CS 730R: Topics in Data and Information Management

1. Summary. In this paper the authors presents two major contributions to the recent work of privacy on streaming data initially introduced by Dwork et al. [2]. First the authors study the family of problems related to the decayed sums. These problems are motivated by the fact that only the most recent part of the stream is often considered important in real applications rather than considering the whole stream. In this scenario the authors proposed three decayed sums results: on fixed length windows (window sum), exponential decay (the sum is computed through a exponential function which decades with the past) and polynomial decay (where the sum decades polynomially). Contrary to the solutions in [1, 2] which consider the whole stream the results in this paper show a considerable improvement in the term of accuracy, since in this case the error is proportional to the length of the window rather than the entire stream. The second major contribution consists in the notion of privacy with expiration where the privacy is relaxed in the past part of the stream while is enforced on the recent data. The authors show that with this relaxation the utility results improve since the error scales with the length of the stream where the privacy is enforced rather than the entire stream.