Minutemen

Sprint 1: Project Summary

 To reduce the sensitivity, we preprocessed the data by subsampling to reduce the max individual contributions. For each epsilon we had different strategies. For epsilon below 1.5 we needed techniques that introduced more modeling assumptions and that for epsilon above 1.5 the Laplace mechanism did quite well. For epsilon below 0.3, we found insignificant incidents from the 2019 data and merged and distributed Laplace noise among them. For epsilon between 0.3 and 1.5, we measured noisy 2-way marginals using the Laplace mechanism. We then used a marginal-based inference engine to estimate a 3-way histogram of the dataset from the marginals.

Place: 1st

Prize amount: $10,000

Team members: Ryan McKenna, Joie Wu, Arisa Tajima, Brett Mullins, Siddhant Pradhan, Cecilia Ferrando