Research map / Time series / Foundation models for time series
ForecastPFN: Synthetically-Trained Zero-Shot 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 |
|---|---|---|---|
| remove noise from training-loss targets (synthetic ground truth) (ablation) | ForecastPFN · Synthetic validation | validation MSE: 33.089 -> 22.692 | Helped |
| use robust scaler instead of min-max scaling (ablation) | ForecastPFN · Synthetic validation | validation MSE loss | Helped |
| scale synthetic noise from 1/6 to full (m_noise=1) (ablation) | ForecastPFN · Synthetic train | train loss: 0.591 -> 0.032 | Helped |
| scale synthetic noise from 2/3 to full (m_noise=1) (ablation) | ForecastPFN · Synthetic train | train loss: 0.384 -> 0.032 | Helped |
| zero-shot ForecastPFN vs models trained on 50 points | ForecastPFN vs best baselines · ETTh1, ETTh2, Weather | MSE: 0.127 vs 0.176 (ETTh1) | Helped |
| zero-shot ForecastPFN vs models trained on 500 points | ForecastPFN vs trained baselines · Traffic, ECL, Illness | MSE: 1.971 vs 0.523 (Traffic) | Hurt |
| ForecastPFN vs Meta-N-BEATS zero-shot | ForecastPFN vs Meta-N-BEATS · ETTh1 | MSE: 0.177 -> 0.127 | Helped |
Closest papers by meaning
- TempoPFN: Synthetic Pre-training of Linear RNNs for Zero-shot Time Series Forecasting (NeurIPS 2026 · similarity 0.92)
- TimePFN: Effective Multivariate Time Series Forecasting with Synthetic Data (AAAI 2025 · similarity 0.91)
- Zero-shot forecasting of chaotic systems (ICLR 2025 · similarity 0.89)
- TiRex: Zero-Shot Forecasting Across Long and Short Horizons with Enhanced In-Context Learning (NeurIPS 2025 · similarity 0.88)
- Rethinking Post-Training Recipes for Multimodal Time-Series Forecasting (NeurIPS 2026 · similarity 0.86)
- FilterNet: Harnessing Frequency Filters for Time Series Forecasting (NeurIPS 2024 · similarity 0.86)
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.