Research map / Time series / Foundation models for time series
TimeCAP: A Channel-Aware Pre-Training Framework for Multivariate Time Series Forecasting
Chuanru Ren, Yao Lu, Tianjin Huang, Haowen Zheng, Hengde Zhu, Yunyin Li, Hengxiao Li, Lu Liu
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 the autoregressive decoding head (ablation) | TimeCAP · Electricity (full-shot) | MSE: 0.186 -> 0.171 | Improved |
| add the autoregressive decoding head (ablation) | TimeCAP · Exchange (full-shot) | MSE: 0.347 -> 0.326 | Improved |
| add the one-shot decoding head (ablation) | TimeCAP · Electricity (full-shot) | MSE: 0.497 -> 0.171 | Improved |
| add the one-shot decoding head (ablation) | TimeCAP · ETT-avg (full-shot) | MSE: 0.399 -> 0.361 | Improved |
| add adaptive meta-routers and channel-aware cross-attention (ablation) | TimeCAP · Exchange (full-shot) | MSE: 0.349 -> 0.326 | Improved |
| add adaptive meta-routers and channel-aware cross-attention (ablation) | TimeCAP · Electricity (full-shot) | MSE: 0.181 -> 0.171 | Improved |
| use sigmoid-based fusion instead of direct weighted averaging (ablation) | TimeCAP · Electricity (full-shot) | MSE: 0.252 -> 0.171 | Improved |
| use sigmoid-based fusion instead of direct weighted averaging (ablation) | TimeCAP · Weather (full-shot) | MSE: 0.251 -> 0.250 | No clear effect |
| add self-supervised multi-domain pre-training | TimeCAP · Electricity (full-shot) | MSE: 0.244 -> 0.171 | Improved |
| add self-supervised pre-training before zero-shot forecasting | TimeCAP · ETTh-avg (zero-shot) | MSE: 0.631 -> 0.565 | Improved |
| add pre-training on 10% few-shot ETTh forecasting versus Timer-XL | Timer-XL · ETTh-avg (few-shot 10%) | MSE: 0.566 -> 0.474 | Improved |
| use TimeCAP instead of Timer-XL for efficiency | Timer-XL · ETTh2 zero-shot, horizon 720 | inference time (s): 9.97 -> 3.63 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 2 of 2 on most shared results.
- TimeDART — TimeDART: A Diffusion Autoregressive Transformer for Self-Supervised Time Series Representation (ICML 2025) · better on 3 of 4 shared results
- Timer-XL — Timer-XL: Long-Context Transformers for Unified Time Series Forecasting (ICLR 2025) · better on 4 of 4 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
- Revitalizing Multivariate Time Series Forecasting: Learnable Decomposition with Inter-Series Dependencies and Intra-Series Variations Modeling (ICML 2024 · similarity 0.91)
- ARM: Refining Multivariate Forecasting with Adaptive Temporal-Contextual Learning (ICLR 2024 · similarity 0.90)
- TimeCHEAT: A Channel Harmony Strategy for Irregularly Sampled Multivariate Time Series Analysis (AAAI 2025 · similarity 0.90)
- TimePFN: Effective Multivariate Time Series Forecasting with Synthetic Data (AAAI 2025 · similarity 0.90)
- Towards a General Time Series Forecasting Model with Unified Representation and Adaptive Transfer (ICML 2025 · similarity 0.89)
- TimePerceiver: An Encoder-Decoder Framework for Generalized Time-Series Forecasting (NeurIPS 2025 · similarity 0.89)
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.
