E4GEN: Event-level Explainable Extreme-Enhanced Time-series Generation
Lin Jiang, Dahai Yu, Ximiao Li, Guang Wang · 2026
Read E4GEN: Event-level Explainable Extreme-Enhanced Time-series Generation by Lin Jiang, Dahai Yu, Ximiao Li, Guang Wang (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/arxiv-2606-01634.txt.
Description
Jiang and co-authors generate time series with extreme events kept faithful. Realistic series matter for science and real-world use. Past methods prize overall distributional fidelity in their output. Extreme events often get lost or distorted in the process. E4GEN is an explainable diffusion frame for extreme-aware generation. It gives systematic control over when, what, and how extremes form. An E-Activator learns when to switch the extreme signal on. It does so without disturbing trend or seasonality parts.
An E-Predictor infers what control signal each sample needs. It derives that signal from latent extreme data during generation. An E-Control network injects the signal layer by layer. Tests on six datasets with 17 metrics favor E4GEN. Gains span fidelity, extreme fidelity, and downstream utility.
Catalog record
| CREATOR | Lin Jiang, Dahai Yu, Ximiao Li, Guang Wang |
|---|---|
| YEAR | 2026 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 150 KB |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2606.01634