The ability to disseminate useful statistics generated from sensitive data while still preserving the privacy of its contributors is an open challenge with highly desirable consequences. In this thesis, we investigate the problem of histogram estimation within the framework of differential privacy. Given a histogram with noisy counts, we seek to generate an estimate of the original, non-noisy histogram such that our estimate is expected to improve upon the identity estimator in terms of sum-squared-error. As a stepping stone towards this goal, this thesis investigates several possible choices of such an estimator as a baseline for comparison.
