Our approach is decomposed into two steps: truncation and reweighting. Truncation is to ignore the incidents that are not in the significant bins learned from the public 2019 data. Reweighting is to reweight each remaining incident so that the total weight associated with each sim_resident is only a constant gamma. The global sensitivity thus is reduced to gamma, so adding laplace noise to each bin with scale gamma / epsilon will satisfy the \epsilon-DP requirement. The value of gamma is chosen carefully to balance the bias from truncation.
Prize amount: $1,000
Team members: Johes Bater, Yuchao Tao