Research map / Time series / Causal and generative time series
From Samples to Scenarios: A New Paradigm for Probabilistic Forecasting
Xilin Dai, Zhijian Xu, Wanxu Cai, Qiang Xu
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 |
|---|---|---|---|
| increase number of scenarios N from 16 to 256 (ablation) | TimePrism · Electricity | CRPS: 0.414 -> 0.162 | Improved |
| increase number of scenarios N from 256 to 625 (ablation) | TimePrism · Electricity | CRPS: 0.162 -> 0.133 | Improved |
| increase number of scenarios N from 625 to 1024 (ablation) | TimePrism · Electricity | CRPS: 0.133 -> 0.139 | No clear effect |
| increase number of scenarios N from 625 to 1024 (ablation) | TimePrism · Solar | Distortion: 0.101 -> 0.0917 | Improved |
| increase number of scenarios N from 1 to 625 (ablation) | TimePrism · Solar | CRPS: 0.199 -> 0.0852 | Improved |
| increase number of scenarios N from 1 to 16 (ablation) | TimePrism · Electricity | Distortion: 0.733 -> 0.911 | Hurt |
| use explicit scenario probabilities with a probability head instead of resampling from fixed set (TimeMCL) | TimePrism vs TimeMCL · Electricity | CRPS: 0.370 -> 0.133 | Improved |
| use TimePrism with only 16 scenarios versus TimeMCL (16 hypotheses) | TimePrism-16 · Solar | CRPS: 0.290 -> 0.137 | Improved |
| replace sampling with single-pass scenario generation | TimePrism vs TimeGrad · Exchange inference, S=1 | inference FLOPs: 1.9 x 10^8 -> 5.1 x 10^5 | Improved |
| Winner-Takes-All reconstruction loss in place of full likelihood | TimePrism · Distortion across 5 datasets | Distortion | Improved |
| TimePrism versus TACTiS-2 on Wikipedia | TimePrism · Wikipedia | CRPS: 0.484 -> 0.506 | Hurt |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- TACTiS-2 — TACTiS-2: Better, Faster, Simpler Attentional Copulas for Multivariate Time Series (ICLR 2024) · better on 5 of 6 shared results
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- Parametric Prior Mapping Framework for Non-stationary Probabilistic Time Series Forecasting (ICML 2026 · similarity 0.88)
- ProbTS: Benchmarking Point and Distributional Forecasting across Diverse Prediction Horizons (NeurIPS 2024 · similarity 0.86)
- Cherry-Picking in Time Series Forecasting: How to Select Datasets to Make Your Model Shine (AAAI 2025 · similarity 0.86)
- Probabilistic Forecasting with Stochastic Interpolants and Föllmer Processes (ICML 2024 · similarity 0.86)
- TimePFN: Effective Multivariate Time Series Forecasting with Synthetic Data (AAAI 2025 · similarity 0.85)
- Winner-takes-all for Multivariate Probabilistic Time Series Forecasting (ICML 2025 · similarity 0.85)
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.
