Neural Active Learning on Heteroskedastic Distributions

TL;DR

This paper investigates the failure modes of active learning algorithms on heteroskedastic data and introduces a new method that effectively filters noise to improve learning accuracy.

Contribution

It identifies the limitations of existing active learning methods on heteroskedastic distributions and proposes a novel filtering approach with a model difference scoring function.

Findings

Existing active learning algorithms fail catastrophically on heteroskedastic data.

The proposed method outperforms existing techniques on heteroskedastic datasets.

Filtering noisy examples improves active learning performance.

Abstract

Models that can actively seek out the best quality training data hold the promise of more accurate, adaptable, and efficient machine learning. Active learning techniques often tend to prefer examples that are the most difficult to classify. While this works well on homogeneous datasets, we find that it can lead to catastrophic failures when performed on multiple distributions with different degrees of label noise or heteroskedasticity. These active learning algorithms strongly prefer to draw from the distribution with more noise, even if their examples have no informative structure (such as solid color images with random labels). To this end, we demonstrate the catastrophic failure of these active learning algorithms on heteroskedastic distributions and propose a fine-tuning-based approach to mitigate these failures. Further, we propose a new algorithm that incorporates a model difference scoring function for each data point to filter out the noisy examples and sample clean examples that maximize accuracy, outperforming the existing active learning techniques on the heteroskedastic datasets. We hope these observations and techniques are immediately helpful to practitioners and can help to challenge common assumptions in the design of active learning algorithms.