Research map / Time series / Causal and generative time series
Reliable Probabilistic Forecasting of Irregular Time Series through Marginalization-Consistent Flows
Vijaya Krishna Yalavarthi, Randolf Scholz, Christian Klötergens, Kiran Madhusudhanan, Stefan Born, Lars Schmidt-Thieme
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 normalizing flows to the mixture of Gaussians (ablation) | MOSES · PhysioNet'12 | njNLL: 1.063 -> -0.491 | Improved |
| add full low-rank covariance to the base Gaussian (ablation) | MOSES · PhysioNet'12 | njNLL: -0.308 -> -0.491 | Improved |
| learn mixture weights instead of fixing them to 1/D (ablation) | MOSES · PhysioNet'12 | njNLL: -0.451 -> -0.491 | Improved |
| use a mixture of D components instead of one component (ablation) | MOSES · PhysioNet'12 | njNLL: -0.493 -> -0.491 | No clear effect |
| use a mixture of D components instead of one component (ablation) | MOSES · MIMIC-III | njNLL: 0.173 -> -0.305 | Improved |
| use a mixture of D components instead of one component (ablation) | MOSES · USHCN | njNLL: -3.255 -> -3.357 | Improved |
| use MOSES (marginalization consistent) instead of ProFITi | ProFITi · MIMIC-III | mNLL: 0.408 -> 0.163 | Improved |
| use MOSES instead of ProFITi | ProFITi · PhysioNet'12 | njNLL: -0.647 -> -0.491 | Hurt |
| use MOSES instead of ProFITi on long horizon PhysioNet'12 (obs 12h, forecast 36h) | ProFITi · PhysioNet'12 12/36 | mNLL: 2.977 -> 0.040 | Improved |
| replace the MOSES encoder with GraFITi (ablation) | MOSES · MIMIC-IV | njNLL: -1.668 -> -2.197 | Improved |
| replace the MOSES encoder with GraFITi (ablation) | MOSES · MIMIC-IV | mNLL: -0.634 -> -0.299 | Hurt |
| use MOSES instead of ProFITi for CRPS | ProFITi · MIMIC-IV | CRPS: 0.279 -> 0.245 | Improved |
Closest papers by meaning
- Probabilistic Forecasting of Irregularly Sampled Time Series with Missing Values via Conditional Normalizing Flows (AAAI 2025 · similarity 0.88)
- Multivariate Probabilistic Time Series Forecasting via Conditioned Normalizing Flows (ICLR 2021 · similarity 0.85)
- Flow Matching with Gaussian Process Priors for Probabilistic Time Series Forecasting (ICLR 2025 · similarity 0.84)
- Multivariate Probabilistic Time Series Forecasting with Correlated Errors (NeurIPS 2024 · similarity 0.84)
- SIN: Selective and Interpretable Normalization for Long-Term Time Series Forecasting (ICML 2024 · similarity 0.83)
- ProbTS: Benchmarking Point and Distributional Forecasting across Diverse Prediction Horizons (NeurIPS 2024 · similarity 0.83)
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.
