摘要
1Introduction Texts data are used to train deep learning models in cloud servers that have the strong computing power and large storage space,which can seriously endanger the user's privacy.Specifically,the attacker can use the intercepted text representations of text data of the primary learning tasks to train an adversarial classifier to infer private attributes or private information,such as gender,age of the user,and the relation between users.
基金
supported by the Guangdong National Key Research and Development Plan(2018YFB1800702,PCL2021A02)
the National Natural Science Foundation of China(Grant Nos.62002077,U20B2046,U1636215,61871140,U1803263)
the Guangdong Higher Education Innovation Group 2020KCXTD007 and Guangzhou Higher Education Innovation Group 202032854,the Guangdong Province Universities and Colleges Pearl River Scholar Funded Scheme(2019)
the China Postdoctoral Science Foundation(2020M682657).
