On Statistical Bias In Active Learning: How and When To Fix It
This addresses a fundamental statistical issue in active learning for practitioners dealing with expensive labeling, though it appears incremental in analyzing existing bias rather than proposing a completely new paradigm.
The paper formalizes the statistical bias introduced by active learning when training data no longer follows the population distribution, showing this bias can be harmful or helpful depending on context. It introduces corrective weights to remove bias when beneficial and demonstrates the bias can be actively helpful when training overparameterized models like neural networks with limited data.
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.