Abstract

We consider the problem of exchanging sensitive information in public and provide a general formulation that can unify and extend various existing scenarios of information exchange, such as the problems of private information extraction and information bottleneck. The formulation also gives rise to a new scenario called secure omniscience (SO), where users want to exchange all their private information with minimum leakage to a wiretapper with side information. Single-letter lower and upper bounds are obtained for the minimum leakage, and the bounds are shown to be tight under the finite linear source model with two users. The bounds are derived in terms of the solutions of the closely related problems of communication for omniscience (CO) and secret key agreement (SKA). However, we find examples where the bounds are not tight, and so the connections to CO and SKA are not precise. In particular, it is possible that any optimal CO scheme that minimizes communication does not minimize leakage, and any optimal SO scheme that minimizes leakage does not attain the capacity for SKA. Nevertheless, we identify a useful notion of information alignment that can modify an optimal CO scheme to reduce leakage for SO.


Presenters

Chung Chan

Department of Computer Science, City University of Hong Kong

Navin Kashyap

Indian Institute of Science

Praneeth Kumar Vippathalla

Indian Institute of Science

Qiaoqiao Zhou

Department of Information Engineering, The Chinese University of Hong Kong.

Session Chair

Matthieu Bloch

Georgia Institute of Technology