摘要
Machine learning models(机器学习模型) are increasingly being deployed in practice. Machine Learning as a Service (MLaaS 机器学习即服务) providers expose such models to queries by third-party developers(第三方开发人员) through application programming interfaces (APIs 应用程序编程接口). Prior work(先前的工作) has developed model extraction attacks(模型提取攻击), in which an attacker extracts an approximation(提取近似值) of an MLaaS model by making black-box queries(黑盒查询) to it. We design ACTIVETHIEF – a model extraction framework for deep neural networks(一个深度神经网络的模型提取框架) that makes use of active learning techniques(主动学习技术) and unannotated public datasets(为注释的公共数据集) to perform model extraction(执行模型提取). It does not expect strong domain knowledge(强大的领域知