1: \begin{abstract}

2: %\boldmath

3: We propose a new active learning strategy designed for deep neural networks.

4: The goal is to minimize the number of data annotation queried from an oracle during training.

5: Previous active learning strategies scalable for deep networks were mostly based on uncertain sample selection.

6: In this work, we focus on examples lying close to the decision boundary.

7: Based on theoretical works on margin theory for active learning, we know that such examples may help to considerably decrease the number of annotations.

8: While measuring the exact distance to the decision boundaries is intractable, we propose to rely on adversarial examples. We do not consider anymore them as a threat instead we exploit the information they provide on the distribution of the input space in order to approximate the distance to decision boundaries.

9: We demonstrate empirically that adversarial active queries yield faster convergence of CNNs trained on \textit{MNIST}, the \textit{Shoe-Bag} and the \textit{Quick-Draw} datasets.

10: \end{abstract}

11: