You have a dataset containing user salary information and need to generate an aggregate report showing average salaries by city while preserving individual privacy. The solution must maintain data accuracy, completeness, and reliability, and the aggregation must be statistically consistent with the original data distribution. You must return an approximate result instead of raw data.

Which differential privacy approach should you implement?