In addition to using a classic output perturbation mechanism, the approach relies on privately answering a selected set of additional queries that are used to improve the accuracy of the data-release mechanism. This post-processing step uses convex optimization to redistribute the noise on the private answers to enforce consistency.
Place: 3rd
Prize amount: $5,000
Team members: Fernando Fioretta, James Kotary, Cuong Tran, Jingwen Xu