Minimax Active Learning
Sayna Ebrahimi, William Gan, Dian Chen, Giscard Biamby, Kamyar Salahi, Michael Laielli, Shizhan Zhu, Trevor Darrell
UC Berkeley

  [Paper]   [BibTeX]   [Code]

Abstract

Active learning aims to develop label-efficient algorithms by querying the most representative samples to be labeled by a human annotator. Current active learning techniques either rely on model uncertainty to select the most uncertain samples or use clustering or reconstruction to choose the most diverse set of unlabeled examples. While uncertainty-based strategies are susceptible to outliers, solely relying on sample diversity does not capture the information available on the main task. In this work, we develop a semi-supervised minimax entropy-based active learning algorithm that leverages both uncertainty and diversity in an adversarial manner. Our model consists of an entropy minimizing feature encoding network followed by an entropy maximizing classification layer. This minimax formulation reduces the distribution gap between the labeled/unlabeled data, while a discriminator is simultaneously trained to distinguish the labeled/unlabeled data. The highest entropy samples from the classifier that the discriminator predicts as unlabeled are selected for labeling. We evaluate our method on various image classification and semantic segmentation benchmark datasets and show superior performance over the state-of-the-art methods.


Motivation

Diversity-based task-agnostic active learning approaches perform unsupervised learning on all data and adversarially train a discriminator to predict whether to label a sample. On the left, the red-circled samples show a failure case for such methods, where the unlabeled samples closest to the discriminator's decision boundary are all from the same class and do not provide a diverse set of data to label. On the right our proposed approach MAL performs semi-supervised learning on all data and uses the label information to learn per-class prototypes and then train a binary classifier to predict which samples to label. In contrast to the diversity-based approach, the selected red-circled samples are chosen because they are farther from the class prototypes, and as a result, come from different classes.


Paper
Sayna Ebrahimi, William Gan, Dian Chen, Giscard Biamby, Kamyar Salahi, Michael Laielli, Shizhan Zhu, Trevor Darrell.
Minimax Active Learning.


[Bibtex]

Correspondance to sayna [at] berkeley [dot] edu.