Research map / Time series / Long-term forecasting
Informer: Beyond Efficient Transformer for Long Sequence 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 ProbSparse attention instead of canonical self-attention (ablation) | Informer · ETTh1, prediction length 336, encoder input 720 | MSE: 0.214 -> 0.225 | No clear effect |
| Use ProbSparse attention instead of canonical self-attention (ablation) | Informer · ETTh1, prediction length 720, encoder input 1440 | memory feasibility: Informer†: - | Helped |
| Remove self-attention distilling (ablation) | Informer† · ETTh1, prediction length 336, encoder input 720 | MSE: 0.214 -> 0.204 | Hurt |
| Add self-attention distilling (ablation) | Informer† · ETTh1, encoder input longer than 720 | memory feasibility: Informer‡: - | Helped |
| Replace generative decoding with dynamic decoding (ablation) | Informer‡ · ETTh1, prediction length 480, prediction offset +0 | MSE: 0.198 -> 0.392 | Hurt |
| Increase ProbSparse sampling factor from small values toward c=5 (ablation) | Informer · ETTh1 | MSE | Helped |
| Use the L and L/4 encoder stacks together (ablation) | Informer · ETTh1 | MSE | Helped |
| Increase encoder input length (ablation) | Informer · ETTh1, long prediction sequences such as 168 | MSE | Helped |
| Increase encoder input length (ablation) | Informer · ETTh1, short prediction sequences such as 48 | MSE | Mixed |
| Use Informer rather than Informer† with canonical self-attention | Informer · Univariate forecasting across ETTh1, ETTh2, ETTm1, Weather, and ECL | MSE: 32>12 | Helped |
| Use Informer rather than DeepAR | Informer · ECL, shorter horizons (≤ 336) | MSE | Hurt |
| Use Informer rather than DeepAR | Informer · ECL, longer horizons | MSE | Helped |
Closest papers by meaning
- Sequence Complementor: Complementing Transformers for Time Series Forecasting with Learnable Sequences (AAAI 2025 · similarity 0.92)
- TimePerceiver: An Encoder-Decoder Framework for Generalized Time-Series Forecasting (NeurIPS 2025 · similarity 0.89)
- SAMformer: Unlocking the Potential of Transformers in Time Series Forecasting with Sharpness-Aware Minimization and Channel-Wise Attention (ICML 2024 · similarity 0.89)
- Are Self-Attentions Effective for Time Series Forecasting? (NeurIPS 2024 · similarity 0.89)
- Introducing Spectral Attention for Long-Range Dependency in Time Series Forecasting (NeurIPS 2024 · similarity 0.89)
- iTransformer: Inverted Transformers Are Effective for Time Series Forecasting (ICLR 2024 · 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.