Structural query-by-committee

TL;DR

This paper introduces a unified framework for various interactive learning tasks, extending the query-by-committee algorithm, and analyzes its theoretical properties and empirical performance under different noise conditions.

Contribution

It generalizes the query-by-committee algorithm to a broader setting and provides theoretical and empirical analysis of its consistency and convergence rates.

Findings

The generalized algorithm is consistent under certain conditions.

Convergence rates are established both theoretically and empirically.

Performance is evaluated with and without noise.

Abstract

In this work, we describe a framework that unifies many different interactive learning tasks. We present a generalization of the {\it query-by-committee} active learning algorithm for this setting, and we study its consistency and rate of convergence, both theoretically and empirically, with and without noise.