Recent research has successfully demonstrated new types of data poisoning attacks. To address this problem, some researchers have proposed both offline and online data poisoning detection defenses which employ machine learning algorithms to identify such attacks. In this work, we take a different approach to preventing data poisoning attacks which relies on cryptographically-based authentication and provenance to ensure the integrity of the data used to train a machine learning model. The same approach is also used to prevent software poisoning and model poisoning attacks. A software poisoning attack maliciously alters one or more software components used to train a model. Once the model has been trained it can also be protected against model poisoning attacks which seek to alter a model's predictions by modifying its underlying parameters or structure. Finally, an evaluation set or test set can also be protected to provide evidence if they have been modified by a second data poisoning attack. To achieve these goals, we propose VAMP which extends the previously proposed AMP system, that was designed to protect media objects such as images, video files or audio clips, to the machine learning setting. We first provide requirements for authentication and provenance for a secure machine learning system. Next, we demonstrate how VAMP's manifest meets these requirements to protect a machine learning system's datasets, software components, and models.
翻译:最近的研究成功地展示了新型数据中毒袭击。为了解决这一问题,一些研究人员提出了离线和在线数据中毒检测防御,采用机器学习算法来识别此类袭击。在这项工作中,我们采取了不同的方法来防止数据中毒袭击,这依赖于加密的认证和源代码,以确保用于培训机器学习模型的数据的完整性。同样的方法也用于防止软件中毒和模型中毒袭击。软件中毒袭击恶意地改变了用于培训模型的一个或多个软件组件。一旦模型经过培训,它也可以受到保护,防止模型中毒袭击,这种袭击试图通过修改其基本参数或结构来改变模型预测。最后,如果数据中毒袭击对数据集或测试集进行了修改,我们也可以加以保护,以提供证据。为了实现这些目标,我们提议VAMP将先前提议的AMP系统扩展至机器学习环境,目的是保护图像、视频文件或音频剪等媒体对象。我们首先为安全机器学习系统提供认证和证明要求。接下来,我们演示了VAMP机器软件模块如何满足这些软件的模版要求。