Privacy-preserving Active Learning on Sensitive Data for User Intent Classification
Machine Learning
2019-03-28 v1 Computation and Language
Machine Learning
Abstract
Active learning holds promise of significantly reducing data annotation costs while maintaining reasonable model performance. However, it requires sending data to annotators for labeling. This presents a possible privacy leak when the training set includes sensitive user data. In this paper, we describe an approach for carrying out privacy preserving active learning with quantifiable guarantees. We evaluate our approach by showing the tradeoff between privacy, utility and annotation budget on a binary classification task in a active learning setting.
Cite
@article{arxiv.1903.11112,
title = {Privacy-preserving Active Learning on Sensitive Data for User Intent Classification},
author = {Oluwaseyi Feyisetan and Thomas Drake and Borja Balle and Tom Diethe},
journal= {arXiv preprint arXiv:1903.11112},
year = {2019}
}
Comments
To appear at PAL: Privacy-Enhancing Artificial Intelligence and Language Technologies as part of the AAAI Spring Symposium Series (AAAI-SSS 2019)