Privacy Aware Learning

Natural tensions between learning and privacy arise whenever a learner must aggregate data across multiple individuals. The learner wishes to make optimal use of each data point, whereas the providers of the data may wish to limit detailed exposure, either to the learner or to other individuals. A characterization of such tensions in the form of quantitative tradeoffs is of great utility: it can inform public discourse surrounding the design of systems that learn from data, and the tradeoffs can be exploited as controllable degrees of freedom whenever such a system is deployed. In this paper, we approach this problem from the point of view of statistical decision theory. The decision-theoretic perspective offers a number of advantages.