Research map / Time series / Causal and generative time series
MoRGen: Mixture-of-Resolutions Generative Forecasting for Irregularly Sampled Medical Time-Series Data
Nassim Oufattole, Matthew B.A. McDermott, Collin Stultz
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 |
|---|---|---|---|
| fuse minute-level and coarse-resolution experts with learned mixture weights | minute-level baseline forecaster · MIMIC-IV death, 30-day | AUROC: 70.27 -> 89.06 | Improved |
| fuse minute-level and coarse-resolution experts | minute-level baseline forecaster · EHRSHOT death, 30-day | AUROC: 49.98 -> 81.46 | Improved |
| fuse resolutions | minute-level baseline forecaster · MIMIC-IV readmission, 30-day | AUROC: 69.98 -> 70.28 | No clear effect |
| fuse resolutions | minute-level baseline forecaster · Hospital A death, 2-year | AUROC: 58.93 -> 77.46 | Improved |
| fuse resolutions at extended horizon | minute-level baseline forecaster · MIMIC-IV death, 20-year | AUROC: 55.36 -> 77.64 | Improved |
| fuse resolutions for joint endpoint | minute-level baseline forecaster · EHRSHOT creatinine and death, 30d/90d | AUROC: 50.00 -> 82.70 | Improved |
| learn task- and horizon-specific mixture weights instead of uniform averaging (ablation) | MoRGen · 3 datasets, 7 tasks, 2 horizons | AUROC win rate vs baseline: 66.7% -> 100.0% | Improved |
| share one weight set across tasks and horizons instead of per-task weights (ablation) | MoRGen · 3 datasets, 7 tasks, 2 horizons | AUROC win rate vs baseline: 100.0% -> 100.0% | No clear effect |
| use all codebook sizes K in the expert pool instead of a single K (ablation) | MoRGen · EHRSHOT 30-day | AUROC: Death 79.42 -> 81.46 | Mixed |
| restrict pool to a single codebook size (ablation) | MoRGen · 3 datasets, 7 tasks, 2 horizons | AUROC win rate vs baseline: 167 / 168 | No clear effect |
| add ETHOS as an extra expert in the mixture (ablation) | MoRGen · 42 task-horizon settings vs ETHOS | AUROC win rate: 27 / 42 (64.3%) -> 40 / 42 (95.2%) | Improved |
| add ETHOS as an extra expert (ablation) | MoRGen · EHRSHOT hematocrit, 30-day | AUROC: 86.54 -> 86.36 | Hurt |
Closest papers by meaning
- Time Series Generation Under Data Scarcity: A Unified Generative Modeling Approach (NeurIPS 2025 · similarity 0.86)
- TARFVAE: Efficient One-Step Generative Time Series Forecasting via TARFLOW based VAE (NeurIPS 2025 · similarity 0.85)
- Breaking the Barrier of Hard Samples: A Data-Centric Approach to Synthetic Data for Medical Tasks (ICML 2025 · similarity 0.85)
- Integrating Sequence and Image Modeling in Irregular Medical Time Series Through Self-Supervised Learning (AAAI 2025 · similarity 0.85)
- TSGM: A Flexible Framework for Generative Modeling of Synthetic Time Series (NeurIPS 2024 · similarity 0.84)
- A Closer Look at Transformers for Time Series Forecasting: Understanding Why They Work and Where They Struggle (ICML 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.
