Worst-case Asymmetric Distributed Source Coding

TL;DR

This paper addresses the worst-case asymmetric distributed source coding problem, introducing a new information measure called information ambiguity, and provides protocols and analysis showing limited benefits of block-coding in this context.

Contribution

It introduces the concept of information ambiguity for worst-case analysis and develops interactive protocols for asymmetric distributed source coding.

Findings

Information ambiguity is a useful measure for worst-case analysis.

Interactive protocols can achieve lossless data gathering with minimal bits.

Block-coding offers negligible advantage in worst-case distributed compression.

Abstract

We consider a worst-case asymmetric distributed source coding problem where an information sink communicates with $N$ correlated information sources to gather their data. A data-vector $\bar{x} = (x_1, ..., x_N) \sim {\mathcal P}$ is derived from a discrete and finite joint probability distribution ${\mathcal P} = p(x_1, ..., x_N)$ and component $x_i$ is revealed to the $i^{\textrm{th}}$ source, $1 \le i \le N$. We consider an asymmetric communication scenario where only the sink is assumed to know distribution $\mathcal P$. We are interested in computing the minimum number of bits that the sources must send, in the worst-case, to enable the sink to losslessly learn any $\bar{x}$ revealed to the sources. We propose a novel information measure called information ambiguity to perform the worst-case information-theoretic analysis and prove its various properties. Then, we provide interactive communication protocols to solve the above problem in two different communication scenarios. We also investigate the role of block-coding in the worst-case analysis of distributed compression problem and prove that it offers almost no compression advantage compared to the scenarios where this problem is addressed, as in this paper, with only a single instance of data-vector.