Research map / Time series / Foundation models for time series
N-BEATS: Neural basis expansion analysis for interpretable 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 generic N-BEATS rather than the M4-winning DL/TS hybrid | M4 forecasting · M4 | OWA: 0.821 -> 0.797 | Helped |
| Ensemble the generic and interpretable configurations | N-BEATS · M4 | OWA: 0.798 -> 0.795 | Helped |
| Use interpretable trend and seasonality bases rather than the generic configuration | N-BEATS · M4 | sMAPE: 11.168 -> 11.174 | No clear effect |
| Increase the generic model from 1 to 3 stacks (ablation) | N-BEATS-G · M4 validation set | sMAPE: 11.154 -> 11.061 | Helped |
| Increase the generic model from 3 to 9 stacks (ablation) | N-BEATS-G · M4 validation set | sMAPE: 11.061 -> 10.998 | Helped |
| Increase the generic model from 9 to 18 stacks (ablation) | N-BEATS-G · M4 validation set | sMAPE: 10.998 -> 10.950 | Helped |
| Increase the generic model from 18 to 30 stacks (ablation) | N-BEATS-G · M4 validation set | sMAPE: 10.950 -> 10.937 | Helped |
| Use both trend and seasonality stacks instead of only seasonality (ablation) | Interpretable N-BEATS · M4 validation set | sMAPE: 11.189 -> 11.040 | Helped |
| Use both trend and seasonality stacks instead of only trend (ablation) | Interpretable N-BEATS · M4 validation set | sMAPE: 11.572 -> 11.040 | Helped |
| Increase both trend and seasonality stacks from 1 block each to 3 blocks each (ablation) | Interpretable N-BEATS · M4 validation set | sMAPE: 11.040 -> 10.986 | Helped |
| Retain backcast residual links instead of using parallel block inputs (ablation) | N-BEATS-G · M4 test set, average | sMAPE: 11.538 -> 11.251 | Helped |
| Retain aggregated partial forecasts instead of using only the last block forecast (ablation) | N-BEATS-G · M4 test set, average | sMAPE: 11.387 -> 11.251 | Helped |
Closest papers by meaning
- Time-MoE: Billion-Scale Time Series Foundation Models with Mixture of Experts (ICLR 2025 · similarity 0.87)
- TimePerceiver: An Encoder-Decoder Framework for Generalized Time-Series Forecasting (NeurIPS 2025 · similarity 0.87)
- ProbTS: Benchmarking Point and Distributional Forecasting across Diverse Prediction Horizons (NeurIPS 2024 · similarity 0.87)
- Enhancing Foundation Models for Time Series Forecasting via Wavelet-based Tokenization (ICML 2025 · similarity 0.87)
- Feature-aligned N-BEATS with Sinkhorn divergence (ICLR 2024 · similarity 0.87)
- Unified Training of Universal Time Series Forecasting Transformers (ICML 2024 · similarity 0.87)
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.