What Can We Learn Privately?

Abstract

Learning problems form an important category of computational tasks that generalizes many of the computations researchers apply to large real-life data sets. We ask: what concept classes can be learned privately, namely, by an algorithm whose output does not depend too heavily on any one input or specific training example? More precisely, we investigate learning algorithms that satisfy differential privacy, a notion that provides strong confidentiality guarantees in the contexts where aggregate information is released about a database containing sensitive information about individuals. We present several basic results that demonstrate general feasibility of private learning and relate several models previously studied separately in the contexts of privacy and standard learning.

Limits of Computational Differential Privacy in the Client/Server Setting

“…The most important are queries with outputs that cannot naturally be thought of as tuples of real numbers. This includes, e.g., queries that return classifiers (as in [9]), graphs, or synthetic databases.…”

“…The standard definition of differential privacy is very strong, requiring unconditional privacy guarantees against computationally unbounded adversaries. Despite this fact, there has been a good amount of success in designing differentially private mechanisms for many types of queries and in various settings [1,5,12,2,9].…”

Personalized Differential Privacy Preserving Data Aggregation for Smart Homes

“…The concept of local differential privacy was first proposed by S. P. Kasiviswanathan et al in [16]. The authors of [4] proposed a scheme RAPPOR that addresses the problem of longitudinal data collection with local differential privacy.…”

“…Local differential privacy (LDP) [16] is a rigorous privacy notion in local setting, which provides a more tough privacy guarantee than the traditional differential privacy. It is because the mechanism needn't any trustable third-party.…”

“…It has been certified that differential privacy can provide rigorous privacy guarantees, regardless of the background knowledge and computing power an adversary has. However, privacy preservation for smart homes belongs to a new differential privacy mechanism which is named local privacy [16]. The local privacy refers that the privacy protection algorithm is performed locally on residents' sensor devices.…”