Differential Privacy Research Agenda

Thoughts on a differential privacy research agenda:

• Anonymous set intersection made differentially private.
• Practical applications for DP secure multiparty computations.
• Tools for deciding on the privacy/accuracy tradeoff
• Tools for building DP models
• Approaches for analyzing programs that provide DP. Analyzing programs that implement randomized algorithms is hard: how do you find bugs?
• More mechanisms.
• DP for text, or a new theory for text de-identification.
• DP theories that provide empirical privacy loss measures by taking into account background information. Would this produce different mechanisms? COuld these mechanisms offer more accuracy without a practical impacts on privacy loss?
• Improved methods for teaching differential privacy.
• Integrating the concepts of differential privacy into the high school curriculum.
• DP theories that take into account computational complexity or numeric precision, so that we can release more aggregate data and know that we aren't impacting privacy loss of individuals.