Research map / Time series / Long-term forecasting
Are Global Dependencies Necessary? Scalable Time Series Forecasting via Local Cross-Variate Modeling
Kun Liu, Renjun Jia, Ruifeng Yang, Xirui Zeng, Yuqi Liang, Cen 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 local cross-variate mixing (variate kernel 1 -> 3) (ablation) | VPNet · Electricity | avg MSE: 0.184 -> 0.171 | Improved |
| add local cross-variate mixing (variate kernel 1 -> 3) (ablation) | VPNet · Traffic | avg MSE: 0.443 -> 0.431 | Improved |
| add local cross-variate mixing (variate kernel 1 -> 3) (ablation) | VPNet · Solar-Energy | avg MSE: 0.224 -> 0.203 | Improved |
| add local cross-variate mixing (variate kernel 1 -> 3) (ablation) | VPNet · ETTh1 | avg MSE: 0.437 -> 0.435 | No clear effect |
| widen variate kernel from 3 to 27 (ablation) | VPNet · Electricity | avg MSE: 0.171 -> 0.160 | Improved |
| widen variate kernel from 3 to 17 (ablation) | VPNet · Weather | avg MSE: 0.250 -> 0.239 | Improved |
| widen variate kernel from 17 to 27 (ablation) | VPNet · Weather | avg MSE: 0.239 -> 0.243 | Hurt |
| random vs original variate ordering (ablation) | VPNet (kv=3, 2 blocks) · Electricity | avg MSE: 0.171 -> 0.171 | No clear effect |
| spectral ordering instead of original (ablation) | VPNet (kv=3, 2 blocks) · Electricity | avg MSE: 0.171 -> 0.174 | No clear effect |
| clustered (topology-aware) ordering instead of shuffled (ablation) | VPNet · Hetero-Mix (Traffic+Electricity+Weather) | avg MSE: 0.388 -> 0.385 | Improved |
| original (ordered) variate ordering instead of shuffled (ablation) | VPNet (kernel 27) · Electricity + 10x noise | avg MSE: 0.178 -> 0.166 | Improved |
| replace depthwise 2D convolution with local-window attention (LANet) (ablation) | VPNet · Weather | avg MSE: 0.238 -> 0.250 | Hurt |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 2 of 2 on most shared results.
- TimeKAN — TimeKAN: KAN-based Frequency Decomposition Learning Architecture for Long-term Time Series Forecasting (ICLR 2025) · better on 7 of 8 shared results
- iTransformer — iTransformer: Inverted Transformers Are Effective for Time Series Forecasting (ICLR 2024) · better on 8 of 8 shared results
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
- Multivariate Time Series Forecasting needs Cross Variable Loss (NeurIPS 2026 · similarity 0.89)
- Unlocking the Power of Patch: Patch-Based MLP for Long-Term Time Series Forecasting (AAAI 2025 · similarity 0.88)
- Revitalizing Multivariate Time Series Forecasting: Learnable Decomposition with Inter-Series Dependencies and Intra-Series Variations Modeling (ICML 2024 · similarity 0.87)
- Addressing Spatial-Temporal Heterogeneity: General Mixed Time Series Analysis via Latent Continuity Recovery and Alignment (NeurIPS 2024 · similarity 0.87)
- TStruct: Learning Shared Temporal Structures for Long-Term Time-Series Forecasting (NeurIPS 2026 · similarity 0.87)
- xLSTM-Mixer: Multivariate Time Series Forecasting by Mixing via Scalar Memories (NeurIPS 2025 · 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.
