Research map / Time series / Foundation models for time series
Test-Time Efficient Pretrained Model Portfolios for Time Series Forecasting
Mert Kayaalp, Ali Caner Turkmen, Oleksandr Shchur, Pedro Mercado, Abdul Fatir Ansari, Michael Bohlke-Schneider, Bernie Wang
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 |
|---|---|---|---|
| partition training data by frequency instead of domain | Chroma portfolio model selection (4m) · BM2 relative WQL | WQL: 0.939 -> 0.880 | Improved |
| partition training data by frequency instead of domain | Chroma portfolio model selection (4m) · BM2 relative MASE | MASE: 0.933 -> 0.887 | Improved |
| use greedy ensemble instead of model selection | Chroma frequency portfolio (4m) · BM2 relative WQL | WQL: 0.880 -> 0.890 | No clear effect |
| use greedy ensemble instead of model selection | Chroma frequency portfolio (4m) · BM2 relative MASE | MASE: 0.887 -> 0.878 | Improved |
| combine several generalists with different seeds instead of one generalist | single generalist (4m) · BM2 relative WQL | WQL: 0.960 -> 0.942 | No clear effect |
| add frequency specialists to the generalist portfolio (vs single generalist) | Chroma 1m · BM2 relative WQL | WQL: 1.000 -> 0.918 | Improved |
| add generalist to portfolio (ablation) | Chroma portfolio · BM2 MASE | MASE: 0.016 | No clear effect |
| add generalist to portfolio (ablation) | Chroma portfolio · BM2 WQL | WQL: 0.017 | No clear effect |
| use model selection or greedy ensemble instead of performance-weighted averaging (ablation) | Chroma portfolio · BM2 MASE | MASE: 0.186 | Improved |
| use model selection or greedy ensemble instead of simple averaging (ablation) | Chroma portfolio · BM2 MASE | MASE: 0.242 | Improved |
| use model selection or greedy ensemble instead of simple averaging (ablation) | Chroma portfolio · BM2 WQL | WQL: 0.180 | Improved |
| post-train specialists from a generalist instead of training from scratch (ablation) | Chroma portfolio · BM2 MASE | MASE: 0.002 | No clear effect |
Closest papers by meaning
- Less is More: Unlocking Specialization of Time Series Foundation Models via Structured Pruning (NeurIPS 2025 · similarity 0.87)
- SEMPO: Lightweight Foundation Models for Time Series Forecasting (NeurIPS 2025 · similarity 0.86)
- Towards a General Time Series Forecasting Model with Unified Representation and Adaptive Transfer (ICML 2025 · similarity 0.86)
- In-Context Fine-Tuning for Time-Series Foundation Models (ICML 2025 · similarity 0.86)
- TimePerceiver: An Encoder-Decoder Framework for Generalized Time-Series Forecasting (NeurIPS 2025 · similarity 0.85)
- Rethinking Post-Training Recipes for Multimodal Time-Series Forecasting (NeurIPS 2026 · similarity 0.85)
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.
