigmoid (ablation) | TLS · Decompensation | AUPRC: 34.9 $\pm$ 0.4 -> $\mathbf{35.5}$ $\pm$ 0.3 | Improved |
| Use exponential temporal smoothing rather than q_concave (ablation) | TLS · Circulatory Failure | AUPRC: 39.4 $\pm$ 0.3 -> $\mathbf{40.6}$ $\pm$ 0.3 | Improved |
| Use multiple horizon-specific outputs rather than shared q_step targets (ablation) | Multi-horizon prediction · Circulatory Failure | Event Recall: 83.4 $\pm$ 1.2 -> 85.2 $\pm$ 1.7 | Improved |
| Use multiple horizon-specific outputs rather than shared q_step targets (ablation) | Multi-horizon prediction · Decompensation | Event Recall: 70.4 $\pm$ 0.7 -> 70.3 $\pm$ 0.6 | No clear effect |
| Apply class reweighting rather than ordinary cross-entropy | Early event prediction · Circulatory Failure and Decompensation | AUPRC | Hurt |
| Replace fixed labels with prediction-dependent temporal targets as in TCSR | Early event prediction · Circulatory Failure and Decompensation | AUPRC | Hurt |
Closest papers by meaning
- XTSFormer: Cross-Temporal-Scale Transformer for Irregular-Time Event Prediction in Clinical Applications (AAAI 2025 · similarity 0.85)
- History-Aware Conformal Prediction Sets for Censored Time-to-Event Outcomes (NeurIPS 2026 · similarity 0.85)
- MOTOR: A Time-to-Event Foundation Model For Structured Medical Records (ICLR 2024 · similarity 0.84)
- Risk Horizons: Structured Hypothesis Spaces for Longitudinal Clinical Prediction (NeurIPS 2026 · similarity 0.84)
- Time-to-Event Pretraining for 3D Medical Imaging (ICLR 2025 · similarity 0.84)
- One Loss to Rule Them All: Marked Time-to-Event for Structured EHR Foundation Models (NeurIPS 2026 · 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.
