Jupyter notebook for producing the plots of:

Comparing Comparators in Generalization Bounds

Fredrik Hellström and Benjamin Guedj

https://arxiv.org/abs/2310.10534

AISTATS 2024

Suggested citation:

@inproceedings{hellstrom2023comparing, title = {Comparing Comparators in Generalization Bounds}, author={Fredrik Hellström and Benjamin Guedj}, year={2024}, booktitle={Proceedings of The 27th International Conference on Artificial Intelligence and Statistics [AISTATS]}, address = {Valencia, Spain}, month = {5}, abstract = {We derive generic information-theoretic and PAC-Bayesian generalization bounds involving an arbitrary convex \emph{comparator} function, which measures the discrepancy between the training loss and the population loss. The bounds hold under the assumption that the cumulant-generating function (CGF) of the comparator is upper-bounded by the corresponding CGF within a family of bounding distributions. We show that the tightest possible bound is obtained with the comparator being the convex conjugate of the CGF of the bounding distribution, also known as the Cram'er function. This conclusion applies more broadly to generalization bounds with a similar structure. This confirms the near-optimality of known bounds for bounded and sub-Gaussian losses and leads to novel bounds under other bounding distributions.}, doi = {10.48550/arXiv.2310.10534} }