Missingness Bias in Model Debugging

TL;DR

This paper investigates how missingness biases affect model debugging in computer vision and proposes transformer-based architectures as a natural solution to improve debugging reliability.

Contribution

It introduces a transformer-based approach for handling missing features in images, reducing bias and enhancing model debugging accuracy.

Findings

Transformers enable more natural missingness implementation.

Using transformers reduces bias in debugging.

Improved reliability in model debugging with transformer architectures.

Abstract

Missingness, or the absence of features from an input, is a concept fundamental to many model debugging tools. However, in computer vision, pixels cannot simply be removed from an image. One thus tends to resort to heuristics such as blacking out pixels, which may in turn introduce bias into the debugging process. We study such biases and, in particular, show how transformer-based architectures can enable a more natural implementation of missingness, which side-steps these issues and improves the reliability of model debugging in practice. Our code is available at https://github.com/madrylab/missingness