報告題目：Consent-based Privacy-preserving Decision Tree Evaluation
報 告 人：薛靚 博士
摘 要：Decision trees are one of the most widely used machine learning algorithms that can be used for data classification. Deploying the decision tree-based models into cloud servers has inspired many real-world applications, such as remote medical diagnoses and face recognition. However, as stringent privacy regulations of personal data, such as GDPR, takes effect, the decision tree evaluation must comply with some requirements. First, the model parameters and user data (input and output) should be protected. Second, different applications should obtain the classification results with users’ consent in the context of user-customized services. In this talk, we present a construction of consent-based privacy-preserving decision tree evaluation scheme. Specifically, to achieve model parameter privacy and user data privacy, the original decision tree evaluation is transformed into a private decision tree evaluation, such that all operations can be performed in the encrypted domain using an additively homomorphic encryption primitive and a secure comparison protocol. In addition, by integrating a proxy re-encryption technique, the scheme enables third-party applications to provide user-dependent services for users based on user’s classification results. The security analysis shows that the proposed scheme achieves the desirable security properties and performance evaluation demonstrates that the scheme is efficient and is suitable for real-world implementations.
報告人簡介：Liang Xue received the B.E. degree on Information Security and the M.S. degree on Computer Science and Technology from University of Electronic Science and Technology of China in 2015 and 2018, respectively. Currently, she is pursuing the PhD degree at the University of Waterloo. Her research interests include applied cryptography, privacy enhancing technologies, and Blockchain.