Research map / Time series / Causal and generative time series
C2FAR: Coarse-to-Fine Autoregressive Networks for Precise Probabilistic Forecasting
Shane Bergsma, Tim Zeyl, Javad Rahimipour Anaraki, Lei Guo
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 |
|---|---|---|---|
| Replace Gaussian output with flat binning (C2FAR-1) (ablation) | DeepAR · Azure | ND%: 3.60 -> 2.95 | Improved |
| Replace Gaussian output with flat binning (C2FAR-1) (ablation) | DeepAR · Wiki | wQL%: 54.7 -> 21.3 | Improved |
| Replace Gaussian output with flat binning (C2FAR-1) (ablation) | DeepAR · Traffic | ND%: 16.1 -> 13.0 | Improved |
| Go from one level to two levels (C2FAR-1 to C2FAR-2) (ablation) | C2FAR · Azure | ND%: 2.95 -> 2.86 | Improved |
| Go from two levels to three levels (C2FAR-2 to C2FAR-3) (ablation) | C2FAR · Azure | ND%: 2.86 -> 2.77 | Improved |
| Go from one level to three levels (ablation) | C2FAR · Electricity | ND%: 6.14 -> 6.00 | Improved |
| Go from one level to two or three levels (ablation) | C2FAR · Traffic | ND%: 12.9 -> 13.3 | Hurt |
| Go from one level to two levels (ablation) | C2FAR · Wiki | wQL%: 21.3 -> 21.0 | Improved |
| Use C2FAR-2 instead of DeepAR-Gaussian for calibration | C2FAR · Azure | Cov80: 89.9/16.9 -> 79.0/8.5 | Improved |
| Use C2FAR-3 vs DeepAR-Gaussian on extreme percentile | C2FAR · Azure | Cov99: 98.0/37.7 -> 98.6/32.7 | Improved |
| Deeper C2FAR on synthetic discrete data (level 1 to 3) (ablation) | C2FAR · Synthetic discrete uniform | test NLL: -1.564 -> -6.664 | Improved |
| Deeper C2FAR on Gaussian mixture (level 1 to 3) (ablation) | C2FAR · Synthetic Gaussian mixture | test NLL: -0.4150 -> -0.4203 | Improved |
Closest papers by meaning
- Enhancing Foundation Models for Time Series Forecasting via Wavelet-based Tokenization (ICML 2025 · similarity 0.86)
- Slimming the Fat-Tail: Morphing-Flow for Adaptive Time Series Modeling (ICML 2025 · similarity 0.86)
- TimePFN: Effective Multivariate Time Series Forecasting with Synthetic Data (AAAI 2025 · similarity 0.85)
- Multivariate Probabilistic Time Series Forecasting with Correlated Errors (NeurIPS 2024 · similarity 0.85)
- xLSTM-Mixer: Multivariate Time Series Forecasting by Mixing via Scalar Memories (NeurIPS 2025 · similarity 0.85)
- Probabilistic Forecasting with Stochastic Interpolants and Föllmer Processes (ICML 2024 · 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.
