Research map / Time series / Causal and generative time series
COSA: Context-aware Output-Space Adapter for Test-Time Adaptation in Time Series Forecasting
Jeonghwan Im, Hyuk-Yoon Kwon
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 CALR learning-rate scheduling (ablation) | COSA · window length study | accuracy: up to 12.13% | Improved |
| add CALR learning-rate scheduling (ablation) | COSA · window length study | wall-clock time: 21.34% reduction | Improved |
| use single linear layer instead of 2-layer MLP adapter (ablation) | COSA · six benchmarks | MSE: 5.71% better | Improved |
| use single linear layer instead of 2-layer MLP adapter (ablation) | COSA · six benchmarks | wall-clock time: 34.95% faster | Improved |
| increase adaptation steps S (ablation) | COSA · six benchmarks | test MSE | Improved |
| increase context length K (ablation) | COSA · six benchmarks | test MSE | Improved |
| decrease batch size B (more frequent adaptation) (ablation) | COSA · six benchmarks | test MSE | Mixed |
| use mean aggregation instead of median for context (ablation) | COSA · six benchmarks | MSE | Improved |
| use online PAAS batch size (COSA-P) instead of fixed B=48 (ablation) | COSA · ETTh1 | MSE: L=720 .5958 -> .5638 | Improved |
| use online PAAS batch size (COSA-P) instead of fixed B=48 (ablation) | COSA · Exchange Rate (L=720) | MSE: .3421 -> .4460 | Hurt |
| use COSA output adapter instead of dual input/output adapters | TAFAS/PETSA · L=720 horizon | MSE: 28.21% and 21.96% (vs other methods) | Improved |
| add COSA on top of normalizers (RevIN, DDN) | normalizers · six benchmarks | MSE: 16.8∼16.9% | Improved |
Closest papers by meaning
- Battling the Non-stationarity in Time Series Forecasting via Test-time Adaptation (AAAI 2025 · similarity 0.89)
- Improving Time Series Forecasting via Instance-aware Post-hoc Revision (NeurIPS 2025 · similarity 0.87)
- AdaPTS: Adapting Univariate Foundation Models to Probabilistic Multivariate Time Series Forecasting (ICML 2025 · similarity 0.87)
- Learning the Context of Errors: Black-Box Online Adaptation of Time Series Foundation Models (NeurIPS 2026 · similarity 0.86)
- ARM: Refining Multivariate Forecasting with Adaptive Temporal-Contextual Learning (ICLR 2024 · similarity 0.86)
- Learning with Calibration: Exploring Test-Time Computing of Spatio-Temporal Forecasting (NeurIPS 2025 · 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.
