Information-theoretic generalization bounds for black-box learning algorithms

Hrayr Harutyunyan, Maxim Raginsky, Greg Ver Steeg, Aram Galstyan

Keywords: generalization, information theory, deep learning theory, stability, memorization

TL;DR: We derive information-theoretic generalization bounds for supervised learning algorithms based on the information contained in predictions rather that in the output of the training algorithm.

Abstract: We derive information-theoretic generalization bounds for supervised learning algorithms based on the information contained in predictions rather than in the output of the training algorithm. These bounds improve over the existing information-theoretic bounds, are applicable to a wider range of algorithms, and solve two key challenges: (a) they give meaningful results for deterministic algorithms and (b) they are significantly easier to estimate. We show experimentally that the proposed bounds closely follow the generalization gap in practical scenarios for deep learning.

Code Of Conduct: I certify that all co-authors of this work have read and commit to adhering to the NeurIPS Statement on Ethics, Fairness, Inclusivity, and Code of Conduct.

Code: https://github.com/hrayrhar/f-CMI