Spurred by developments such as cloud computing, there has been considerable interest in the data-mining-as-a-service (DMaS) paradigm in which a client outsources his/her data mining needs to a third-party service provider. However, this raises a few security concerns. One of the security concerns is that the service provider may return plausible but incorrect mining results to the client. There is a crucial need for techniques that enable the client to verify, without much effort, that the service provider has performed the outsourced computations faithfully and returned correct mining results. Despite the recent intensive efforts on verifiable general-purpose computations, efficient result verification of data mining computations remains a largely unexplored territory.
This CAREER proposal aims at designing efficient and practical verification techniques for data mining computations outsourced to an untrusted service provider. Research activities include developing (1) innovative verification approaches for data mining computations without any privacy preservation mechanisms; (2) new verification approaches for privacy-preserving data mining computations; (3) novel methods for the analysis of attack types and modeling of the collusion behaviors of service providers; and (4) a full system approach in developing, deploying, and evaluating the proposed techniques.
Advances in verifiable outsourcing of data mining computations can spur wider adoption of cloud services. This project also includes curriculum development and the training of high school, undergraduate, graduate, women and students in underrepresented groups.
