Explaining an image classifier with a generative model conditioned by uncertainty

TL;DR

This paper introduces a method to explain image classifier behavior by conditioning a generative model on the classifier's uncertainty, demonstrated through experiments on synthetic and MNIST data.

Contribution

It presents a novel approach to interpret classifier decisions by integrating uncertainty into generative models, offering new insights into model behavior.

Findings

Effective in illustrating classifier uncertainty on synthetic data.

Applicable to real-world datasets like MNIST with corrupted images.

Provides a new tool for model interpretability.

Abstract

We propose to condition a generative model by a given image classifier uncertainty in order to analyze and explain its behavior. Preliminary experiments on synthetic data and a corrupted version of MNIST dataset illustrate the idea.