Research map / Time series / Causal and generative time series
From Observations to States: Latent Time Series Forecasting
Jie Yang, Yifan Hu, Yuante Li, Kexin Zhang, Kaize Ding, Philip S. Yu
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 |
|---|---|---|---|
| Train backbone to forecast latent states (LatentTSF) instead of observations | iTransformer · Electricity (avg) | MSE: 0.268 -> 0.194 | Improved |
| Train backbone to forecast latent states (LatentTSF) | PatchTST · Electricity (avg) | MSE: 0.389 -> 0.207 | Improved |
| Train backbone to forecast latent states (LatentTSF) | TimeXer · Traffic (avg) | MSE: 1.270 -> 0.636 | Improved |
| Train backbone to forecast latent states (LatentTSF) | DLinear · Traffic (avg) | MSE: 0.525 -> 0.552 | Hurt |
| Train backbone to forecast latent states (LatentTSF) | iTransformer · ETTh1 (avg) | MSE: 0.485 -> 0.441 | Improved |
| Train backbone to forecast latent states without Align loss versus vanilla DLinear (ablation) | DLinear · Electricity | MSE: 8.8% MSE reduction | Improved |
| Add alignment loss in latent space on top of latent prediction (ablation) | LatentTSF DLinear · five datasets | MSE: ranking (a) > (b) | Improved |
| Apply alignment loss in observation space instead (ablation) | DLinear · five datasets | MSE: (c) approx (d) | No clear effect |
| Add observation-space perceptual loss and fine-tune the autoencoder (ablation) | LatentTSF iTransformer · Electricity | MSE: best 0.297 vs default 0.194 | Hurt |
| Replace pre-trained autoencoder with randomly initialised 2-layer MLPs (ablation) | LatentTSF iTransformer · Electricity | MSE: best 0.351 vs 0.297 with pre-trained AE | Hurt |
| Increase Pred loss weight from 0 to moderate value (5), then 10 (ablation) | LatentTSF · multiple datasets | MAE: substantial MAE reduction, marginal gain at 10 | Improved |
| Raise Align loss weight from 0 to 5 (ablation) | LatentTSF · multiple datasets | MAE: sharp MAE drop, plateau beyond 5-10 | Improved |
Closest papers by meaning
- SynTSBench: Rethinking Temporal Pattern Learning in Deep Learning Models for Time Series (NeurIPS 2025 · similarity 0.88)
- TimeBase: The Power of Minimalism in Efficient Long-term Time Series Forecasting (ICML 2025 · similarity 0.88)
- TStruct: Learning Shared Temporal Structures for Long-Term Time-Series Forecasting (NeurIPS 2026 · similarity 0.88)
- Enhancing Foundation Models for Time Series Forecasting via Wavelet-based Tokenization (ICML 2025 · similarity 0.88)
- Attractor Memory for Long-Term Time Series Forecasting: A Chaos Perspective (NeurIPS 2024 · similarity 0.87)
- Online Time Series Forecasting with Theoretical Guarantees (NeurIPS 2025 · 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.
