Identification of Systematic Errors of Image Classifiers on Rare Subgroups
This addresses fairness and robustness issues for demographic minorities and domain shifts in AI systems, representing a novel method for a known bottleneck.
The paper tackled the problem of identifying systematic errors in image classifiers on rare, unannotated subgroups by using text-to-image models and combinatorial testing to search for low-performance prompts, finding that it identifies these errors with high accuracy and uncovering novel systematic errors on ImageNet classifiers.
Despite excellent average-case performance of many image classifiers, their performance can substantially deteriorate on semantically coherent subgroups of the data that were under-represented in the training data. These systematic errors can impact both fairness for demographic minority groups as well as robustness and safety under domain shift. A major challenge is to identify such subgroups with subpar performance when the subgroups are not annotated and their occurrence is very rare. We leverage recent advances in text-to-image models and search in the space of textual descriptions of subgroups ("prompts") for subgroups where the target model has low performance on the prompt-conditioned synthesized data. To tackle the exponentially growing number of subgroups, we employ combinatorial testing. We denote this procedure as PromptAttack as it can be interpreted as an adversarial attack in a prompt space. We study subgroup coverage and identifiability with PromptAttack in a controlled setting and find that it identifies systematic errors with high accuracy. Thereupon, we apply PromptAttack to ImageNet classifiers and identify novel systematic errors on rare subgroups.