Sample-efficient fine-tuning with textual prompts for time series forecasting
Kaibin Wei, Jianqiang Jing, Jiawei Liu, Qing Liu, Xiannian Xie · 2026
Read Sample-efficient fine-tuning with textual prompts for time series forecasting by Kaibin Wei, Jianqiang Jing, Jiawei Liu, Qing Liu, Xiannian Xie (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/plos-pone-0353130.txt.
Description
Kaibin Wei and four colleagues present this 2026 PLOS One methods paper on forecasting time series when training data are scarce. Their system turns number sequences into short written descriptions, lets a language model read those descriptions for extra context, and then adjusts a standard forecasting model's output accordingly. Most of the underlying model stays frozen, so adapting it to a new domain is cheap.
Across four datasets, including weather records, the prompt-guided system often beat the plain forecaster, with error reductions of around a tenth in some low-data settings. Gains were not uniform across every dataset and measure, which the authors note openly. The work suggests that written context can help forecasting models adjust to new fields without full retraining.
Catalog record
| CREATOR | Kaibin Wei, Jianqiang Jing, Jiawei Liu, Qing Liu, Xiannian Xie |
|---|---|
| YEAR | 2026 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 51138 |
| LICENSE | CC_BY |
| RIGHTS | CC BY |
| SOURCE | PLOS |