Visualizing Classification Structure of Large-Scale Classifiers

TL;DR

This paper introduces a novel measure for class similarity in large-scale classifiers, using visualization of class relationships to reveal hierarchical structures and improve understanding of classification behavior.

Contribution

It proposes a new class similarity measure based on prediction scores and demonstrates how visualizing this matrix uncovers class hierarchies and aids analysis.

Findings

Revealed hierarchical class structures through visualization

Identified potential corner cases in classifiers

Enhanced understanding of classification behavior

Abstract

We propose a measure to compute class similarity in large-scale classification based on prediction scores. Such measure has not been formally pro-posed in the literature. We show how visualizing the class similarity matrix can reveal hierarchical structures and relationships that govern the classes. Through examples with various classifiers, we demonstrate how such structures can help in analyzing the classification behavior and in inferring potential corner cases. The source code for one example is available as a notebook at https://github.com/bilalsal/blocks