Research map / Time series / Long-term forecasting
Adaptive Normalization for Non-stationary Time Series Forecasting: A Temporal Slice Perspective
Zhiding Liu, Mingyue Cheng, Zhi Li, Zhenya Huang, Qi Liu, Yanhu Xie, Enhong Chen
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 |
|---|---|---|---|
| add SAN slice-level normalization | FEDformer · Exchange avg MSE | MSE: 0.474 -> 0.298 (vs RevIN) | Improved |
| use SAN instead of RevIN/NST | FEDformer · Weather avg MSE | MSE: 0.279 vs 0.268 (RevIN) | Hurt |
| add SAN to backbone | DLinear · Exchange/ILI/ETTh2 multivariate | avg MSE: 7.67%, 11.13%, 21.29% MSE reduction | Improved |
| add SAN to backbone | SCINet · ETTh2 horizon 720 | MSE: 70.37% reduction | Improved |
| add SAN for long input lengths | Informer · Electricity, Lin 720, Lout 720 | MSE: 0.9426 -> 0.2090 | Improved |
| add residual learning of mean in statistics prediction (vs removing) (ablation) | SAN on Autoformer · Exchange 720 | MSE: 1.456 -> 0.689 | Improved |
| add residual learning of mean (ablation) | SAN on Autoformer · ETTh2 720 | MSE: 0.889 -> 0.471 | Improved |
| add individual per-variable preference weights (ablation) | SAN on Autoformer · Exchange 336 | MSE: 0.340 -> 0.262 | Improved |
| add individual per-variable preference weights (ablation) | SAN on Autoformer · ETTh2 (96-720) | MSE: 0.483 -> 0.471 at 720 | No clear effect |
| vary slicing length from 6 to 48 (ablation) | SAN on SCINet · Electricity/Traffic/Weather/ETTh1/ETTh2 | MSE: Electricity 0.210 / 0.207 / 0.206 / 0.208 | No clear effect |
| slicing length 12 vs 24 (ablation) | SAN on SCINet · ETTm1 720 | MSE: 0.611 (24) vs 0.450 (12) | Mixed |
| replace joint training (Dish-TS) with SAN two-stage training | FEDformer · Exchange avg MSE | MSE: 0.704 -> 0.298 | Improved |
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.
- Informer — Informer: Beyond Efficient Transformer for Long Sequence Time-Series Forecasting (AAAI 2021) · better on 1 of 1 shared result
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
- SIN: Selective and Interpretable Normalization for Long-Term Time Series Forecasting (ICML 2024 · similarity 0.91)
- DDN: Dual-domain Dynamic Normalization for Non-stationary Time Series Forecasting (NeurIPS 2024 · similarity 0.89)
- PAMod: Modeling Cyclical Shifts via Phase-Amplitude Modulation for Non-stationary Time Series Forecasting (NeurIPS 2026 · similarity 0.89)
- Unified Training of Universal Time Series Forecasting Transformers (ICML 2024 · similarity 0.87)
- TimePFN: Effective Multivariate Time Series Forecasting with Synthetic Data (AAAI 2025 · similarity 0.87)
- FilterNet: Harnessing Frequency Filters for Time Series Forecasting (NeurIPS 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.
