Meta-Statistical Learning: Supervised Learning of Statistical Estimators
Maxime Peyrard ♢ \diamondsuit Kyunghyun Cho ♣ \clubsuit ♢ \diamondsuit Université Grenoble Alpes, CNRS, Grenoble INP, LIG ♣ \clubsuit New York University, Prescient Design, Genentech [email protected] [email protected] · 2025
Read Meta-Statistical Learning: Supervised Learning of Statistical Estimators by Maxime Peyrard ♢ \diamondsuit Kyunghyun Cho ♣ \clubsuit ♢ \diamondsuit Université Grenoble Alpes, CNRS, Grenoble INP, LIG ♣ \clubsuit New York University, Prescient Design, Genentech [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-2502-12088.txt.
Description
Maxime Peyrard and Kyunghyun Cho suggest teaching neural networks to invent statistical tests, in this 2025 paper. The field runs on estimators, functions that turn samples into guesses about hidden quantities. Crafting one with provable virtues — unbiased, low-variance, well-calibrated — is often very hard. For some targets, an estimator with no bias in every case provably does not exist.
Their meta-statistical learning recasts design as training. Whole data sets flow into permutation-invariant networks such as Set Transformers, which learn to output the target property in one step. The trained network is the estimator, and it can still be audited with classical frequentist tools later. Two worked cases anchor the claim: a learned test for normality, framed as classification, and a learned estimator of mutual information, framed as regression. Both performed strongly even with small models. The authors see a road toward automating estimator discovery itself.
Catalog record
| CREATOR | Maxime Peyrard ♢ \diamondsuit Kyunghyun Cho ♣ \clubsuit ♢ \diamondsuit Université Grenoble Alpes, CNRS, Grenoble INP, LIG ♣ \clubsuit New York University, Prescient Design, Genentech [email protected] [email protected] |
|---|---|
| YEAR | 2025 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 60 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2502-12088