On Conditional Stochastic Interpolation for Generative Nonlinear Sufficient Dimension Reduction
Shuntuo Xu, Zhou Yu, Jian [email protected], [email protected], [email protected] · 2025
Read On Conditional Stochastic Interpolation for Generative Nonlinear Sufficient Dimension Reduction by Shuntuo Xu, Zhou Yu, Jian [email protected], [email protected], [email protected] (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-2512-18971.txt.
Description
Shuntuo Xu, Zhou Yu, and Jian Huang use generative AI to compress data without losing what matters. Sufficient dimension reduction seeks a few combinations of variables that carry all the information about a response, but nonlinear versions lacked guarantees of finding everything, in the population or in samples. Their method, GenSDR, posted to arXiv in December 2025, builds the reduction with deep generative models trained through conditional stochastic interpolation.
GenSDR provably recovers all information in the central sigma-field, at both levels. At the sample level they prove the estimator consistent, in the sense of conditional distributions, using what deep generative models learn from data. An ensemble extension handles non-Euclidean responses, such as shapes or networks, widening its reach. Numerical results back the theory with strong test performance. Statisticians drowning in high-dimensional predictors gain a reduction they can trust not to throw away the signal.
Catalog record
| CREATOR | Shuntuo Xu, Zhou Yu, Jian [email protected], [email protected], [email protected] |
|---|---|
| YEAR | 2025 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 77 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2512-18971