Keywords: model debugging, vision transformers, missingness
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.
One-sentence Summary: We investigate how current missingness approximations for model debugging can impose undesirable biases on the model predictions and hinder our ability to debug models, and we show how transformer-based architectures can side-step these issues.
Community Implementations: [![CatalyzeX](/images/catalyzex_icon.svg) 2 code implementations](https://www.catalyzex.com/paper/arxiv:2204.08945/code)