A note on active learning for smooth problems

TL;DR

This paper investigates the disagreement coefficient in active learning for smooth hypothesis classes, establishing that it scales linearly with the dimension of the hypothesis space, which has implications for learning efficiency.

Contribution

It provides a theoretical bound on the disagreement coefficient for smooth hypothesis classes, addressing an open question in the field.

Findings

Disagreement coefficient is $O(m)$ for certain smooth classes

Answers an open question from previous research

Implications for active learning efficiency

Abstract

We show that the disagreement coefficient of certain smooth hypothesis classes is $O(m)$, where $m$ is the dimension of the hypothesis space, thereby answering a question posed in \cite{friedman09}.