What If We Let Forecasting Forget? A Sparse Bottleneck for Cross-Variable Dependencies
Fan Zhang, Shiming Fan, Hua Wang · 2026
Read What If We Let Forecasting Forget? A Sparse Bottleneck for Cross-Variable Dependencies by Fan Zhang, Shiming Fan, Hua 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-2605-08289.txt.
Description
Fan Zhang, Shiming Fan, and Hua Wang argue that forecasting models should forget most cross-variable links. In multivariate series, real links depend on the current state. They also come with noise. Dense interaction layers can amplify fake correlations. They can smooth representations until predictions turn unreliable in the hard cases.
Their MS-FLOW framework treats interaction as a capacity-limited flow. The flow passes through a sparse bottleneck. Only a few key dependency paths stay open. Cross-variable signals pass under a strict communication budget. The routing keeps the links that matter for the present regime. It drops the spurious ones that dense models would amplify.
Catalog record
| CREATOR | Fan Zhang, Shiming Fan, Hua Wang |
|---|---|
| YEAR | 2026 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 77 KB |
| LICENSE | http://creativecommons.org/licenses/by-sa/4.0/ |
| RIGHTS | CC BY SA |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2605.08289