Research map / Time series / Long-term forecasting
SCINet: Time Series Modeling and Forecasting with Sample Convolution and Interaction
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 SCINet's downsample-convolve-interact architecture instead of the strongest listed baseline | SCINet · Short-term forecasting, Solar-Energy, Traffic, Electricity, and Exchange-Rate, horizon 3 | RSE: Solar-Energy: 0.1803 -> 0.1775; Traffic: 0.4487 -> 0.4216; Electricity: 0.0823 -> 0.0740; Exchange-Rate: 0.0174 -> 0.017 | Helped |
| Use SCINet instead of Autoformer | SCINet · Exchange-Rate, long-term forecasting, horizon 96 | MSE: 0.197 -> 0.061 | Helped |
| Use SCINet instead of Autoformer | SCINet · ETTh1 multivariate, horizon 24 | MSE: 0.406 -> 0.300 | Helped |
| Remove interactive learning between the two downsampled subsequences (ablation) | SCINet SCI-Block · ETTh1 and PEMS08, multiple prediction horizons | prediction accuracy | Hurt |
| Share the convolution weights across the SCI-Block modules (ablation) | SCINet SCI-Block · ETTh1 and PEMS08, multiple prediction horizons | prediction accuracy | Hurt |
| Remove the residual connection (ablation) | SCINet · ETTh1 and PEMS08 | prediction accuracy | Hurt |
| Remove the fully connected decoder (ablation) | SCINet · ETTh1 and PEMS08 | prediction accuracy | Hurt |
| Increase the number of levels from 3 to 4 (ablation) | SCINet · ETTh1, horizon 24, look-back 192 | MSE: 0.330 -> 0.325 | Helped |
| Increase the number of stacks from 1 to 3 (ablation) | SCINet · ETTh1, horizon 24, look-back 24 | MSE: 0.405 -> 0.374 | Helped |
| Increase the number of stacks from 1 to 3 (ablation) | SCINet · ETTh1, horizon 24, look-back 192 | MSE: 0.330 -> 0.363 | Hurt |
| Vary addition/subtraction choices in the interaction update (ablation) | SCINet SCI-Block · PEMS03, PEMS04, PEMS07, and PEMS08 | MAE: PEMS03: 15.06–15.30; PEMS04: 19.21–19.32; PEMS07: 21.63–21.77; PEMS08: 15.72–15.84 | No clear effect |
Closest papers by meaning
- FilterNet: Harnessing Frequency Filters for Time Series Forecasting (NeurIPS 2024 · similarity 0.89)
- TVNet: A Novel Time Series Analysis Method Based on Dynamic Convolution and 3D-Variation (ICLR 2025 · similarity 0.89)
- TStruct: Learning Shared Temporal Structures for Long-Term Time-Series Forecasting (NeurIPS 2026 · similarity 0.88)
- TimePerceiver: An Encoder-Decoder Framework for Generalized Time-Series Forecasting (NeurIPS 2025 · similarity 0.88)
- HDT: Hierarchical Discrete Transformer for Multivariate Time Series Forecasting (AAAI 2025 · similarity 0.88)
- TimePFN: Effective Multivariate Time Series Forecasting with Synthetic Data (AAAI 2025 · similarity 0.88)
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.