Disentanglement based Active Learning

TL;DR

Disentanglement based Active Learning (DAL) uses self-supervision and InfoGAN to automatically label most data points, significantly reducing human labeling effort in GAN-based active learning for image classification.

Contribution

The paper introduces DAL, a novel active learning method leveraging disentanglement and InfoGAN to automate labeling and improve performance over existing GAN-based approaches.

Findings

DAL reduces human labeling effort significantly.

DAL outperforms existing GAN-based active learning methods.

Effective label correction reduces noise from automatic labels.

Abstract

We propose Disentanglement based Active Learning (DAL), a new active learning technique based on self-supervision which leverages the concept of disentanglement. Instead of requesting labels from human oracle, our method automatically labels the majority of the datapoints, thus drastically reducing the human labeling budget in Generative Adversarial Net (GAN) based active learning approaches. The proposed method uses Information Maximizing Generative Adversarial Nets (InfoGAN) to learn disentangled class category representations. Disagreement between active learner predictions and InfoGAN labels decides if the datapoints need to be human-labeled. We also introduce a label correction mechanism that aims to filter out label noise that occurs due to automatic labeling. Results on three benchmark datasets for the image classification task demonstrate that our method achieves better performance compared to existing GAN-based active learning approaches.