On Statistical Bias In Active Learning: How and When To Fix It

TL;DR

This paper analyzes the bias introduced by active learning, formalizes its effects, and proposes novel correction methods, revealing when bias removal improves model training, especially for neural networks with limited data.

Contribution

It formalizes the bias in active learning, investigates its effects, and introduces corrective weights to mitigate bias when beneficial, explaining empirical successes.

Findings

Bias can be harmful or helpful depending on the context.

Corrective weights can effectively remove bias when needed.

Bias may aid training overparameterized models with limited data.

Abstract

Active learning is a powerful tool when labelling data is expensive, but it introduces a bias because the training data no longer follows the population distribution. We formalize this bias and investigate the situations in which it can be harmful and sometimes even helpful. We further introduce novel corrective weights to remove bias when doing so is beneficial. Through this, our work not only provides a useful mechanism that can improve the active learning approach, but also an explanation of the empirical successes of various existing approaches which ignore this bias. In particular, we show that this bias can be actively helpful when training overparameterized models -- like neural networks -- with relatively little data.