Research map / Time series / Long-term forecasting
MSDformer: Multi-Scale Discrete Transformer for Time Series Generation
Shibo Feng, Zhicheng Chen, Xi Xiao, Zhong Zhang, Qing Li, Xingyu Gao, Peilin Zhao
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 codebook reset (ablation) | MSDformer · fMRI | Discriminative score: 0.423 -> 0.005 | Improved |
| add codebook reset (ablation) | MSDformer · fMRI | Context-FID: 14.664 -> 0.009 | Improved |
| add similarity-driven vector quantization (ablation) | MSDformer · fMRI | Context-FID: 0.019 -> 0.009 | Improved |
| add EMA codebook update (ablation) | MSDformer · fMRI | Context-FID: 0.011 -> 0.009 | Improved |
| add token type encoding (TTE) (ablation) | MSDformer · fMRI | Discriminative score: 0.007 -> 0.005 | Improved |
| add token type encoding (TTE) (ablation) | MSDformer · fMRI | Context-FID: 0.011 -> 0.009 | Improved |
| use two scales [128,512] instead of single codebook 512 (ablation) | SDformer · fMRI (length 24) | Discriminative score: 0.017 -> 0.005 | Improved |
| enlarge single-scale codebook from 512 to 640 (ablation) | SDformer · fMRI (length 24) | Discriminative score: 0.017 -> 0.017 | No clear effect |
| use downsampling factor 8 instead of 2 in single scale (ablation) | MSDformer K=1 · fMRI | Context-FID: 0.015 -> 0.092 | Hurt |
| add a second scale r=[2,4] to single scale r=[2] (ablation) | MSDformer · fMRI | Discriminative score: 0.017 -> 0.005 | Improved |
| add third scale to [2,4] on Stocks (ablation) | MSDformer · Stocks | Discriminative score: 0.005 -> 0.012 | Hurt |
| use multi-scale MSDformer instead of single-scale SDformer (ablation) | MSDformer · fMRI (short-term) | Discriminative score: 0.017 -> 0.005 | 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.
- Diffusion-TS — Diffusion-TS: Interpretable Diffusion for General Time Series Generation (ICLR 2024) · better on 6 of 6 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
- SDformer: Similarity-driven Discrete Transformer For Time Series Generation (NeurIPS 2024 · similarity 0.89)
- HDT: Hierarchical Discrete Transformer for Multivariate Time Series Forecasting (AAAI 2025 · similarity 0.89)
- Time-MMD: Multi-Domain Multimodal Dataset for Time Series Analysis (NeurIPS 2024 · similarity 0.87)
- Enhancing Foundation Models for Time Series Forecasting via Wavelet-based Tokenization (ICML 2025 · similarity 0.87)
- Multi-Scale Finetuning for Encoder-based Time Series Foundation Models (NeurIPS 2025 · similarity 0.87)
- Forging Time Series with Language: A Large Language Model Approach to Synthetic Data Generation (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.
