Effective Version Space Reduction for Convolutional Neural Networks

TL;DR

This paper introduces a new diameter-based active learning method for convolutional neural networks that effectively reduces the version space, outperforming existing approaches on multiple datasets.

Contribution

It proposes a novel diameter-based querying method called minimum Gibbs-vote disagreement and analyzes version space evolution in neural networks.

Findings

Diameter reduction methods outperform prior mass reduction.

Gibbs vote disagreement matches the best query methods.

Effective version space reduction improves active learning performance.

Abstract

In active learning, sampling bias could pose a serious inconsistency problem and hinder the algorithm from finding the optimal hypothesis. However, many methods for neural networks are hypothesis space agnostic and do not address this problem. We examine active learning with convolutional neural networks through the principled lens of version space reduction. We identify the connection between two approaches---prior mass reduction and diameter reduction---and propose a new diameter-based querying method---the minimum Gibbs-vote disagreement. By estimating version space diameter and bias, we illustrate how version space of neural networks evolves and examine the realizability assumption. With experiments on MNIST, Fashion-MNIST, SVHN and STL-10 datasets, we demonstrate that diameter reduction methods reduce the version space more effectively and perform better than prior mass reduction and other baselines, and that the Gibbs vote disagreement is on par with the best query method.