Our new X account is live! Follow @wizwand_team for updates
WorkDL logo mark

Unsupervised Learning of Debiased Representations with Pseudo-Attributes

About

Dataset bias is a critical challenge in machine learning since it often leads to a negative impact on a model due to the unintended decision rules captured by spurious correlations. Although existing works often handle this issue based on human supervision, the availability of the proper annotations is impractical and even unrealistic. To better tackle the limitation, we propose a simple but effective unsupervised debiasing technique. Specifically, we first identify pseudo-attributes based on the results from clustering performed in the feature embedding space even without an explicit bias attribute supervision. Then, we employ a novel cluster-wise reweighting scheme to learn debiased representation; the proposed method prevents minority groups from being discounted for minimizing the overall loss, which is desirable for worst-case generalization. The extensive experiments demonstrate the outstanding performance of our approach on multiple standard benchmarks, even achieving the competitive accuracy to the supervised counterpart.

Seonguk Seo, Joon-Young Lee, Bohyung Han• 2021

Related benchmarks

TaskDatasetResultRank
Attribute ClassificationCelebA (test)
Worst-group Accuracy94.94
48
Attribute Classification (Blond Hair)CelebA
Unbiased Accuracy92.62
36
Attribute Classification (Blurry)CelebA
Unbiased Accuracy89.79
36
Object ClassificationWaterbirds (test)
Worst-Group Accuracy71.39
22
Image ClassificationCMNIST 0.5% bias ratio unbiased (test)
Accuracy73.34
17
Image ClassificationCMNIST 5% bias ratio unbiased (test)
Accuracy97.13
11
Image ClassificationCMNIST 2% bias ratio unbiased (test)
Accuracy89.42
11
Image ClassificationCIFAR10C 0.5% bias ratio unbiased (test)
Accuracy25.5
11
Image ClassificationCMNIST 1% bias ratio unbiased (test)
Accuracy87.21
11
Image ClassificationCIFAR10C 1% bias ratio unbiased (test)
Accuracy26.86
11
Showing 10 of 28 rows

Other info

Code

Follow for update