Optimal quantization for nonuniform discrete distributions

TL;DR

This paper investigates optimal quantization strategies for various nonuniform discrete distributions, providing theoretical insights and computational methods for data approximation and distribution reconstruction.

Contribution

It introduces new methods for identifying optimal representative points for both finite and infinite discrete distributions, including reciprocal and natural number distributions.

Findings

Optimal sets of representatives are computed for specific distributions.

Quantization quality is assessed for different approximation levels.

The reverse problem of distribution identification from optimal sets is addressed.

Abstract

This paper explores the process of optimal quantization for several types of discrete probability distributions. Quantization is a technique used to approximate a complex distribution with a smaller set of representative points, which is important in fields such as data compression and signal processing. We begin by examining two specific nonuniform distributions over a finite set of values and identify the best representative points for different levels of approximation. We then extend our analysis to two infinite discrete distributions: one supported on the reciprocals of natural numbers and another on the natural numbers themselves. For these distributions, we compute the optimal sets of representatives and assess how well they approximate the original distributions. Finally, we address the reverse problem-determining the underlying distribution when the optimal sets are known. Our results provide both theoretical insights and computational techniques useful in information theory and data analysis.