A multidimensional analysis of social biases in vision transformers

Brinkmann, Jannik ; Swoboda, Paul ; Bartelt, Christian

DOI:	https://doi.org/10.1109/ICCV51070.2023.00453
URL:	https://ieeexplore.ieee.org/document/10377502
Additional URL:	https://www.researchgate.net/publication/377427167...
Document Type:	Conference or workshop publication
Year of publication:	2023
Book title:	2023 IEEE/CVF International Conference on Computer Vision : ICCV 2023, Paris, France, 2-6 October 2023, proceedings
Page range:	4891-4900
Conference title:	2023 IEEE/CVF International Conference on Computer Vision, ICCV
Location of the conference venue:	Paris, France
Date of the conference:	02.-06.10.2023
Publisher:	Jurie, Frédéric ; Sharma, Gaurav
Place of publication:	Piscataway, NJ
Publishing house:	IEEE
ISBN:	979-8-3503-0719-1 , 979-8-3503-0718-4
Related URLs:	https://openaccess.thecvf.com/content/IC... https://iccv2023.thecvf.com/ https://ieeexplore.ieee.org/stamp/stamp....
Publication language:	English
Institution:	Außerfakultäre Einrichtungen > Institut für Enterprise Systems (InES)
Subject:	004 Computer science, internet
Abstract:	The embedding spaces of image models have been shown to encode a range of social biases such as racism and sexism. Here, we investigate specific factors that contribute to the emergence of these biases in Vision Transformers (ViT). Therefore, we measure the impact of training data, model architecture, and training objectives on social biases in the learned representations of ViTs. Our findings indicate that counterfactual augmentation training using diffusion-based image editing can mitigate biases, but does not eliminate them. Moreover, we find that larger models are less biased than smaller models, and that models trained using discriminative objectives are less biased than those trained using generative objectives. In addition, we observe inconsistencies in the learned social biases. To our surprise, ViTs can exhibit opposite biases when trained on the same data set using different self-supervised objectives. Our findings give insights into the factors that contribute to the emergence of social biases and suggests that we could achieve substantial fairness improvements based on model design choices.