Research map / Time series / Causal and generative time series
Time-Gated Multi-Scale Flow Matching for Time-Series Imputation
Hangtian Wang, Mahito Sugiyama
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 multi-scale heads (vs single scale s=1) (ablation) | TG-MSFM · Electricity | MSE: 0.116 -> 0.101 | Improved |
| add multi-scale heads (vs single scale s=1) (ablation) | TG-MSFM · ETTh1 | MSE: 0.158 -> 0.126 | Improved |
| add time gate (vs static mixing) (ablation) | TG-MSFM · Electricity | MSE: 0.212 -> 0.101 | Improved |
| add time gate (vs static mixing) (ablation) | TG-MSFM · ETTh1 | MSE: 0.147 -> 0.126 | Improved |
| use Heun integrator instead of Euler (ablation) | TG-MSFM · Electricity | MSE: 0.115 -> 0.101 | Improved |
| use Heun integrator instead of Euler (ablation) | TG-MSFM · ETTh1 | MAE: 0.257 -> 0.231 | Improved |
| increase ODE steps N from 150 to 300 (ablation) | TG-MSFM · Illness long-gap | RMSE: 0.334 -> 0.318 | Improved |
| increase ODE steps N from 300 to 500 (ablation) | TG-MSFM · Illness long-gap | RMSE: 0.318 -> 0.317 | No clear effect |
| add scale 1 to pyramid (from {2,4} to {1,2,4}) (ablation) | TG-MSFM · Illness long-gap | RMSE: 0.329 -> 0.318 | Improved |
| add scale 8 to pyramid (from {1,2,4} to {1,2,4,8}) (ablation) | TG-MSFM · Illness long-gap | RMSE: 0.318 -> 0.317 | No clear effect |
| reduce Transformer layers from 6 to 4 (ablation) | TG-MSFM · Illness long-gap | RMSE: 0.318 -> 0.323 | Hurt |
| TG-MSFM vs CSDI diffusion baseline across 5 seeds | CSDI · Electricity | MSE: 0.568 -> 0.112 | 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.
- CSDI — CSDI: Conditional Score-based Diffusion Models for Probabilistic Time Series Imputation (NeurIPS 2021) · better on 3 of 3 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
- MissPath-FM: Flow Matching with Structured Priors for Partially Observed Time Series (NeurIPS 2026 · similarity 0.89)
- PrismFlow: Residual Dynamics for Flow Matching in Time-Series Generation (NeurIPS 2026 · similarity 0.89)
- What Time Is It? How Data Geometry Makes Time Conditioning Optional for Flow Matching (NeurIPS 2026 · similarity 0.89)
- Discretizing Continuous Time Series for Imputation with Masked Diffusion Training (NeurIPS 2026 · similarity 0.89)
- Zero-shot Imputation with Foundation Inference Models for Dynamical Systems (ICLR 2025 · similarity 0.88)
- ProCTI: Prototype-Refined Global Conditioning for Diffusion-Based Time Series Imputation (NeurIPS 2026 · 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.
