Forecasting extreme temperature events by Hankel-augmented contrastive learning
Chengyang Qin, Yueyang Ding, Peng Tao, Luonan Chen · 2026
Read Forecasting extreme temperature events by Hankel-augmented contrastive learning by Chengyang Qin, Yueyang Ding, Peng Tao, Luonan Chen (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/PMC13639192.txt.
Description
Qin and colleagues presented in 2026 a deep learning system, Hankelformer, for forecasting extreme temperature events such as sudden freezes and heatwaves. Standard time-series models often miss abrupt shifts because extreme events break sharply from past patterns. Their design builds Hankel matrices, structured tables of lagged values, to make extra views of the input data that keep time order intact while exposing local patterns.
Qin processed these views alongside the raw input in a dual-stream transformer model. Contrastive learning then trained the model to treat matching views as alike, hardening its features against shifting conditions. Tests covered three famous events, the Texas freeze, the Pacific Northwest heatwave, and an Antarctic heat event, plus six standard data sets. The system beat leading rivals throughout, cutting forecast error by up to about a third over the best baselines in mean squared error terms.
Catalog record
| CREATOR | Chengyang Qin, Yueyang Ding, Peng Tao, Luonan Chen |
|---|---|
| YEAR | 2026 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 53 KB |
| LICENSE | https://creativecommons.org/licenses/by/4.0 |
| RIGHTS | CC BY |
| SOURCE | Europe PMC |