<ul data-eligibleForWebStory="true"><li>Monotone classification involves identifying a function that classifies points in a set according to hidden labels.</li><li>The goal is to find a monotone function with minimal error in classification.</li><li>The error is measured by the number of points whose labels differ from the classifier's predicted values.</li><li>The cost of an algorithm in this context is determined by the number of points requiring label revelation.</li><li>This article explores the minimum cost needed to identify a monotone classifier with error at most a specified factor above the optimal error.</li><li>It presents nearly matching upper and lower bounds for different error factors.</li><li>Previous approaches to the problem could only achieve higher errors than the optimal by a fixed amount.</li></ul>

Monotone Classification with Relative Approximations

Discover more