Private classification with limited labeled data

作者：

Highlights：

• A private classification algorithm when labeled data is limited is proposed.

• A pool of approximately correct label assignments is constructed based on Transductive Support Vector Machines.

• Exponential mechanism is applied for sampling the output label assignment from the generated pool.

• Utility analysis is given for the proposed algorithm.

摘要

•A private classification algorithm when labeled data is limited is proposed.•A pool of approximately correct label assignments is constructed based on Transductive Support Vector Machines.•Exponential mechanism is applied for sampling the output label assignment from the generated pool.•Utility analysis is given for the proposed algorithm.

论文关键词：Unlabeled data,TSVM,Differential privacy,Exponential mechanism

论文评审过程：Received 20 January 2017, Revised 28 June 2017, Accepted 2 July 2017, Available online 8 July 2017, Version of Record 4 September 2017.

论文官网地址：https://doi.org/10.1016/j.knosys.2017.07.006