A Stein Identity for q q -Gaussians with Bounded Support
Sophia Sklaviadis, Thomas Möllenhoff, André F. T. Martins, Mário A. T. Figueiredo, Mohammad Emtiyaz Khan · 2026
Read A Stein Identity for q q -Gaussians with Bounded Support by Sophia Sklaviadis, Thomas Möllenhoff, André F. T. Martins, Mário A. T. Figueiredo, Mohammad Emtiyaz Khan (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-2603-03673.txt.
Description
Sophia Sklaviadis and four colleagues extend a workhorse identity of statistics beyond the Gaussian world. Stein's identity converts expectations into gradient formulas and powers generative modeling and stochastic optimization, but it was built for Gaussian distributions. Their 2026 paper derives the identity for q-Gaussians with bounded support, a flexible family that lives on a finite interval instead of the whole line.
The proof extends earlier Bonnet and Price type theorems and simplifies them using escort distributions, close cousins of the original law. The resulting gradient estimators look nearly identical to the Gaussian ones and are just as easy to implement. Experiments add a practical lure: bounded-support distributions can lower the variance of gradient estimators. The authors point to Bayesian deep learning and sharpness-aware minimization as likely beneficiaries.
Catalog record
| CREATOR | Sophia Sklaviadis, Thomas Möllenhoff, André F. T. Martins, Mário A. T. Figueiredo, Mohammad Emtiyaz Khan |
|---|---|
| YEAR | 2026 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 88718 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2603.03673