Research map / Time series / Causal and generative time series
Predict, Refine, Synthesize: Self-Guiding Diffusion Models for Probabilistic Time Series Forecasting
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 |
|---|---|---|---|
| use quantile (asymmetric Laplace) guidance instead of Gaussian MSE guidance (ablation) | TSDiff self-guidance · Solar | CRPS: 0.391 -> 0.358 | Helped |
| use quantile guidance instead of MSE guidance (ablation) | TSDiff self-guidance · Electricity | CRPS: 0.062 -> 0.049 | Helped |
| use quantile guidance instead of MSE guidance (ablation) | TSDiff self-guidance · M4 | CRPS: 0.045 -> 0.036 | Helped |
| unconditional TSDiff-Q with self-guidance instead of task-trained conditional TSDiff-Cond | TSDiff · KDDCup | CRPS: 0.754 -> 0.311 | Helped |
| unconditional TSDiff-Q instead of task-trained TSDiff-Cond | TSDiff · Traffic | CRPS: 0.094 -> 0.098 | Hurt |
| refine Seasonal Naive forecasts with LMC-Q | Seasonal Naive · Solar | CRPS: 0.512 -> 0.480 | Helped |
| refine Transformer forecasts with ML-Q | Transformer · KDDCup | CRPS: 0.411 -> 0.384 | Helped |
| refine DeepAR forecasts with LMC-MS (Gaussian regularizer) | DeepAR · Traffic | CRPS: 0.099 -> 0.111 | Hurt |
| use ML-Q refinement instead of LMC-Q for probabilistic base models (ablation) | DeepAR · Solar | CRPS: 0.388 -> 0.386 | Helped |
| inference-only handling of random missing values vs conditional model trained for it | TSDiff-Q · KDDCup RM missing | CRPS: 0.757 -> 0.397 | Helped |
| use TSDiff synthetic samples instead of TimeVAE/TimeGAN to train linear forecaster | Linear (LPS) · Solar | CRPS: 0.933 -> 0.581 | Helped |
| self-guidance instead of conditional model at inference | TSDiff · Exchange | inference time (s): 162.97 -> 201.67 | Hurt |
Closest papers by meaning
- Conditional Diffusion Model with Nonlinear Data Transformation for Time Series Forecasting (ICML 2025 · similarity 0.91)
- Auto-Regressive Moving Diffusion Models for Time Series Forecasting (AAAI 2025 · similarity 0.91)
- Retrieval-Augmented Diffusion Models for Time Series Forecasting (NeurIPS 2024 · similarity 0.89)
- Diffusion-TS: Interpretable Diffusion for General Time Series Generation (ICLR 2024 · similarity 0.89)
- MG-TSD: Multi-Granularity Time Series Diffusion Models with Guided Learning Process (ICLR 2024 · similarity 0.89)
- Non-stationary Diffusion For Probabilistic Time Series Forecasting (ICML 2025 · similarity 0.89)
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.