Research map / Time series / Long-term forecasting
See More, Forecast Better and Faster: Enhancing Time Series Foundation Models via Inference-Time Plug-and-Play Downsampling
Longlong Xu, Zeyan Li, Xiao He, Zhaoyang Yu, Dazhong Wen, Mingze Sun, Changhua Pei, Dan Pei
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 SPRINT downsampling wrapper | Chronos · ETTm1 | MSE: 0.615 -> 0.414 | Improved |
| add SPRINT downsampling wrapper | VisionTS · ETTm1 | MSE: 0.709 -> 0.452 | Improved |
| add SPRINT downsampling wrapper | Moirai · ETTm1 | MSE: 0.369 -> 0.356 | Improved |
| add SPRINT downsampling wrapper | Timer · Service | MSE: 0.745 -> 0.494 | Improved |
| add SPRINT downsampling wrapper | ToTo · Service | MSE: 0.166 -> 0.171 | Hurt |
| add SPRINT downsampling wrapper | VisionTS · Service | MSE: 0.426 -> 0.491 | Hurt |
| add SPRINT downsampling wrapper | TSFMs (ultra-long L=1440) · Ultra-long-term forecasting | average MSE reduction: 18.45% to 32.39% | Improved |
| add SPRINT so the model forecasts beyond its native length | Timer · ETTm1 L>720 | validity of forecast: NaN -> valid up to L=5760 | Improved |
| add SPRINT downsampling wrapper | Chronos · ETTm1 efficiency | inference time (ms): 1707.72 -> 28.45 | Improved |
| add SPRINT downsampling wrapper | VisionTS · ETTm1 efficiency | inference time (ms): 6.43 -> 7.33 | Hurt |
| add seasonal-trend decomposition before downsampling (ablation) | SPRINT · ETTh2, ETTm2, Weather, Service | MSE | Improved |
| use moving-average instead of STL decomposition (ablation) | SPRINT · ETTh2, ETTm2, Weather, Service | MSE | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- VisionTS — VisionTS: Visual Masked Autoencoders Are Free-Lunch Zero-Shot Time Series Forecasters (ICML 2025) · better on 2 of 2 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
- Less is More: Unlocking Specialization of Time Series Foundation Models via Structured Pruning (NeurIPS 2025 · similarity 0.89)
- TS-RAG: Retrieval-Augmented Generation based Time Series Foundation Models are Stronger Zero-Shot Forecaster (NeurIPS 2025 · similarity 0.89)
- SEMPO: Lightweight Foundation Models for Time Series Forecasting (NeurIPS 2025 · similarity 0.88)
- TimeBase: The Power of Minimalism in Efficient Long-term Time Series Forecasting (ICML 2025 · similarity 0.87)
- Learning Pattern-Specific Experts for Time Series Forecasting Under Patch-level Distribution Shift (NeurIPS 2025 · similarity 0.87)
- Multi-Scale Finetuning for Encoder-based Time Series Foundation Models (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.
