English

Information-Distilling Quantizers

Information Theory 2019-10-30 v2 math.IT

Abstract

Let XX and YY be dependent random variables. This paper considers the problem of designing a scalar quantizer for YY to maximize the mutual information between the quantizer's output and XX, and develops fundamental properties and bounds for this form of quantization, which is connected to the log-loss distortion criterion. The main focus is the regime of low I(X;Y)I(X;Y), where it is shown that, if XX is binary, a constant fraction of the mutual information can always be preserved using O(log(1/I(X;Y)))\mathcal{O}(\log(1/I(X;Y))) quantization levels, and there exist distributions for which this many quantization levels are necessary. Furthermore, for larger finite alphabets 2<X<2 < |\mathcal{X}| < \infty, it is established that an η\eta-fraction of the mutual information can be preserved using roughly (log(X/I(X;Y)))η(X1)(\log(| \mathcal{X} | /I(X;Y)))^{\eta\cdot(|\mathcal{X}| - 1)} quantization levels.

Keywords

Cite

@article{arxiv.1812.03031,
  title  = {Information-Distilling Quantizers},
  author = {Alankrita Bhatt and Bobak Nazer and Or Ordentlich and Yury Polyanskiy},
  journal= {arXiv preprint arXiv:1812.03031},
  year   = {2019}
}
R2 v1 2026-06-23T06:35:24.633Z