# Spread Spurious Attribute: Improving Worst-group Accuracy with Spurious Attribute Estimation Source: [Spread Spurious Attribute: Improving Worst-group Accuracy with Spurious Attribute Estimation]() Collection index: [llms.txt]() Data updated: 2025-10-28T13:56:52.603Z Authors: Junhyun Nam, Jaehyung Kim, Jaeho Lee, Jinwoo Shin Published: 2022-04-05T09:08:30.000Z ## Abstract The paradigm of worst-group loss minimization has shown its promise in avoiding to learn spurious correlations, but requires costly additional supervision on spurious attributes. To resolve this, recent works focus on developing weaker forms of supervision -- e.g., hyperparameters discovered with a small number of validation samples with spurious attribute annotation -- but none of the methods retain comparable performance to methods using full supervision on the spurious attribute. In this paper, instead of searching for weaker supervisions, we ask: Given access to a fixed number of samples with spurious attribute annotations, what is the best achievable worst-group loss if we "fully exploit" them? To this end, we propose a pseudo-attribute-based algorithm, coined Spread Spurious Attribute (SSA), for improving the worst-group accuracy. In particular, we leverage samples both with and without spurious attribute annotations to train a model to predict the spurious attribute, then use the pseudo-attribute predicted by the trained model as supervision on the spurious attribute to train a new robust model having minimal worst-group loss. Our experiments on various benchmark datasets show that our algorithm consistently outperforms the baseline methods using the same number of validation samples with spurious attribute annotations. We also demonstrate that the proposed SSA can achieve comparable performances to methods using full (100%) spurious attribute supervision, by using a much smaller number of annotated samples -- from 0.6% and up to 1.5%, depending on the dataset. ## Recorded benchmark results A research result applies to its stated dataset, split, metric and evaluation configuration. Compare like-for-like settings, preserve the metric's unit and direction, and distinguish paper-reported results from independent reproduction. The best result recorded here is not a claim of exhaustive, current global coverage. - [Object Classification on Waterbirds (test)](): Object Classification; Waterbirds (test); Worst-Group Accuracy: 89. - [Image Classification on CelebA original unshifted](): Image Classification; CelebA original unshifted; Worst Acc: 89.8. - [Classification on CelebA HairColor (test)](): Classification; CelebA HairColor (test); In-Distribution Accuracy: 92.8. - [Image Classification on Waterbirds original unshifted](): Image Classification; Waterbirds original unshifted; Worst Accuracy: 89. - [Image Classification on CMNIST original unshifted](): Image Classification; CMNIST original unshifted; Worst Accuracy: 71.1. - [Toxicity detection on CivilComments-WILDS (test)](): Toxicity Detection; CivilComments-WILDS (test); Average Accuracy: 89.1. - [Natural Language Inference on MultiNLI (test) (Matched & Mismatched)](): Natural Language Inference; MultiNLI (test); Average Worst-Group Accuracy: not recorded. - [Attribute Classification on CelebA (test)](): Attribute Classification; CelebA (test); Average Worst-Group Accuracy: not recorded. ## Original sources and implementation [Original paper]() [PDF]()