Research map / Time series / Long-term forecasting
Efficiently Enhancing Long-term Series Forecasting via Adaptive Lookback with Wavelets
Suxin Tong, Jingling Yuan
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 ALW plug-in (L=512 to H=256) to backbones | TQNet, CycleNet, iTransformer, DLinear · 8 datasets (ETT, Weather, Exchange, ECL, Traffic) | MSE: average MSE reduction 3.2% | Improved |
| add ALW to DLinear | DLinear · 8 datasets | MSE: 5.1% average reduction | Improved |
| add ALW to DLinear | DLinear · ETTh2 (F=720) | MSE: 0.605 -> 0.426 | Improved |
| add ALW to ECL/Exchange backbones | CycleNet · ECL (F=96) | MSE: 0.126 -> 0.126 | No clear effect |
| add Multi-scale Decomposition Module (ablation) | CycleNet+ALW · Weather | MSE: 0.240 -> 0.220 | Improved |
| add Adaptive Lookback Window Learning Module (ablation) | CycleNet+ALW · ETTh1 | MSE: 0.419 -> 0.412 | Improved |
| add Weighted Reconstruction Module (ablation) | CycleNet+ALW · ETTh1 | MSE: 0.448 -> 0.412 | Improved |
| add positional encoding in reconstruction (ablation) | CycleNet+ALW · ETTh1 | MSE: 0.461 -> 0.412 | Improved |
| use learnable weights instead of uniform averaging (ablation) | CycleNet+ALW · ETTh1 | MSE: 0.472 -> 0.412 | Improved |
| use db6 wavelet and J=3 instead of other wavelets/levels (ablation) | CycleNet+ALW · ETTm1 | MSE: 0.347 vs 0.349-0.352 | No clear effect |
| extend fixed lookback from 96 to 512 without ALW (ablation) | iTransformer · ETTh1 (F=720) | MSE: 0.503 -> 0.553 | Hurt |
| extend fixed lookback from 96 to 512 without ALW (ablation) | CycleNet · ETTh1 (F=720) | MSE: 0.520 -> 0.473 | Improved |
Closest papers by meaning
- Enhancing Foundation Models for Time Series Forecasting via Wavelet-based Tokenization (ICML 2025 · similarity 0.88)
- WaveletMixer: A Multi-Resolution Wavelets Based MLP-Mixer for Multivariate Long-Term Time Series Forecasting (AAAI 2025 · similarity 0.88)
- TimeBase: The Power of Minimalism in Efficient Long-term Time Series Forecasting (ICML 2025 · similarity 0.88)
- Efficiently Enhancing Long-term Series Forecasting via Ultra-long Lookback Windows (AAAI 2025 · similarity 0.87)
- TimeEmb: A Lightweight Static-Dynamic Disentanglement Framework for Time Series Forecasting (NeurIPS 2025 · similarity 0.87)
- Introducing Spectral Attention for Long-Range Dependency in 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.
