By collecting sensor data from individuals in a user community, e.g., using their smartphones, it is possible to learn the behavior of communities, for example locations, activities, and events. Similarly, using data from personal health monitoring sensors, it is possible to learn about the health risks and responses to treatments for population groups. But is it possible to use the valuable information for the greater good without disclosing information about the individuals contributing the data? What about protecting this information from improper access? This project uses cloud computing augmented with a combination of data reduction techniques and methods from differential privacy and homomorphic encryption to address such questions. The key ideas are (1) to use coresets as a way of mitigating the computational challenges around the state of the art in differential privacy and homomorphic encryption and to ensure private secure computation on the server side and the client side, and (2) to give the data owners control over setting access to their data as a trade-off between data access control guarantees and computation accuracy. Combining coresets, differential privacy, and homomorphic encryption has the potential for practical private and secure computation in the cloud. This proposal builds on previous results in coresets, private coresets, and their implementations for the cloud. Coresets are a data reduction technique for computing a function f on a large data set D efficiently by compressing the initial data into a small data set C (possibly on the cloud), and then solving the problem f on the reduced set C (now, possible also at the client). The reduced data set C is chosen so that it is fast to compute f(C) and f(C) ~ f(D). A particular type is coreset is the private coreset, which preserves privacy but must be constructed and sanitized on the client side. On the other hand, fully homomorphic encryption allows encrypted computation on the server side but it is usually impractical. This project develops (i) New private coresets for broad classes of practical problems, with focus on generic frameworks as for non-private coresets; (ii) Novel algorithms and techniques for efficient homomorphic encryption on the cloud using coresets; (iii) Private Encrypted Coresets which are new type of coresets that simultaneously preserve privacy and can be computed securely on the cloud; (iv) Life-logging systems that implement and combine the above techniques for simultaneous secure and private computation in the cloud, with appropriate benchmarks and real-world testing.
NSFSaTC-BSF: TWC: Small: Enabling Secure and Private Cloud Computing using Coresets