Research map / Time series / Long-term forecasting
PhaseFormer: From Patches to Phases for Efficient and Effective Time Series Forecasting
Yiming Niu, Jinliang Deng, Yongxin Tong
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 cross-phase routing vs none (w/o Routing) (ablation) | PhaseFormer · Weather | MSE: 0.1907 -> 0.1503 | Improved |
| add cross-phase routing vs none (ablation) | PhaseFormer · Traffic | MSE: 0.3892 -> 0.3721 | Improved |
| use routers instead of linear mixing (ablation) | PhaseFormer · Electricity | MSE: 0.1403 -> 0.1290 | Improved |
| use routers instead of full attention (ablation) | PhaseFormer · Traffic | MSE: 0.3791 -> 0.3721 | Improved |
| use routers instead of full attention (ablation) | PhaseFormer · Traffic | MFLOPs: 131.452 -> 113.356 | Improved |
| routing vs linear mixing on compute (ablation) | PhaseFormer · Traffic | MFLOPs: 37.776 -> 113.356 | Hurt |
| increase number of routers M (ablation) | PhaseFormer · Electricity/Traffic/Weather | MSE | Mixed |
| phase tokens instead of patch tokens (PhaseFormer vs PatchTST) | PhaseFormer · Traffic | MSE: 0.394 -> 0.386 | Mixed |
| PhaseFormer vs FITS | PhaseFormer · ETTh2 | MSE: 0.334 -> 0.346 | Hurt |
| PhaseFormer vs TimeBase (similar efficiency) | PhaseFormer · Traffic | MSE: 0.418 -> 0.386 | Improved |
Later papers that compared with this method
2 later accepted papers report PhaseFormer as a baseline in their results tables; 2 of them report a better number on most of the results they share with it.
- TStruct: Learning Shared Temporal Structures for Long-Term Time-Series Forecasting (NeurIPS 2026 · better on 7 of 7 shared results)
- Rethinking Time Series Tokenization from a Frequency Perspective (NeurIPS 2026 · better on 1 of 1 shared result)
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 3 of 3 on most shared results.
- TimeBase — TimeBase: The Power of Minimalism in Efficient Long-term Time Series Forecasting (ICML 2025) · better on 3 of 4 shared results
- SparseTSF — SparseTSF: Modeling Long-term Time Series Forecasting with *1k* Parameters (ICML 2024) · better on 2 of 2 shared results
- iTransformer — iTransformer: Inverted Transformers Are Effective for Time Series Forecasting (ICLR 2024) · better on 1 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
- TimeBase: The Power of Minimalism in Efficient Long-term Time Series Forecasting (ICML 2025 · similarity 0.88)
- DeformableTST: Transformer for Time Series Forecasting without Over-reliance on Patching (NeurIPS 2024 · similarity 0.87)
- PAMod: Modeling Cyclical Shifts via Phase-Amplitude Modulation for Non-stationary Time Series Forecasting (NeurIPS 2026 · similarity 0.87)
- TimePerceiver: An Encoder-Decoder Framework for Generalized Time-Series Forecasting (NeurIPS 2025 · similarity 0.87)
- Enhancing Foundation Models for Time Series Forecasting via Wavelet-based Tokenization (ICML 2025 · similarity 0.86)
- CycleNet: Enhancing Time Series Forecasting through Modeling Periodic Patterns (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.
