Research map / Time series / Causal and generative time series
CombinationTS: A Modular Framework for Understanding Time-Series Forecasting Models
Xiaorui Wang, Fanda Fan, Chenxi Wang, Yuxuan Yang, Rui Tang, Kuoyu Gao, simiao pang, Yuanfeng Shang, Zhipeng Liu, Wanling Gao, Lei Wang, Jianfeng Zhan
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 |
|---|---|---|---|
| use parameter-free Identity encoder instead of Transformer encoder (ablation) | modular forecasting pipeline with RevIN and linear decoder · ETTh1 | avg marginalized MSE: 0.5095 -> 0.4392 | Improved |
| use Identity encoder instead of Transformer encoder (ablation) | modular forecasting pipeline · ETTm2 | avg marginalized MSE: 0.2909 -> 0.2621 | Improved |
| use Identity encoder instead of Transformer encoder (ablation) | modular forecasting pipeline · Electricity | avg marginalized MSE: 0.1776 -> 0.1888 | Hurt |
| use Identity encoder instead of Transformer, all datasets combined (ablation) | modular forecasting pipeline · six datasets combined | MSE (Mann-Whitney U test): p<0.0001 vs Transformer; p=0.0115 vs MLP | Improved |
| use structured tokenization (patch-wise or variate-wise) instead of point-wise embedding (ablation) | modular forecasting pipeline · ETTh1 | avg marginalized MSE: 0.4367 -> 0.4277 (patch); 0.4285 (variate) | Improved |
| use parameter-free Time-as-Feature reshape instead of learnable embeddings (ablation) | modular forecasting pipeline · ETT datasets | marginalized MSE: ETTh1 avg 0.4398 vs point-wise 0.4367 | No clear effect |
| add Cycle (learnable periodic) input transformation instead of RevIN-only baseline (ablation) | point/variate-wise embedding models · ETTh1 | marginalized MSE: 0.3975 -> 0.3891 | Improved |
| add Cycle input transformation (ablation) | point/variate-wise embedding models · Weather | marginalized MSE: 0.1882 -> 0.1917 | Hurt |
| add naive multi-scale downsampling instead of RevIN-only (ablation) | point/variate-wise embedding models · ETTh2 and Weather | marginalized MSE: ETTh2 0.3013 -> 0.3071; Weather 0.1882 -> 0.1950 | Hurt |
| add naive multi-scale downsampling instead of RevIN-only (ablation) | point/variate-wise embedding models · Electricity | marginalized MSE: 0.2061 -> 0.1952 | Improved |
| add trend-seasonal decomposition instead of RevIN-only (ablation) | point/variate-wise embedding models · ETTh1 | marginalized MSE: 0.3975 -> 0.3947 | Mixed |
| replace attention with spectral (SimpleTM) encoder under variate-wise embedding (ablation) | iTransformer · ETTm2 | avg marginalized MSE: 0.2870 -> 0.2697 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- iTransformer — iTransformer: Inverted Transformers Are Effective for Time Series Forecasting (ICLR 2024) · better on 2 of 2 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
- Unlocking the Power of Patch: Patch-Based MLP for Long-Term Time Series Forecasting (AAAI 2025 · similarity 0.87)
- TimeXer: Empowering Transformers for Time Series Forecasting with Exogenous Variables (NeurIPS 2024 · similarity 0.87)
- Improving Time Series Forecasting via Instance-aware Post-hoc Revision (NeurIPS 2025 · similarity 0.87)
- A Closer Look at Transformers for Time Series Forecasting: Understanding Why They Work and Where They Struggle (ICML 2025 · similarity 0.87)
- An Analysis of Linear Time Series Forecasting Models (ICML 2024 · similarity 0.86)
- TimeEmb: A Lightweight Static-Dynamic Disentanglement Framework for Time Series Forecasting (NeurIPS 2025 · 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.
