News Research

Paper Accepted to IEEE International Symposium on Information Theory (ISIT) 2021

Title: “Information-Theoretic Bounds on the Moments of the Generalization Error of Learning Algorithms” 
Authors: Gholamali Aminian, Laura Toni, Miguel R. D. Rodrigues

Link: https://arxiv.org/abs/2102.02016

Abstract: Generalization error bounds are critical to under- standing the performance of machine learning models. In this work, building upon a new bound of the expected value of an arbitrary function of the population and empirical risk of a learning algorithm, we offer a more refined analysis of the generalization behaviour of a machine learning models based on a characterization of (bounds) to their generalization error moments. We discuss how the proposed bounds – which also encompass new bounds to the expected generalization error – relate to existing bounds in the literature. We also discuss how the proposed generalization error moment bounds can be used to construct new generalization error high-probability bounds.

Index Terms: Population Risk, Empirical Risk, Generalization Error, Generalization Error Moments, Information Measures

0 comments on “Paper Accepted to IEEE International Symposium on Information Theory (ISIT) 2021

Leave a Reply

Fill in your details below or click an icon to log in:

WordPress.com Logo

You are commenting using your WordPress.com account. Log Out /  Change )

Google photo

You are commenting using your Google account. Log Out /  Change )

Twitter picture

You are commenting using your Twitter account. Log Out /  Change )

Facebook photo

You are commenting using your Facebook account. Log Out /  Change )

Connecting to %s

%d bloggers like this: