Research map / Time series / Foundation models for time series
CoST: Contrastive Learning of Disentangled Seasonal-Trend Representations for Time Series Forecasting
Gerald Woo, Chenghao Liu, Doyen Sahoo, Akshat Kumar, Steven Hoi
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 |
|---|---|---|---|
| learn disentangled trend and seasonal representations with contrastive losses | CoST · Multivariate forecasting, average | MSE: 0.750 -> 0.590 | Improved |
| add seasonal representations to the trend-only model (ablation) | CoST · ETT datasets, multivariate | MSE: 0.789 -> 0.781 | Improved |
| add the mixture of autoregressive experts (ablation) | CoST trend branch · ETT datasets, multivariate | MSE: 0.882 -> 0.789 | Improved |
| add the learnable Fourier layer (ablation) | CoST seasonal branch · ETT datasets, multivariate | MSE: 0.895 -> 0.862 | Improved |
| add frequency-domain contrastive loss (ablation) | CoST seasonal branch · ETT datasets, multivariate | MSE: 0.905 -> 0.862 | Improved |
| train the architecture end-to-end with supervised forecasting loss instead of the representation-learning pipeline (ablation) | CoST · ETT datasets, multivariate | MSE: 0.781 -> 1.376 | Hurt |
| add contrastive loss to end-to-end supervised forecasting training (ablation) | CoST · ETT datasets, multivariate | MSE: 1.376 -> 1.477 | Hurt |
| replace the TCN backbone with LSTM (ablation) | CoST · ETT datasets, multivariate | MSE: 0.781 -> 0.928 | Hurt |
| replace the TCN backbone with Transformer (ablation) | CoST · ETT datasets, multivariate | MSE: 0.781 -> 0.863 | Hurt |
| replace ridge regression with linear regression (ablation) | CoST · ETT datasets, multivariate | MSE: 0.781 -> 1.472 | Hurt |
| replace ridge regression with kernel ridge regression (ablation) | CoST · ETT datasets, multivariate | MSE: 0.781 -> 0.868 | Hurt |
| vary the seasonal-loss weight alpha (ablation) | CoST · ETT datasets, univariate and multivariate | MSE: Multivariate: 0.810 to 0.780; Univariate: 0.120 to 0.102 | Mixed |
Later papers that compared with this method
7 later accepted papers report CoST as a baseline in their results tables; 7 of them report a better number on most of the results they share with it.
- Divide and Contrast: Learning Robust Temporal Features without Augmentation (ICML 2026 · better on 1 of 1 shared result)
- Parametric Augmentation for Time Series Contrastive Learning (ICLR 2024 · better on 4 of 4 shared results)
- TimesURL: Self-Supervised Contrastive Learning for Universal Time Series Representation Learning (AAAI 2024 · better on 2 of 2 shared results)
- Segment, Shuffle, and Stitch: A Simple Layer for Improving Time-Series Representations (NeurIPS 2024 · better on 2 of 2 shared results)
- PowerPM: Foundation Model for Power Systems (NeurIPS 2024 · better on 1 of 1 shared result)
- SimMTM: A Simple Pre-Training Framework for Masked Time-Series Modeling (NeurIPS 2023 · better on 2 of 2 shared results)
- Learning Latent Seasonal-Trend Representations for Time Series Forecasting (NeurIPS 2022 · better on 5 of 5 shared results)
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.
- TS2Vec — TS2Vec: Towards Universal Representation of Time Series (AAAI 2022) · better on 4 of 5 shared results
- Informer — Informer: Beyond Efficient Transformer for Long Sequence Time-Series Forecasting (AAAI 2021) · better on 1 of 1 shared result
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
- Self-Supervised Contrastive Learning for Long-term Forecasting (ICLR 2024 · similarity 0.89)
- Inference via Interpolation: Contrastive Representations Provably Enable Planning and Inference (NeurIPS 2024 · similarity 0.87)
- Soft Contrastive Learning for Time Series (ICLR 2024 · similarity 0.86)
- TimeCMA: Towards LLM-Empowered Multivariate Time Series Forecasting via Cross-Modality Alignment (AAAI 2025 · similarity 0.86)
- HDT: Hierarchical Discrete Transformer for Multivariate Time Series Forecasting (AAAI 2025 · similarity 0.86)
- Parametric Augmentation for Time Series Contrastive Learning (ICLR 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.
