Research map / Time series / Foundation models for time series
Beyond Observations: Reconstruction Error-Guided Irregularly Sampled Time Series Representation Learning
Jiexi Liu, Meng Cao, Songcan Chen
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 pseudo-observations from reconstruction-error mixup (ablation) | iTimER pretraining · PhysioNet interpolation | MSE x1e-3: 5.03 -> 2.86 | Improved |
| add pseudo-observations (ablation) | iTimER pretraining · PhysioNet forecasting | MSE x1e-3: 6.10 -> 3.64 | Improved |
| add pseudo-observations (ablation) | iTimER pretraining · P12 | AUROC: 84.6 -> 85.7 | Improved |
| use mixup of error with last observation instead of random or constant values (ablation) | iTimER · PhysioNet interpolation | MSE x1e-3: 4.01 / 3.89 -> 2.86 | Improved |
| anchor error samples to an observed value instead of error only (ablation) | iTimER · PhysioNet interpolation | MSE x1e-3: 3.08 -> 2.86 | Improved |
| add Wasserstein loss (ablation) | iTimER · PhysioNet interpolation | MSE x1e-3: 3.17 -> 2.86 | Improved |
| add contrastive loss (ablation) | iTimER · PhysioNet interpolation | MSE x1e-3: 3.02 -> 2.86 | Improved |
| use moving average anchor instead of last observation (ablation) | iTimER · PhysioNet interpolation | MSE x1e-3: 2.86 -> 2.83 | No clear effect |
| use iTimER instead of ViTST/Warpformer | classification · P19 | AUROC: AUROC 87.1 vs ViTST 89.2 | Hurt |
| use iTimER instead of ISTS-PLM | forecasting · Human Activity | MSE x1e-3: 2.75 vs 2.58 | Hurt |
| use iTimER instead of ISTS-PLM | interpolation · MIMIC | MSE x1e-2: 1.47 -> 0.13 | 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.
- PrimeNet — PrimeNet: Pre-training for Irregular Multivariate Time Series (AAAI 2023) · better on 5 of 5 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
- T-Rep: Representation Learning for Time Series using Time-Embeddings (ICLR 2024 · similarity 0.86)
- TimeDART: A Diffusion Autoregressive Transformer for Self-Supervised Time Series Representation (ICML 2025 · similarity 0.85)
- TimeCHEAT: A Channel Harmony Strategy for Irregularly Sampled Multivariate Time Series Analysis (AAAI 2025 · similarity 0.84)
- IMTS is Worth Time $\times$ Channel Patches: Visual Masked Autoencoders for Irregular Multivariate Time Series Prediction (ICML 2025 · similarity 0.84)
- Learning to Embed Time Series Patches Independently (ICLR 2024 · similarity 0.84)
- Not All Data are Good Labels: On the Self-supervised Labeling for Time Series Forecasting (NeurIPS 2025 · similarity 0.84)
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.
