Abstract

We study the guessing variants of two distributed source coding problems: the Gray-Wyner network and the Slepian-Wolf network. Building on the former, we propose a new definition of the Rényi common information as the least attainable common rate in the Gray-Wyner guessing problem under the no-excess-rate constraint. We then provide a variational characterization of this quantity. In the Slepian-Wolf setting, we follow up the work of Bracher-Lapidoth-Pfister with the case where the expected number of guesses need not converge to one but must be dominated by some given exponential.


Presenters

Amos Lapidoth

ETH Zurich

Session Chair

Michael Gastpar

École Polytechnique Fédérale de Lausanne