2022/03/23 by Natalie Dullerud, Dullerud, Natalie, Karsten Roth +7
Computer Science · #AI in cancer detection #Artificial Intelligence (cs.AI) #Computers and Society (cs.CY) #Domain Adaptation and Few-Shot Learning #FOS: Computer and information sciences #Face recognition and analysis #Machine Learning (cs.LG) #Machine Learning (stat.ML)
paper · pdf · doi:10.48550/arxiv.2203.12748
openalex publication_date 2022/03/23 · openalex created_date 2022/04/03 · openalex updated_date 2026/07/28
Deep metric learning (DML) enables learning with less supervision through its emphasis on the similarity structure of representations. There has been much work on improving generalization of DML in settings like zero-shot retrieval, but little is known about its implications for fairness. In this paper, we are the first to evaluate state-of-the-art DML methods trained on imbalanced data, and to show the negative impact these representations have on minority subgroup performance when used for downstream tasks. In this work, we first define fairness in DML through an analysis of three properties of the representation space -- inter-class alignment, intra-class alignment, and uniformity -- and propose finDML, the fairness in non-balanced DML benchmark to characterize representation fairness. Utilizing finDML, we find bias in DML representations to propagate to common downstream classification tasks. Surprisingly, this bias is propagated even when training data in the downstream task is re-balanced. To address this problem, we present Partial Attribute De-correlation (PARADE) to de-correlate feature representations from sensitive attributes and reduce performance gaps between subgroups in both embedding space and downstream metrics.