Comment on Deterministic Information Bottleneck

TL;DR

The paper critiques the use of Deterministic Information Bottleneck for lossy compression without blocklength and proposes a new objective function to address this limitation.

Contribution

It introduces a new objective function that enables deterministic information bottleneck to be used effectively for lossy compression with blocklength.

Findings

Deterministic Information Bottleneck is limited for lossy compression without blocklength.

A new objective function is proposed to improve its applicability.

Further testing of the new objective is suggested for future work.

Abstract

We make the case that although Deterministic Information Bottleneck may be a contribution to clustering, it should not be used to aid lossy compression without the addition of blocklength. We therefore suggest a new objective function that does so and leave its testing to future work.