Machine learning and data mining algorithms have now penetrated our everyday lives in a wide range of applications, like web search, social networking, communication networks, mobility data, advertising, cloud computing, and business intelligence. Each application can have its own privacy requirements, which include protection of personal information and statistical disclosure control, protection of business secrets, and knowledge hiding in general. Adaptations of traditional privacy definitions such as randomized response, k-anonymity, and differential privacy to these new applications do not always adequately protect sensitive information. Attacks on supposedly anonymized data and additional privacy issues raised in the literature have shown that developing a robust privacy definition for a new application requires a tremendous feat of engineering. Another challenge in privacy preserving data analysis is maintaining accuracy (or utility) of the algorithms. There is an inherent trade-off between privacy and utility which needs to be formally captured for each data mining task. Often, the utility guarantee provided by an algorithm is dependent on the privacy notion it satisfies. Comparing the utilities of algorithms that implement different privacy definitions is still an open challenge. Where theoretical utility analysis is difficult or impossible, it is important to develop performance and data benchmarks, facilitating reliable comparison of competing methods. Finally, there is a need to address the new privacy challenges presented by emerging applications such as mobility data mining, social network analysis, advertising, and cloud computing.
PADM will be a full-day workshop that will be held in
conjunction with the IEEE ICDM 2011 conference in Vancouver, Canada. The purpose of this workshop is to encourage principled research that develops methodologies to address open privacy problems.