Two4Two: Evaluating Interpretable Machine Learning - A Synthetic Dataset For Controlled Experiments

TL;DR

This paper introduces a synthetic dataset generator for controlled experiments in interpretable machine learning, enabling systematic human and algorithm evaluations with customizable biases.

Contribution

The authors present a novel library that creates synthetic 3D animal images with controllable biases, simplifying the design of human-subject experiments in ML interpretability.

Findings

Generated biases are predictive enough for classifiers.

Biases are subtle and detectable by about half of human inspectors.

The approach lowers barriers for conducting human evaluations.

Abstract

A growing number of approaches exist to generate explanations for image classification. However, few of these approaches are subjected to human-subject evaluations, partly because it is challenging to design controlled experiments with natural image datasets, as they leave essential factors out of the researcher's control. With our approach, researchers can describe their desired dataset with only a few parameters. Based on these, our library generates synthetic image data of two 3D abstract animals. The resulting data is suitable for algorithmic as well as human-subject evaluations. Our user study results demonstrate that our method can create biases predictive enough for a classifier and subtle enough to be noticeable only to every second participant inspecting the data visually. Our approach significantly lowers the barrier for conducting human subject evaluations, thereby facilitating more rigorous investigations into interpretable machine learning. For our library and datasets see, https://github.com/mschuessler/two4two/