Non-uniform quantization with linear average-case computation time

TL;DR

The paper introduces a novel non-uniform quantization method that achieves linear average-case computation time by avoiding binary search, resulting in over four times faster binning for unknown data distributions.

Contribution

It proposes a new binning method that skips binary search, providing linear average-case performance for non-uniform quantization tasks.

Findings

Achieves over fourfold speedup compared to binary search binning.

Proven to have linear average-case computation time.

Experimental results confirm theoretical analysis.

Abstract

A new method for binning a set of $n$ data values into a set of m bins for the case where the bins are of different sizes is proposed. The method skips binning using a binary search across the bins all the time. It is proven the method exhibits a linear average-case computation time. The experiments' results show a speedup factor of over four compared to binning by binary search alone for data values with unknown distributions. This result is consistent with the analysis of the method.