State evolution beyond first-order methods I: Rigorous predictions and finite-sample guarantees
2025
Read State evolution beyond first-order methods I: Rigorous predictions and finite-sample guarantees (2025) 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-2507-19611.txt.
Description
Michael Celentano and co-authors extend a powerful prediction tool for iterative algorithms into harder territory. Many modern methods, from signal recovery to machine learning, can be tracked by state evolution, a deterministic recursion that forecasts their random behavior. Earlier theory covered only simple first-order updates, applied coordinate by coordinate.
The new toolbox handles algorithms that mix first-order steps with saddle-point updates and that refuse to separate by coordinate. The authors prove the prediction is exact in the large-system limit. They also bound how far a real, finite run can stray from the forecast. Technical ingredients include a Hilbert space lifting that pins down the recursion's form and a Gaussian comparison argument adapted to sequential data.
Catalog record
| YEAR | 2025 |
|---|---|
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 269 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2507-19611