Block-Sample MAC-Bayes Generalization Bounds
Matthias Frey, Jingge Zhu, Michael C. Gastpar · 2026
Read Block-Sample MAC-Bayes Generalization Bounds by Matthias Frey, Jingge Zhu, Michael C. Gastpar (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-2602-12605.txt.
Description
PAC-Bayes bounds give strong guarantees on generalisation error, but a related family bounds expected error. Matthias Frey led this work with two co-authors, giving block-sample MAC-Bayes bounds, where MAC stands for mean approximately correct. The family generalises an expectation version of PAC-Bayes bounds, with divergence terms depending only on subsets, or blocks, of training data.
That block structure promises tighter bounds than older bounds. A numerical example illustrates the gain. The original bound stays vacuous for any prior, while the new family gives finite bounds for fitting block sizes. The paper asks whether high-probability versions are possible. It answers no. An example shows no PAC-Bayes bound can do both at once. It cannot vanish faster than one over log n when the MAC-Bayes bound vanishes like n to the minus one half, and depend on allowed error chance in log form.
Catalog record
| CREATOR | Matthias Frey, Jingge Zhu, Michael C. Gastpar |
|---|---|
| YEAR | 2026 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 61 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2602-12605