Abstract:Deep neural networks generalize well despite being heavily overparameterized, in apparent contradiction with classical learning theory based on uniform convergence over fixed hypothesis spaces. Uniform bounds over the entire parameter space are vacuous in this regime, and recent work has shown that non-vacuous guarantees can be recovered by restricting attention to the part of parameter space that the algorithm actually visits. This survey paper organizes this line of work around three steps: extending PAC-Bayesian theory to random, data-dependent hypothesis sets (arXiv:2404.17442); refining the complexity term with geometric and topological descriptors of the optimization trajectory, including fractal dimensions, alpha-weighted lifetime sums, and positive magnitude (arXiv:2006.09313, arXiv:2302.02766, arXiv:2407.08723); and replacing the resulting information-theoretic terms by stability assumptions (arXiv:2507.06775). We unify these contributions around a single template inequality and a head-to-head comparison of the resulting bounds.
| Comments: | 15 pages, 4 figures, 3 tables. The LaTeX source uses the JMLR preprint style (this http URL) and BibTeX (this http URL). Central references in arXiv form include arXiv:2404.17442, arXiv:2006.09313, arXiv:2302.02766, arXiv:2407.08723, and arXiv:2507.06775 |
| Subjects: | Machine Learning (stat.ML); Machine Learning (cs.LG) |
| ACM classes: | I.2.6; F.2.0; G.3 |
| Cite as: | arXiv:2605.13913 [stat.ML] |
| (or arXiv:2605.13913v1 [stat.ML] for this version) | |
| https://doi.org/10.48550/arXiv.2605.13913 arXiv-issued DOI via DataCite |
Submission history
From: Hubert Leroux [view email]
[v1]
Wed, 13 May 2026 09:03:50 UTC (72 KB)