Privacy-Preserving Algorithms for Outsourced Data Mining in Cloud Computing

Description

Due to ever increasing data size, many data owners choose to host their data in thecloud. Since the cloud server is not fully trusted, the privacy of the data becomes a majorconcern when outsourcing data mining services to the cloud. In traditional data miningsystems, the data owner itself provides data mining services to the users and its privacy-preservingis only against the users. In the outsourced data mining, the cloud serverplays the role of data miner, while it is not fully trusted. Therefore, the data owner needsto preserve privacy against both the cloud server and the users. Existing techniques forprivacy-preservation, such as differential privacy, bucketization and result auditing, arebased on mechanisms of adding quantized noise to the raw data or mining results.However, they are not strong enough to protect the privacy against the cloud server. Theprincipal goal of this project is to investigate privacy-preserving issues in outsourceddata mining. This project has three major tasks: 1) Design privacy-preservingalgorithms for outsourced data mining in cloud. 2) Develop a measurement model toquantify privacy and utility of data mining algorithms. To protect data privacy, sensitiveinformation is often removed or hidden from the raw data, which leads to the loss ofdata utility for data mining. There is no common model to measure the privacy or theutility of data mining algorithms. We plan to develop a measurement model that canprovide quantitative analysis of privacy and utility for data mining algorithms. 3) Designtradeoff mechanisms between privacy and utility. Privacy and utility are the twocontradictory metrics of data mining algorithms. It is important to strike the balancebetween privacy and utility. We plan to develop tradeoff mechanisms between privacyand utility for different data mining applications.