Research map / Time series / Long-term forecasting
What If We Let Forecasting Forget? A Sparse Bottleneck for Cross-Variable Dependencies
Fan Zhang, Shiming Fan, Hua Wang
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 the sparse dependency bottleneck (Top-K cross-variable routing) (ablation) | MS-FLOW · Solar-Energy | MSE: 0.210 -> 0.173 | Improved |
| add patch-wise temporal interaction (ablation) | MS-FLOW · Electricity | MSE: 0.135 -> 0.132 | Improved |
| add the feed-forward network after dependency injection (ablation) | MS-FLOW · Electricity | MSE: 0.146 -> 0.132 | Improved |
| use Top-K sparse routing instead of dense (K=C) interaction (ablation) | MS-FLOW · Electricity | MSE: 0.139 -> 0.132 | Improved |
| use Top-K instead of dense interaction (ablation) | MS-FLOW · ETTm2 | MSE: 0.166 -> 0.166 | No clear effect |
| select neighbors by Top-K similarity instead of random-K (ablation) | MS-FLOW · Solar-Energy | MSE: 0.179 -> 0.173 | Improved |
| use sample-adaptive graph instead of static graph (ablation) | MS-FLOW · Solar-Energy | MSE: 0.196 -> 0.173 | Improved |
| use average pooling instead of max pooling for variable states (ablation) | MS-FLOW · Weather | MSE: 0.172 -> 0.150 | Improved |
| use shared window-level graph instead of per-patch graphs (ablation) | MS-FLOW · Weather | MSE: 0.155 -> 0.150 | Improved |
| Top-K sparse routing reduces edges touching noisy channels (ablation) | MS-FLOW · Electricity with 25 noisy channels | pollution hit rate (%): 7.79% -> 3.80% | Improved |
| use MS-FLOW instead of TimeFilter on Traffic | MS-FLOW · Traffic (L=96) | MSE: 0.407 -> 0.440 | Hurt |
| increase encoder depth (layers) (ablation) | MS-FLOW · ETTh1, Electricity, Traffic | MSE | No clear effect |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 2 on most shared results.
- TimeFilter — TimeFilter: Patch-Specific Spatial-Temporal Graph Filtration for Time Series Forecasting (ICML 2025) · better on 10 of 12 shared results
- iTransformer — iTransformer: Inverted Transformers Are Effective for Time Series Forecasting (ICLR 2024) · better on 0 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
- 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 0.89)
- CSformer: Combining Channel Independence and Mixing for Robust Multivariate Time Series Forecasting (AAAI 2025 · similarity 0.89)
- SOFTS: Efficient Multivariate Time Series Forecasting with Series-Core Fusion (NeurIPS 2024 · similarity 0.88)
- xLSTM-Mixer: Multivariate Time Series Forecasting by Mixing via Scalar Memories (NeurIPS 2025 · similarity 0.87)
- From Similarity to Superiority: Channel Clustering 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.
