Research map / Time series / Causal and generative time series
Rethinking Time-Series Imputation as Conditional Inference along Temporal Evolution
Yu Fan, Yang Yang, Yufan Guo, Huazhong Yang, Pengjun 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 |
|---|---|---|---|
| use CBiT instead of the best baseline (ImputeFormer), 10% missing | CBiT · ETTh1 10% missing (joint supervision) | Masked MAE: 0.2369 -> 0.1658 | Improved |
| use CBiT instead of the best baseline (ImputeFormer), 10% missing | CBiT · Power 10% missing (joint supervision) | Masked MAE: 0.0485 -> 0.0336 | Improved |
| use CBiT instead of the best baseline (SAITS), 50% missing | CBiT · WTH 50% missing (joint supervision) | Masked MAE: 0.1514 -> 0.1487 | Improved |
| use CBiT instead of the best baseline (SSSD), 10% missing | CBiT · ETTh1 10% missing (missing-only supervision) | Masked MAE: 0.2821 -> 0.2002 | Improved |
| use CBiT instead of the best baseline (UNiTS), 10% missing | CBiT · ETTh2 10% missing (missing-only supervision) | Masked MAE: 0.1680 -> 0.1597 | Improved |
| use CBiT instead of ImputeFormer, 50% missing | CBiT · Power 50% missing (missing-only supervision) | Masked MAE: 0.0538 -> 0.0596 | No clear effect |
| add reconstruction loss on observed entries (joint vs missing-only supervision) (ablation) | CBiT · ETTh1 10% missing | Masked MAE: 0.2002 -> 0.1658 | Improved |
| add reconstruction loss on observed entries (joint vs missing-only supervision) (ablation) | CBiT · Power 50% missing | Masked MAE: 0.0596 -> 0.0557 | Improved |
| add reconstruction loss on observed entries (joint vs missing-only supervision) (ablation) | CBiT · WTH 50% missing | Masked MAE: 0.1777 -> 0.1487 | Improved |
| use CBiT instead of UNiTS in compute | CBiT · Computational cost | FLOPs (G): 2.2620 -> 0.0586 | Improved |
| use CBiT instead of SAITS in parameters | CBiT · Computational cost | Params (M): 0.2045 -> 0.2753 | Hurt |
Closest papers by meaning
- Discretizing Continuous Time Series for Imputation with Masked Diffusion Training (NeurIPS 2026 · similarity 0.89)
- ProCTI: Prototype-Refined Global Conditioning for Diffusion-Based Time Series Imputation (NeurIPS 2026 · similarity 0.88)
- Diffusion Transformers for Imputation: Statistical Efficiency and Uncertainty Quantification (NeurIPS 2025 · similarity 0.88)
- BayOTIDE: Bayesian Online Multivariate Time Series Imputation with Functional Decomposition (ICML 2024 · similarity 0.87)
- Glocal Information Bottleneck for Time Series Imputation (NeurIPS 2025 · similarity 0.87)
- Frequency-aware Generative Models for Multivariate Time Series Imputation (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.
